Datasets:

kejian

/

SciReviewGen

like 0

Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> The foreseeable evolution of electric networks within the next years walks towards a progressive and increasing use of Distributed Generation (DG), specially small clustered photovoltaic (PV) inverters on building rooftops thanks to the economic support in many countries, such as Spain. Plugged-in Electric Vehicle (EV) should be considered as well, having seen the big interest of the automotive industry on it. The separated introduction of these two elements may create benefits but also risks, both economical and technical. As benefits, the improvement of voltage profile along feeders and losses decrease. Overloading at equipments, excessive voltage raise or drop, increasing losses or the mandatory equipments reinforcement in order to satisfy the new requirements should be considered as possible risks. Quality of electrical supply must also be considered. This paper discusses the synergy of the coordinated introduction of both elements which would compensate the caused damages of the separated addition. <s> BIB001 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> Net-metering is commonly known as a practice by which owners of distributed generation (DG) units may offset their electricity consumption from the grid with local generation. The increasing number of prosumers (consumers that both produce and consume electricity) with solar photovoltaic (PV) generation combined with net-metering results in reduced incomes for many network utilities worldwide. Consequently, this pushes utilities to increase charges per kWh in order to recover costs. For non-PV owners, this could result into inequality issues due to the fact that also non-PV owners have to pay higher chargers for their electricity consumed to make up for netted costs of PV-owners. In order to provide insight in those inequality issues caused by net-metering, this study presents the effects on cross-subsidies, cost recovery and policy objectives evolving from different applied netmetering and tariff designs for a residential consumer. Eventually this paper provides recommendations regarding tariffs and metering that will result in more explicit incentives for PV, instead of the current implicit incentives which are present to PV owners due to net-metering. <s> BIB002 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> With conventional generating units being replaced by renewable sources which are not required to provide same high level of ancillary services, there is an increasing need for additional resources to achieve certain standards regarding frequency and voltage. This paper investigates the potential of incorporating electric vehicles (EVs) in a low voltage distribution network with high penetration of photovoltaic installations (PVs), and focuses on analysing potential voltage support functions from EVs and PVs. In addition, the paper evaluates the benefits that reactive power control may provide with addressing the issues regarding voltage control at the expense of increased loading. Analysed real Danish low voltage network has been modelled in Matlab SimPowerSystems and is based on consumption and PV production data measured individually for number of households. <s> BIB003 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> An analysis of the operating conditions of a power transformer with the introduction of photovoltaic (PV) panels and Electric Vehicles (EV) in an eco-district is conducted. The study focuses on: (a) determining the optimal transformer sizing when there is neither PV nor EV; and (b) characterizing the overloading periods of the transformer induced by the penetration of PV sources and EVs in the district. We compare two EV integration strategies: they are either considered as unmanaged loads, or controlled by an Energy Management System (EMS) with Vehicle-to-Grid (V2G) capabilities. Transformer operating curves and rated power are deduced from technical standards supplied by Schneider Electric. The results show that EVs and PV introduction has a substantial impact on overloading periods. However, the latters are seriously mitigated both in terms of energy flows and peak power with the implementation of the EMS. <s> BIB004 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> Photovoltaic systems (PVSs) and plug-in electric vehicles (PEVs) are becoming increasingly common and expected to be further integrated into electric power distribution systems (EDSs) in the near future. In this paper, the combined effects of PVSs and PEVs on the feeder and transformer loads, voltage profiles and harmonic distortions of an urban area EDS are investigated. The results indicate that the simultaneous connection of PVSs and PEVs, at the correct ratio, can impact load reduction and produce a reduction in voltage variations while potentially resulting in an increase the total harmonic distortion of voltage (THDV). In addition, this paper presents changes in the characteristic daily load and voltage profiles and changes in the THDV of the system caused by the simultaneous operation of PVSs and PEVs. The test EDS with 19 nodes is based on a real EDS. Two seasons, summer and winter, and several different degrees of penetration of PVSs and PEVs were studied. We used a stochastic approach to model the loads, PVSs and PEVs based on data measured in real systems and data found in the literature. <s> BIB005 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> Strategic use of domestic battery storage in households with a V2G-capable electric vehicle and photovoltaic generation is shown to be capable of shaving the peak loading on the electricity distribution grid under realistic operating conditions by up to 37%. The additional degree of freedom provided by combining the EV and domestic storage is shown to enable additional reductions in peak grid load. A smart control algorithm is described for coordinating the system components to reduce peak grid load and improve power quality. <s> BIB006 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> High penetration of photovoltaic (PV) generation in micro-grid may cause additional challenges for grid control. When clouds are moving over PV panels, it will result in a significant voltage drop. Aimed at this case, electric vehicle (EV) can be a strategic source to mitigate the effects of voltage drop on account of the fluctuating cloud cover in PV micro-grid. This paper proposes a new methodology to determine the optimal EV charging-discharging strategy based on the uncertainty of voltage drop. Moreover, the optimal policy is obtained by Markov Decision Process model, with the aim of maximizing power quality benefits. The control strategy of EV is raised to help the PV micro-grid react the external changes and future events proactively. <s> BIB007 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> This paper investigates the impact of the grid integration of roof-top Photovoltaic (PV) generation and Electric Vehicles (EVs) energy storage on the demand response. The risk indices are introduced and risk map is created to predict the potential weather impact on the PV and EV owners. Based on the predictions, the aggregator bidding strategies to enable the EVs' stored energy and PV generation to participate in the ancillary service market considering the stochastic behavior are proposed. The role of programs for demand-side management (DSM) (in daily operation) and outage management (OM) (when fault happens) to mitigate the negative weather impacts on the customers is explored. Numerical experiments are implemented to validate the proposed approach and illustrate the impact of PV generation and EVs' ability to charge/discharge the stored energy on the flexibility of the electricity customer demand. <s> BIB008 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> Depending on charging locations and business interest, EV charging facilities can be classified into Commercial Charging (CC), Business Charging (BC), and Home Charging (HC). Diurnal load distribution in these EV charging facilities can be dissimilar on the account of different customers' behavior. Therefore, these charging facilities will put forth different degrees of impacts on the grid. A portion of the impacts can be reduced through onsite deployment of PV and battery energy storage system (BESS). However, the extent of reduction depends on wide range of factors such as costs, correlation between EV load profile and PV generation profile etc. Thus, the objective of this work is to find out which one of these three charging facilities is technically and economically feasible for PV and BESS based EV charging station. <s> BIB009 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> The aim of this paper is to avoid overloading a private customer distribution transformer (DT) in a Portuguese insular area through the means of solar PV microgeneration. Firstly, the consequence of the penetration of electric vehicles (EV) on dielectric oil deterioration of a DT in an industrial unit is estimated. The workplace has local PV generation, allowing the EVs to charge while their owners are working at three different working shifts during a day. Secondly, the model is tested and the resulting scenarios are analyzed. This paper shows that the solar PV microgeneration decreases the overloading of the DT due to a lower daily load profile. It also contributes to the reduction of the loss-of-life (LoL) of the DT. <s> BIB010 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> Abstract This study investigates the impact of single-phase plug-in electric vehicles charging on increasing the rate at which center-tapped distribution transformers experience aging. Distribution transformer aging is investigated considering varying rooftop solar photovoltaic generation penetration rates. Monte Carlo methods are used to probabilistically estimate the transformer’s loss of life considering the effect of time-of-use (TOU) pricing. The results of applying the proposed method have revealed that plug-in battery electric vehicle charging impact on both transformer aging and neutral current is largest in the case that vehicles charge based on time-of-use pricing methods. Further application has shown that while rooftop solar photovoltaic generation reduces transformer aging, no significant reduction in neutral current is observed. <s> BIB011 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> High penetrations of photovoltaic (PV) systems, energy storage (ES) devices, and electric vehicle (EV) charging may significantly affect the operational constraints of substation power transformers. In high penetrations these applications can flatten a transformer's daily load profile, which minimizes the cooling down period for the unit's paper insulation. Also, because these applications rely upon power electronics to interface with the electric grid, high penetrations can increase the volume of harmonic currents propagating through the distribution system, which can also impact transformer aging. Although the initial impact of PV and ES applications may reduce a unit's peak energy demand, long-term system planning and emergency operating conditions may require derating of existing capacity limits to prevent aging that impacts overall life expectancy. To identify transformer aging characteristics as a function of load profile and harmonic content, the authors developed a transformer transient model based upon the tested attributes of a 50MVA SPX Waukesha transformer and the modeling methods described in IEEE Std. C57.90 and IEEE Std. C57.110. Utilizing the model, the authors identified the relationship between aging, harmonic distortion, and load profile characteristics associated with high penetrations of PV, ES applications, and EV charging. <s> BIB012 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with distribution grids <s> Abstract The direct coupling of solar electricity with electric vehicles is very topical due to their many co-benefits, such as reducing the electricity flow into and from the grid. This work illustrates this solution using a park-and-ride lot outside Lisbon to study its solar energy potential to charge electric vehicles. The solar resource, its exploitation and coordination with the vehicles under different charging approaches are discussed in detail, using methods extendible to other cases. A genetic-algorithm optimisation was performed to maximise the solar energy generation. To ascertain the techno-economic feasibility of the concept, several sets of assumptions are analysed, including the use of energy storage and smart charging. For current market conditions, the payback time is found to be 14 years; a modest financial public incentive would significantly improve the project economics – making the payback drop to 7 years – enabling conditions to put this inescapable and green approach at the forefront. <s> BIB013

The purpose of distribution systems is to bring electricity from transmission grid to end-users. Therefore, distribution system operators (DSO) design distribution grids and organize investments in grid reinforcements in order to optimize power distribution. Those heavy investments in equipment are made for several decades. Regulation is then designed to remunerate DSO according to its costs. As EV and PV units are generally directly connected to the distribution grid, the combined effects of EV and PV should be the most visible there. To study the impact of EV/PV coupling, we start by reminding the individual effects of both technologies on the distribution grid and then analyze how beneficial the coupling could be to it. [90] provides an overview of the technical and economic effects of distributed generation on the distribution grid. Technically, photovoltaic generation in distribution grids leads to trouble in grid operations (with congestion and over-voltage issues), difficulties in maintaining power quality (with increased harmonics) and in keeping power losses to low level (losses are proportional to power variations). For the DSO, this results in additional grid upgrading and missing revenues at short and medium terms. As BIB002 points out, any additional costs for the DSO leads to an increase of distribution fees. With current regulation, this leads to higher costs for non-PV users, while PV users do not participate fairly to their own impacts on the grid. Regarding EV impacts, technical issues are similar: more difficult grid operations (phase unbalances, overvoltage, congestion), worse power quality (harmonics) and higher power losses. The EV impact is strongly dependent on the charging power (from 3kW at home to 50-150kW with fast chargers) . As an effect of EV/PV synergy is to increase the system self-consumption and to reduce grid power imports, at first sight, wealth transfers should be reinforced. Nonetheless, EV charging could create an opposite effect. As EV raises substantially the owner's power consumption, DSO's revenues could be increased. These new revenues could balance the DSO's missing revenues due to PV self-consumption in the grid. To our knowledge, the resulting trade-off and the appropriate network tariff designs have not been investigated yet. [92] distinguishes EV services in distribution grids between load (congestion management and power loss reduction) and voltage services (magnitude regulation and unbalance reduction). A part of the EV/PV coupling literature deals with improving its integration in distribution grids by diminishing its stress on it. Although this literature is mainly technical, it underlines the beneficial effect of EV/PV coupling on distribution grids. Controlling local peak load is a prime matter in order to stabilize the grid. Some the authors have studied local solutions to smooth load curves BIB006 . Due to its high initial cost (millions of euros), transformer aging is an important feature to watch for DSOs. Congestion problems at transformer level (provoked either by PV or EV) tend to shorten its lifetime, which diminish the investment profitability. The authors of BIB010 BIB011 BIB012 BIB004 develop strategies for an EV/PV coupling to fulfill charging requirements and self-consumption with a softer impact on the transformer. BIB001 proposes a smart strategy based on technical constraints. It aims at smoothing EV/PV coupling stress to the distribution grid. A smart strategy manages EV charging such as PV and EV profiles are more synchronized which leads to a smooth transformer load curve. EV charging led by voltage can avoid overloads issues and ensure the charging of a larger number of EV. Smoothing the curve leads to lower power losses, as lesser high current and power variations happen. A decentralized method of smart control strategies is to measure directly voltage profile of the system, in order to schedule EV to smooth it BIB003 BIB007 . This results in lesser power losses in the grid . In the selected literature, is one of the rare study to consider techno-economic aspects as exchanges of reactive power between EVs, PV systems and the distribution grid. The authors' method shows a 50% decrease in power losses in the grid, buy only in costs decrease of 3% for the EV/PV system. For instance, inject EV reactive power into the system, as it does not cause damage to the EV battery. As pointed out in BIB005 , EV/PV system benefits towards distribution grid may have its limits. The authors provide a technical study of the effects EV/PV system on several aspects of power quality (voltages profiles and harmonics voltage distortions). They show that joint effect of PV and EV can reduce the load while improving voltage profiles on the grid, but does not really improve power quality (harmonics distortions). For larger scales, references BIB008 BIB009 contain results that point out that a EV/PV synergy would be particularly beneficial in weak parts of the distribution grid. Those results indicate that EV/PV synergy could alleviate the distribution grid needs for reinforcements due to EV and PV. Nonetheless, this could require a specific contractual cooperation between the DSO and the EV/PV system manager. In their study on a PVCS system, BIB009 considers a grid in which grid strength (based on voltage magnitude) differs across areas. Results show that the influence of a smart strategy is even more beneficial in the weakest areas of the grid. Indeed, high power flows due to EV charging and PV power, and causes higher losses in a weak grid area and maximizing self-consumption in this area manages to improve the EV/PV system efficiency. This result is very intesting for the DSO, and hence could lead it to promote smart control strategy in EV/PV systems. BIB008 studies coordinating strategies of EV and PV in a distribution system focusing on demand side management and outage management. The distribution system faces weather change and risk is diversely distributed in the fed zones of the grid. Demand-side management strategy appears to be more efficient in the riskiest part of the grid with costs reduction above 94%. On the contrary, cost reductions in the safest parts of the grid are in the range 76-90%. The authors of BIB013 show that allowing voltage support from the PVCS (with a regulation price of 25.6e/MW/h) increases PVCS profits by 27%. As pointed out earlier, novel power grid components as solid-state transformers or DC microgrid are particularly adapted to EV/PV systems. As they constitute an valuable solution for its distribution management, DSO could be tempted to install those technologies. This would indirectly create an incentive to exploit EV/PV synergy. We can see through literature that a EV/PV system can interact with the distribution in a mutually beneficial way. Technical advantages for the grid are obvious, but literature lacks a quantified study of these benefits. The recognition of this systemic benefits should influence the potential regulation for EV/PV system in terms of entry conditions, pricing etc.

Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with transmission grids <s> Abstract A number of studies have found that plug-in hybrid electric vehicles (PHEVs) will have relatively small grid impacts if charged with overnight off-peak electricity. However the greatest petroleum displacement will occur if vehicles are allowed to charge during the day, maximizing all-electric range. Charging during the day also allows for a smaller, lower cost battery. Mid-day charging will add to peak electricity demands and may occur in locations where it is difficult to construct new generation and transmission capacity. Solar photovoltaics (PV) provide an option to provide mid-day peaking capacity. Mid-day charging of PHEVs also may absorb low value or even curtailed PV generation during periods of low demand. This study identifies possible co-benefits of large scale PV and PHEV deployment by simulating the Texas grid and identifying changes in peak capacity requirements and PV curtailment. A modest deployment of PV is able to avoid most of the increase in capacity requirements associated with very large PHEV penetrations. PHEVs are also able to reduce curtailment at high PV penetration, especially if charging can be controlled to improve the coincidence of consumer charging demand with normal PV generation patterns. <s> BIB001 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with transmission grids <s> Abstract This paper proposes a collaborative strategy between the photovoltaic (PV) participants and electric vehicle (EV) owners to reduce the forecast uncertainties and improve the predictability of PV power. The PV generation is predicted using an auto regressive moving average (ARMA) time series model. Fuzzy C-means (FCM) clustering is used to group the EVs into fleets with similar daily driving patterns. Uncertainties of the PV power and stochastic nature of driving patterns are characterized by a Monte Carlo simulation (MCS) technique. A particle swarm optimization (PSO) algorithm is developed to optimally use the vehicle-to-grid (V2G) capacities of EVs and minimize the penalty cost for PV power imbalances between the predicted power and actual output. The proposed method provides a coordinated charging/discharging scheme to realize the full potential of V2G services and increase the revenues and incentives for both PV producers and EV drivers. An economic model is developed to include the V2G expenses and revenues to provide a complete picture of the cost–benefit analysis. The proposed model is used to evaluate the economic feasibility of V2G services for PV power integration. <s> BIB002 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with transmission grids <s> In 2013, the California Independent System Operator published the "duck chart,"" which shows a significant drop in mid-day net load on a spring day as solar photovoltaics (PV) are added to the system. The chart raises concerns that the conventional power system will be unable to accommodate the ramp rate and range needed to fully utilize solar energy, particularly on days characterized by the duck shape. This could result in "overgeneration"" and curtailed renewable energy, increasing its costs and reducing its environmental benefits. This paper explores the duck chart in detail, examining how much PV might need to be curtailed if additional grid flexibility measures are not taken, and how curtailment rates can be decreased by changing grid operational practices. It finds that under business-as-usual types of assumptions and corresponding levels of grid flexibility in California, solar penetrations as low as 20 percent of annual energy could lead to marginal curtailment rates that exceed 30 percent. However, by allowing (or requiring) distributed PV and storage (including new installations that are part of the California storage mandate) to provide grid services, system flexibility could be greatly enhanced. Doing so could significantly reduce curtailment and allow much greater penetration of variable generationmore » resources in achieving a 50 percent renewable portfolio standard. Overall, the work described in this paper points to the need to fully integrate distributed resources into grid system planning and operations to allow maximum use of the solar resource.« less <s> BIB003 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with transmission grids <s> Distributed energy resources are able to provide services to grid operators, possibly with competing objectives. With the development of active distribution grid management, various market designs arise. Here, a reference market framework is considered, which allocates the available flexibility products according to requests coming from both distribution and transmission system operators. The goal of this paper is to provide an identification procedure that is able to detect, identify and catalogue possible conflicts among the involved stakeholders that take place when requesting and/or acquiring ancillary services from flexible units. The investigation is carried out considering a 3-area power system which allows to take into account local constraints as well as system-wide needs. As outcome, this paper identifies the conflicts from both a theoretical and a practical point of view, by means of descriptions/identification procedure and by visual examples, respectively. <s> BIB004 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Interaction with transmission grids <s> Workplace charging of electric vehicles (EVs) from photovoltaic (PV) panels installed on an office building can provide several benefits. This includes the local production and use of PV energy for charging the EV and making use of dynamic tariffs from the grid to schedule the energy exchange with the grid. The long parking time at the workplace provides the chance for the EV to support the grid via vehicle-to-grid technology, the use of a single EV charger for charging several EVs by multiplexing and the offer of ancillary services to the grid for up and down regulation. Further, distribution network constraints can be considered to limit the power and prevent the overloading of the grid. A single mixed integer linear programming (MILP) formulation that considers all the above applications has been proposed in this paper for a charging a fleet of EVs from PV. The MILP is implemented as a receding-horizon model predictive energy management system. Numerical simulations based on market and PV data in Austin, TX, USA, have shown 32% to 651% reduction in the net cost of EV charging from PV when compared to immediate and average rate charging policies. <s> BIB005

EVs can provide services to the transmission system. Those would mainly be frequency regulation and peak load mitigation. Large penetration of intermittent and variable power as PV brings the opportunity to contribute for EV fleets to smooth the PV generation at large scale. The first issue is the control of the global load curve. In many regions, PV generation peaks during low demand periods and is low during high-demand periods. As PV capacity increases, this leads to high power ramps in mornings and evenings, namely the so-called duck-curve effect BIB003 . In addition, a large penetration of commuter EV that charge during the evening should increase this effect. BIB001 studies a duck curve effect adding to a large EV fleet charging. The authors show that a smart charging displacing all EV loads from the global peak load, could diminish it up to 8% with 10% PV. [102] presents a method to design a cost-optimal planning locations and capacities of EV charging stations and PV plants in a territory. With a case study considering a transmission grid including 200000 EV charging in around 35 charging stations and 5 PV plants with a total capacity 90MWp, the authors show that PV energy manage to reduce the EV impact on the grid, and therefore allow substantial investments deferring. In BIB002 , PV producers are penalized for their forecasting errors. Penalties are defined by the product of the real-time electricity price, the forecasting error and a penalty factor which varies between three deviation band errors. We already saw that the EV/PV coupling could reduce PV power imbalances by a factor 10 and penalties by a factor 15. Moreover, an interesting sensibility analysis is made over forecast accuracy. Transmission system operator would rather deal with participants with good prediction accuracy and low forecasting deviations. A smart strategy would erase total penalties in an even more efficient way when forecast error is low. Nonetheless, more money is saved when forecast error is high. Moreover, with uncontrolled strategy, cost total penalty over accumulated forecast error is not dependent on the accumulated forecast error. On tre contrary, with the smart strategy, it increases quickly. Therefore, once a smart control strategy is used, it is beneficial to aim the smallest forecast error. In BIB005 , the authors compare the different effects of allowing a PVCS to trade energy, to offer reserve service, with an without V2G. They show that without V2G, they show that once offering reserve services, trading energy on the power market does not bring much advantage in comparison. On the other hand, trading is much more interesting once V2G is enabled. Another issue to be investigated is the possibility of competition between the participation of EV in ancillary services and in the EV/PV synergy. Generally, frequency regulation is the most needed when the transmission system is facing sudden large variation as during the daily peak load. Our insight is that the correlation of the peak load with the PV generation peak is an important feature to estimate if EVs can participate in frequency regulation while profiting its synergy with PV energy. This case of conflict can be related to those identified in BIB004 between TSO and DSO services. On the other hand, frequency markets are likely to be quickly saturated as soon as there will be millions of potential EV participants. As PV stabilization needs will keep on increasing, this hypothetical competition issue may be irrelevant on the long run.

Interactions between electric mobility and photovoltaic generation: A review <s> Regulations and policies for efficient EV/PV coupling <s> As part of France's political intentions to reduce CO2 emissions in building and transport, this work offers an analysis from the perspective of a public authority on the various innovation policies for the deployment of electric vehicles powered by solar photovoltaic electricity (PV) by the year 2030. This innovation involves enhancing the synergy between buildings and mobility by associating positive energy houses with electric vehicles. The vehicle battery, charged primarily using PV electricity, could also provide a way of storing intermittent solar-generated electricity for later use. <s> BIB001 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Regulations and policies for efficient EV/PV coupling <s> This study assesses the impact of electric vehicle (EV) uptake and large-scale photovoltaic (PV) investment on the economics of future electricity-generation portfolios. A Monte-Carlo-based portfolio modeling tool was used to assess the expected overall industry cost, associated cost uncertainty, and CO2 emissions of future generation portfolios, where both EVs and PV generation have achieved major deployment. The Australian National Electricity Market (NEM) was used as a case study under uncertain future fuel and carbon prices, electricity demand, and plant capital costs. Two EV charging scenarios were considered: 1)unmanaged charging which commences immediately as the EVs arrive at suitable charging infrastructure and 2)managed charging where EV charging loads are managed so that they better align with PV output. Results show that there are potentially valuable synergies between PV generation and EV charging demand in minimizing future electricity industry costs, cost uncertainties, and emissions, particularly when EV charging loads can be managed. The value of PV generation and managed EV charging is greater for higher EV fleet size and moderate carbon prices. <s> BIB002 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Regulations and policies for efficient EV/PV coupling <s> This paper presents a novel, empirical analysis of the most common business models for the deployment of demand response and energy management systems, electricity and thermal storage, and solar PV distributed energy resources. We classify the revenue streams, customer segments, electricity services provided, and resources for 144 business models. We use this assessment to identify a set of business model “archetypes” in each resource category. Our analysis leads us to five observations that have important implications for policymakers and regulators. First, our analysis highlights that business models are deeply embedded in myriad policy and regulatory frameworks. Second, current DER business models are driven more by regulatory and policy factors than by technological factors. Third, the relatively small set of well-defined DER business model archetypes suggests that the determinants of success within a given archetype may include executional capabilities, culture, and other activities that are not captured in our framework. Fourth, continued cost declines, technological innovation, and changing policy and regulatory landscapes mean the business models of tomorrow will likely look very different than the business models of today. Finally, DER business models compete within archetypes for market share in providing a limited set of electricity services. <s> BIB003

In former sections, we showed that electric mobility and photovoltaic generation would have an interesting cooperation potential, and would able to reduce total costs of both technology, along with their respective ecological impact. Also, the EV/PV coupling could even participate to the power system stabilization at distribution and transmission levels. In power systems, regulation and policies have a huge impact on technological development. For instance, recent studies are showing that regulatory and policy components have a prominent influence on distributed energy business models BIB003 . It is therefore crucial to identify which regulatory and political tools and frameworks are appropriate in order to exploit efficiently potentialities of EV/PV coupling. To our knowledge, only BIB001 provides a framework to analyze policies in favor of solar-powered electric mobility. Such policies are not necessarily specific to EV/PV synergy, such as carbon taxes BIB002 .

Interactions between electric mobility and photovoltaic generation: A review <s> Pricings <s> This paper investigates the opportunity for electricity loss minimization in commercial facilities where a photovoltaic system and electric vehicle charging stations coexist, through combined operation. A new metering architecture is proposed that can be used for coupling the two independent circuits and consequently reduce energy losses in grid side converters. The metering topology is valid for every kind of pricing (flat rates, feed-in tariffs, feed-in premiums). The problem is modeled as a non-linear optimization problem using BARON solver, taking into account the non-linear efficiency curve of the grid side converters. A case study with a sensitivity analysis on both photovoltaic system and electric vehicle charging fleet is carried out and useful results regarding losses minimization and profit increments are concluded. <s> BIB001 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Pricings <s> Renewable energy is increasingly replacing carbon-based technologies worldwide in electricity networks. This increases the challenge of balancing intermittent generation with demand fluctuation. DR (Demand response) is recognized as a way to address this by adapting consumption to supply patterns. By using DR technology, grid withdrawal of DSM (demand side management) devices such as heat pumps, electric vehicles or stationary batteries can be temporally shifted. Yet, the development of an accurate control and market design is still one of the greatest remaining DR challenges. <s> BIB002 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Pricings <s> The developments of battery storage technology together with photovoltaic (PV) roof-top systems might lead to far-reaching changes in the electricity demand structures and flexibility of households. The implications are supposed to affect the generation mix of utilities, distribution grid utilization, and electricity price. Using a techno-economic optimization model of a household system, we endogenously dimension PV system and stationary battery storage (SBS). The results of the reference scenario show positive net present values (NPV) for PV systems of approx. 500–1,800EUR/kWp and NPV for SBS of approx. 150–500EUR/kWh. Main influences are the demand of the households, self-consumption rates, investment costs, and electricity prices. We integrate electric vehicles (EV) with different charging strategies and find increasing NPV of the PV system and self-consumption of approx. 70%. With further declining system prices for solar energy storage and increasing electricity prices, PV systems and SBS can be profitable in Germany from 2018 on even without a guaranteed feed-in tariff or subsidies. Grid utilization substantially changes by households with EV and PV-SBS. We discuss effects of different incentives and electricity tariff options (e. g. load limits or additional demand charges). Concluding, solar energy storage systems will bring substantial changes to electricity sales. <s> BIB003

It is generally accepted that dynamic pricing leads to a better economic efficiency, and particularly in demand response studies BIB002 . Especially because they reflect a better reality of the different costs. In our literature, BIB001 provides a comparison of the efficiency of a EV/PV system with different tariff structure. The authors find that revenues of a EV/PV system are increased by 0.5-1% with dynamic tariffs compared to flat tariffs, while power losses decrease by 10-20%. BIB003 studies the effectiveness of different electricity pricing on the profitability of PV-ESS system for German households. First, the high price of electricity in Germany makes investments in self-consumption more profitable. Then, the authors study different network tariff designs, which account for 22% of the electricity price in Germany. Those tariffs are a reference volumetric tariff and the following alternative tariffs: a fixed charge, a maximum peak load tariff and a tariff combining the two previous tariffs. The authors find comparable decrease of the net present value of the installation for all alternative impact of the alternative tariffs compared to the volumetric tariff. Nevertheless, the decrease is lesser when there is a smart charging strategy for EVs. In addition, the authors determine that time-of-use tariffs does not significantly improve the performance of the EV smart control.

Interactions between electric mobility and photovoltaic generation: A review <s> Entry conditions <s> The electric power sector is once again evolving. A variety of distributed energy resources and improving computation, communication, and control technologies create an unprecedented degree of choice for electricity consumers, choices that are poorly guided by electricity rates and other incentives designed for a comparatively simpler era. These technologies also create new tools for regulated utilities, competitive suppliers, and other businesses to employ in the provision of electricity services. This paper summarizes the findings of a two-year, multidisciplinary MIT Energy Initiative research effort, the Utility of the Future study, and outlines a framework for proactive electricity regulation, market, and policy reform designed to enable the efficient evolution of the power sector over the next decade and beyond. Recommendations include a comprehensive system of efficient prices and charges for all electricity users, enhanced regulation of distribution utilities, careful reconsideration of industry structure to avoid conflicts of interest, and improvements to electricity markets. Together, this framework is intended to establish a level playing field for the provision and consumption of electricity services and enable the integration of a cost-effective combination of centralized generation, conventional network assets, and emerging distributed resources, whatever that mix may be. <s> BIB001 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Entry conditions <s> Increasing environmental concerns are driving an evolution of the energy system in which electric vehicles (EVs) play an important role. Still, as the EV number increases, the adverse impact of charging is observed more widely, especially at the low-voltage level where high EV concentrations cause various detrimental effects due to the coincidence between EV charging and residential peak load. However, if managed properly, EVs become flexible resources which can improve the system operation, making them an attractive asset for the distribution system operator. With the recent technology development, new forms of local EV support can be developed, provided that an appropriate regulatory framework is established. Whereas the technical value of such EV distribution grid services has already been proven, integrating them into the European regulatory context is not straightforward. In the context where active distribution grid management schemes are still to be developed, it is important to recognise the barriers for active EV involvement in the early stage of the development. This manuscript focuses on identifying these barriers from a technology and infrastructure perspective as well as from the regulatory and market aspect. Various policy recommendations are provided for the stakeholders involved in the EV value chain. <s> BIB002 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Entry conditions <s> As the share of intermittent renewable energy sources increases, it will be necessary to increase the volume of frequency-regulation reserves. New sources of reserves can be found in distributed energy resources: controllable loads, Electric Vehicles (EVs), distributed generation units. However, it is necessary to adapt the frequency-regulation reserve market design to allow these new resources to participate through new market players known as “aggregators”. The aim of this paper is to provide a modular framework to analyze frequency regulation markets or mechanisms in order i) to identify barriers to entry for aggregators and ii) to identify some options to overcome these barriers. A case study of the situation in 2016 of four major European frequency regulation markets is performed to show how this framework could be used by aggregators or policy-makers. <s> BIB003

Regulation of the power system must adapt to facilitate new technologies diffusion (distributed renewable energies and information and communication technologies) BIB001 . The design of those regulation reforms would have to embrace lots of elements, while being conceived dynamically. In EV/PV literature, regulatory is barely formally addressed, but some insights for appropriate regulation tools can be interpreted. In the aforementioned literature, we saw that a new actor, the aggregator, was often invoked to manage jointly EV fleets charging and distributed photovoltaic generation. defines the different types of aggregators in evolving power systems. The authors distinguish fundamental, transitory and opportunistic values of aggregation. A fundamental value lies in intrinsic economies of scale of the aggregator services. A transitory value is an aggregator service that could be potentially substituted by an appropriate regulation. An opportunistic value comes from inefficient or inappropriate regulation, and is more likely to disturb the power system rather than help it. EV/PV aggregators require a new regulation whether they would operate at the transmission level (frequency regulation...) or at the distribution level. Aggregated EV/PV systems in energy and ancillary markets require a novel regulation at transmission and distribution level, that should include all distributed energy resources. The entry conditions of such systems should be similar to the ones required by EV aggregators in order to participate in those markets. Frequency market is thought to be the most realistic market for EVs to participate . For many countries (France, USA), this would require a deep reform of energy markets design BIB003 . Nonetheless, allowing EV aggregators to contract with a PV powerstation to lower its forecast errors could be an alternative and/or complementary solution as EV participation in ancillary markets. We saw that EV/PV coupling could be particularly beneficial for the distribution system and its operator. EV/PV systems in households, buildings and charging station should have entry issues as the site manager is in charge of the power system of the spatial configuration. For larger configuration, there is not yet any framework enabling DSO to contract flexibility services . A recent work ( BIB002 ) proposes a regulation framework and road-map for EV integration in distribution systems.

Interactions between electric mobility and photovoltaic generation: A review <s> Support mechanisms <s> The paper analyses the economic and environmental benefits of charging electric vehicles (EV) at workplaces in the Netherlands using photovoltaic panels (PV). A 10kW EV-PV charging system is used to charge the electric cars directly from photovoltaic panels. The cost of using a gasoline vehicle is compared with that of an electric vehicle that is charged from the grid or from solar panels. It is found that charging EV from PV results in huge savings in fuel cost, taxes and lower CO2 emissions. A comparison is made for solar panels installed on rooftops and as a solar carport and the impact of feed-in tariffs on PV generation revenues is evaluated. <s> BIB001 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Support mechanisms <s> This research uses multi-objective optimisation to determine the optimal mixture of energy and transportation technologies, while optimising economic and environmental impacts. We demonstrate the added value of using multi-objective mixed integer linear programming (MOMILP) considering economies of scale versus using continuous multi-objective linear programming assuming average cost intervals. This paper uses an improved version to solve MOMILPs exactly. To differentiate optimal solutions with and without subsidies, the impact of policy on the Pareto frontier is assessed. We distinguish between minimising economic life cycle costs (complete rationality) and required investments (bounded rationality). The approach is illustrated using a Belgian company with demands for electricity and transport. Electricity technologies are solar photovoltaics and the grid; transportation includes internal combustion engine vehicles, grid powered battery electric vehicles (BEVs), and solar-powered BEVs. The impact of grid powered BEVs to reduce GHG emissions is limited, yet they are less costly than solar panels to decrease emissions. Current policy measures are found to be properly targeting rational investors who consider life cycle costs, while private (potentially bounded rational) investors often focus on required investments only. <s> BIB002 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Support mechanisms <s> Abstract The direct coupling of solar electricity with electric vehicles is very topical due to their many co-benefits, such as reducing the electricity flow into and from the grid. This work illustrates this solution using a park-and-ride lot outside Lisbon to study its solar energy potential to charge electric vehicles. The solar resource, its exploitation and coordination with the vehicles under different charging approaches are discussed in detail, using methods extendible to other cases. A genetic-algorithm optimisation was performed to maximise the solar energy generation. To ascertain the techno-economic feasibility of the concept, several sets of assumptions are analysed, including the use of energy storage and smart charging. For current market conditions, the payback time is found to be 14 years; a modest financial public incentive would significantly improve the project economics – making the payback drop to 7 years – enabling conditions to put this inescapable and green approach at the forefront. <s> BIB003 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Support mechanisms <s> Rising greenhouse gas emissions raise the risk of severe climate change. The household sector׳s greenhouse gas emissions have increased over time as more people drive gasoline cars and consume electricity generated using coal and natural gas. The household sector׳s emissions would decline if more households drove electric vehicles and owned solar panels. In recent years automobile manufacturers have been producing high-performance electric vehicles, and solar panels are becoming more efficient and less expensive. Using several data sets from California, we document evidence of the growth of the joint purchase of electric and hybrid vehicles and solar panels. We discuss pricing and quality trends for these green durable goods. <s> BIB004 </s> Interactions between electric mobility and photovoltaic generation: A review <s> Support mechanisms <s> Corrections were made to author information for the paper, “Energy management system with PV power forecast to optimally charge EVs at the workplace,” (van Der Meer, D., et al), IEEE Trans. Ind. Informat., vol. 14, no. 1, pp. 311–320, Jan. 2018. <s> BIB005

Feed-in tariffs and other subsidies are commonly used by governments to stimulate PV technology development while EV is also heavily subsidized in many countries. By taking different levels of FITs, BIB005 shows that the lower the FIT is, the more efficient the EV/PV coupling will be. Indeed, high FITs incentivize PV capacity owners to sell their whole production to the grid instead of cooperating with an EV fleet, that have to charge only using the grid. BIB001 shows that low FIT make profitable investments on both PV and EV compared to PV and classical vehicles. In a Belgium context, BIB002 uses a multi-objective framework (investment costs and ecological footprint under the constraints of transport and power needs) to design company transport fleet and power system. The authors exhibit the Pareto of technology combinations investment in function of life-cycle CO 2 emissions and either initial required investments (bounded rationality) or life-cycle economic cost (complete rationality). Without subsidy, the authors show a clear trade-off between economical (classical vehicle and grid electricity) and ecological (EV and PV electricity) options. Applying the actual governmental subsidies on both PV and EV make the EV/PV coupling the economic optimum as well as ecological for an investor with complete rationality. Nonetheless, if the investor has a bounded rationality, the latter result is less clear. In a similar way, BIB003 studies the influence of investment subsidies on the payback time of a PVCS. Payback time decreases linearly with the level of subsidy, from 14 years without subsidy to 6 years with a 50% investment subsidy. Synergy between EV and PV raises an issue on the efficiency of public policy. In many countries, both technologies are subsidized, while the synergy between them is not considered in the policy. Thus the efficiency of combined separated subsidy policies seems questionable, especially since these technologies are more and more likely to be acquired jointly, as pointed out in BIB004 . Therefore, it is also possible that subsiding features of the EV/PV synergy, as PVCS, smart buildings, or energy management systems, would be particularly efficient for both EV and PV development. To our opinion, this issue deserves a more detailed investigation.

Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> In this paper, we bring techniques from operations research to bear on the problem of choosing optimal actions in partially observable stochastic domains. We begin by introducing the theory of Markov decision processes (mdps) and partially observable MDPs (pomdps). We then outline a novel algorithm for solving pomdps off line and show how, in some cases, a finite-memory controller can be extracted from the solution to a POMDP. We conclude with a discussion of how our approach relates to previous work, the complexity of finding exact solutions to pomdps, and of some possibilities for finding approximate solutions. <s> BIB001 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> In this paper, we make a brief study of some of the important requirements of a structural monitoring system for civil infrastructures and explain the key issues that are faced in the design of a suitable wireless monitoring strategy. Two-tiered wireless sensor network architecture is proposed as a solution to these issues and the protocol used for the communication in this network is described. The power saving strategies at various levels, from the network architecture, to communication protocol, to the sensor unit architecture are explained. A detailed analysis of the network is done and the implementation of this network in a laboratory setting is described. <s> BIB002 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> Optimal use of energy is a primary concern in fielddeployable sensor networks. Artificial intelligence algorithms offer the capability to improve the performance or sensor networks in dynamic environments by minimizing energy utilization while not compromising overall performance. However, they have been used only to a limited extent in sensor networks primarily due to their expensive computing requirements. We describe the use of Markov decision processes for the adaptive control of sensor sampling rates in a sensor network used for human health monitoring. The MDP controller is designed to gather optimal information about the patient's health while guaranteeing a minimum lifetime of the system. At every control step, the MDP controller varies the frequency at which the data is collected according to the criticality of the patient's health at that time. We present a stochastic model that is used to generate the optimal policy offline. In cases where a model of the observed process is not available a-priori. we descrihe a Q-learning technique to learn the control policy, by using a pre-existing master controller. Simulation results that illustrate the performance of the controller are presented. <s> BIB003 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> This paper addresses the optimal power and rate allocation control in multiple-input multiple-output (MIMO) wireless systems over Markovian fading channels. The problem is posed as an infinite horizon average-cost constrained Markov decision process (CMDP) with the goal of minimizing the average transmission power subject to delay constraints. By using a Lagrangian formulation of the CMDP, we use the concepts of stochastic dominance, submodularity, and multimodularity to prove that the optimal randomized policies are monotone. Three important structural results on the nature of the optimal randomized policies are derived. First, we show that the action space can be exponentially reduced by decomposing the rate allocation problem into bit-loading problem across individual antennas and the total rate allocation based on the current buffer occupancy and channel state. Second, we show that the optimal rate allocation policy is a randomized mixture of two pure policies that are monotonically increasing in the buffer occupancy. Finally, we show that the optimal power allocation is piecewise linear in the delay constraint. These three structural results can be exploited to devise efficient online reinforcement learning algorithms for optimal rate allocation. <s> BIB004 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> Wireless Body Area Networks (WBANs) of intelligent sensors emerged as the most promising architecture for ambulatory health monitoring. Integrated in an m-Health system they provide means for unobotrusive ubiquitous monitoring. However, proper design of wireless communication system is application dependent and critically influences system performance and user acceptance. In this paper we present a survey of existing and emerging wireless communication technologies and critical parameters for the system design. <s> BIB005 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> Body area sensors can enable novel applications in and beyond healthcare, but research must address obstacles such as size, cost, compatibility, and perceived value before networks that use such sensors can become widespread. <s> BIB006 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> Wireless body area networks (WBANs) offer many promising new applications in the area of remote health monitoring. An important element in the development of a WBAN is the characterization of the physical layer of the network, including an estimation of the delay spread and the path loss between two nodes on the body. This paper discusses the propagation channel between two half-wavelength dipoles at 2.45 GHz, placed near a human body and presents an application for cross-layer design in order to optimize the energy consumption of different topologies. Propagation measurements are performed on real humans in a multipath environment, considering different parts of the body separately. In addition, path loss has been numerically investigated with an anatomically correct model of the human body in free space using a 3-D electromagnetic solver. Path loss parameters and time-domain channel characteristics are extracted from the measurement and simulation data. A semi-empirical path loss model is presented for an antenna height above the body of 5 mm and antenna separations from 5 cm up to 40 cm. A time-domain analysis is performed and models are presented for the mean excess delay and the delay spread. As a cross-layer application, the proposed path loss models are used to evaluate the energy efficiency of single-hop and multihop network topologies. <s> BIB007 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> With the growing amount of old people in world population, the concept of Body Area Network(BAN) is expected to be an important technology in the near future, which could provide the mobility health care for the patients with wearable devices who are roaming in the coverage of a medical monitoring system. However, there exist some challenges to the application of BAN, such as power consumption, bandwidth limitations, tissue protection (absorbed by human body), diverse data rates supporting and coexistence and interference. In order to provide a better understanding of the research challenges of BAN, this article presents a detailed investigation. <s> BIB008 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> We present a framework based on Markov decision process to optimize software on mobile phones. Unlike previous approaches in literature that focus on energy optimization while meeting a specific task-related time constraint, we model the desired talk-time as an explicit user given parameter and formulate the optimization of resources such as battery-life on a mobile phone as a decision processes that maximizes a user specified application specific reward or utility metric while meeting the talk-time constraint. We propose efficient techniques to solve the optimization problem based on dynamic programming and illustrate how it can be used in the context of realistic applications such as WiFi radio power optimization and email synchronization. We present a design methodology to use the proposed technique and experimental results using the Android platform from Google running on the HTC mobile phone. <s> BIB009 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> Advances in wireless communication technologies, such as wearable and implantable biosensors, along with recent developments in the embedded computing area are enabling the design, development, and implementation of body area networks. This class of networks is paving the way for the deployment of innovative healthcare monitoring applications. In the past few years, much of the research in the area of body area networks has focused on issues related to wireless sensor designs, sensor miniaturization, low-power sensor circuitry, signal processing, and communications protocols. In this paper, we present an overview of body area networks, and a discussion of BAN communications types and their related issues. We provide a detailed investigation of sensor devices, physical layer, data link layer, and radio technology aspects of BAN research. We also present a taxonomy of BAN projects that have been introduced/proposed to date. Finally, we highlight some of the design challenges and open issues that still need to be addressed to make BANs truly ubiquitous for a wide range of applications. <s> BIB010 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> In this paper, we provide a comprehensive survey of recent energy-efficient medium access control (MAC) protocols for wireless body area networks (WBANs) and presents a comparison of the various approaches pursued. At the outset, we outline the crucial attributes for a good MAC. Several sources that contribute to the energy inefficiency are identified. Then, we investigate few MAC protocols devised for WBAN by emphasizing their salient features. As a conclusion, we put forward a number of open research challenges with regard to prospects of medium access techniques and other issues. <s> BIB011 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Introduction <s> Body Area Networks (BANs) are only recently beginning to revolutionize e-healthcare by making monitoring of various patients accessible and convenient for physicians and patients. Past research in BANs has focused on biosensor design, signal processing, power efficiency, and wireless protocols. However, numerous challenges plague its development and implementation to this day. One of these is the dependability — a conjunction of reliability, security and availability — from both an intra-BAN and extra-BAN communications perspective. Using a generic, divided system model (intra-BAN and extraBAN), we identify potential areas of failure. The paper discusses and classifies issues and schemes proposed to increase dependability. Finally, upcoming challenges and new directions of research for improving the dependability of BANs are presented. <s> BIB012

The recent advancements in hardware miniaturization facilitate the realization of many applications especially in the medical and health fields. Recently, Wireless Body Area Networks (WBANs) attracted significant attention. It consists of small, intelligent devices attached on or implanted in the body which are capable of establishing a wireless communication link. These devices provide continuous health monitoring and real-time feedback to the user or medical personnel BIB010 . These networks typically have two types of devices: sensors and actuators. Sensors are used to acquire or measure certain conditions of the human body externally such as body temperature, or internally such as glucose level in the blood. On the other hand, the actuators are concerned in the next step to be taken after examining the recorded measurements. These two devices sometimes work in conjunction with each other. For example, for a diabetic patient who requires insulin injection, when the sensor measures the glucose level, the actuator can be equipped with a built-in reservoir and pump which administers the correct dose of insulin to be given based on the glucose level measurements by the sensor BIB010 . WBAN has numerous possible applications; some are shown in Fig. 1 . BIB006 The Wireless Body Area Networks can be differentiated from the regular Wireless sensor networks by the following distinguished characteristics BIB010 . Firstly, the devices used are very small (less than 1cm BIB012 ) and have limited energy resources. Secondly, although the wanted lifetime is expected to extend to several years or even decades for implanted devices, for most devices, it is impossible to recharge or change the batteries. Thirdly, all devices are equally important and no redundant devices are available. Fourthly, to cope with health concerns and minimize interference, an extremely low transmit power per node is needed. Fifthly, the surrounding environment (i.e. human body) is a (very) lossy medium BIB007 and consequently, the waves are attenuated considerably before they reach the receiver. In addition, because the devices are located on the human body that can be in motion, the deployed WBANs should be robust against frequent changes in the network topology. More importantly, WBAN should be very dependable BIB010 BIB012 where high reliability, low delay and stringent security mechanisms are required. Finally, the devices operating in WBANs are often very heterogeneous and they may have very different demands or may require different resources of the network in terms of data rates, power consumption and reliability BIB010 . Having stated the above challenges, the literature is rich with works trying to find solutions for these problems. Actually, several papers have surveyed these works BIB008 - BIB006 The authors in BIB005 surveyed the existing and emerging wireless communication technologies and critical parameters for the system design. While the authors in BIB011 focused on surveying the recent energy-efficient medium access control (MAC) protocols for wireless body area networks (WBANs) along with a number of open research challenges with regard to prospects of medium access techniques and other issues. In , a survey of wireless Body Area Network (WBAN) was preseneted and the authors introduced the notion of a virtual doctor server (VDS) in existing WBAN architecture. This archticture supports various patient health care services and it keeps the historical data about the patient, generate the daily tips and advices for him, call the doctor or emergency squad if required and can provide first aid assistance instructions on patient or any of his close relative's PDA's. In BIB006 , the authors surveyed the expected challenges in realizing WBANs for diverse applications. However, according to the authors' knowledge, there is no published work that surveys the recent literature for new techniques or methods that solve the related optimization problems. One of the main challenges that face the WBAN designer is the high uncertainty in the status of the monitored system which leads to major difficulties in decision making process related to optimizing the activities of the implanted sensors, for example. Partially observable Markov decision process (POMDP) is a generalization of Markov decision processes (MDPs) that provide a mathematical framework for modeling decision-making in situations where outcomes are partly random and partly under the control of a decision maker. It provides a rich framework for agent planning under uncertainty BIB001 . Implantation of devices like biosensors in human body is a critical operational issue and application that requires efficient and accurate monitoring and management. MDPs have been applied in different facets of life such as medicine, education, quality control, scheduling, data communication, scheduling and so on . In most of these domains, the full and exact state of the system is known. Specifically, amongst the many applications are optimizations of software in mobile phones BIB009 . The authors formulate the optimization of resources such as battery life in a mobile phone as a decision processes and proposed a dynamic programming techniques to solve the problem. In BIB004 , the authors presented an optimal power and rate allocation control in multiple input and multiple output (MIMO) wireless system over Markovian fading channels. Earlier works for the use sensors for patient health monitoring include; lifeGuard BIB002 codeBlue , and Ubimon . However, energy management and the environmental constraints are challenges facing these systems. To tackle this energy resource management problem, MDPs was introduced in the effort to design optimally performing system. In BIB003 , MDPs was used to design MDP controllers for optimal energy utilization and information gathering in a health monitoring sensor network. However, the author concluded that POMDPs has greater representational power in the quest to manage this sort of wireless networks. Due to arrays of possible applications, various algorithms, methods and techniques have been used or proposed to solve this type of Markov decision processes.

Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> A Motivating Problem <s> In many practical reinforcement learning problems, the state space is too large to permit an exact representation of the value function, much less the time required to compute it. In such cases, a common solution approach is to compute an approximation of the value function in terms of state features. However, relatively little attention has been paid to the cost of computing these state features. For example, search-based features may be useful for value prediction, but their computational cost must be traded off with their impact on value accuracy. To this end, we introduce a new cost-sensitive sparse linear regression paradigm for value function approximation in reinforcement learning where the learner is able to select only those costly features that are sufficiently informative to justify their computation. We illustrate the learning behavior of our approach using a simple experimental domain that allows us to explore the effects of a range of costs on the cost-performance trade-off. <s> BIB001 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> A Motivating Problem <s> Wireless Sensor Networks (WSNs) are embedded networks made up of tiny wireless devices referred to as sensors. They can be distinguished based on the type of sensing elements used in their sensors. Two important classes of WSNs are directional and biological WSNs. The former is given its name because the sensing region of a directional sensor is viewed as a sector in a two-dimensional plane. The latter, however, is given its name because its sensing elements are biological materials like enzymes. One of the major challenges introduced by directional WSNs is the fact that the existing mathematical models developed for conventional WSNs cannot directly be used for solving the directional sensor placement and configuration problems. Therefore, new optimization models which capture the primary parameters characterizing directional sensors are necessary. On the other hand, a major challenge introduced by biological WSNs is the heat generated as a result of power consumption and wireless radiation. When biological WSNs are operated in temperature-sensitive environments like the human body, the temperature of the surrounding tissues might rise. The tissues might be damaged if the maximum safe temperature level is exceeded. Therefore, thermal management techniques are indispensible. The first objective of the work presented in this dissertation is to develop optimization models for the planning of directional WSNs. Toward that end, the existing literature is critiqued and gaps are identified. Then, three fundamental planning problems are presented. The problems are formulated as Integer Linear Programming <s> BIB002

In this section, we present an example of WBAN problem that can be formulated and solved using POMDP. Consider the problem of implanting a subcutaneous biosensor in a human body. This biosensor in required to measure some physiological and metabolic vital signals such as temperature, pulse rate and glucose level. The operation and monitoring of these biosensors need to be accurate and efficient due to the critical nature of the application. The biosensor communicates wirelessly with a base station outside the body. The state of the system can be model as three variables vector (temperature, Energy and channel state which changes over time. The objective is to minimize the expected total energy consumption and maximize the number of measurements (or samples) subject to constraints such as the maximum temperature increase and quality of service. In order to study the above system, it is modeled as a POMDP since its state cannot be completely observed. The formulated model is solved using available techniques to obtain optimal policy, then these policies will be characterized and new heuristics are proposed based on reinforcement learning BIB001 . There are limited research works on planning and management of biosensors, considering the increase in research for the design and development of biosensors and the possible increase in area of related applications. Therefore, there is a need for a comprehensive research on the optimization approaches of this class of wireless sensors. Most of the existing research efforts are dedicated to traditional wireless sensor network. Biosensors, in addition, to the general features of wireless sensor network, they have additional external factors from the environment such as temperature that can affect its performance. The background of this work is found in BIB002 . In this work, the author modeled the problem as Markov Decision Process (MDP) where the state of the system is assumed to be deterministic, but this is not the case in real life situations where decisions are made based on the observation of the current state which unknown apriori. The most suitable model is to use partial observable markov decision process. The remaining part of this article is organized as follows; Section 2 described in detail, the POMDPs and some solutions techniques. In Section 3, the use of POMDPs framework in wireless sensor network is presented. Section 4 presented two recent real systems that employed POMDP for optimizing the perfromance of WBANs. An overview of applications of POMDPs in other fields of human endeavor was provided in section 5, while the challenges facing this framework is presented in section 6 and finally section 7 concluded the article.

Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Partially Observable Markov Decision Process (POMDPs) <s> In this paper, we bring techniques from operations research to bear on the problem of choosing optimal actions in partially observable stochastic domains. We begin by introducing the theory of Markov decision processes (mdps) and partially observable MDPs (pomdps). We then outline a novel algorithm for solving pomdps off line and show how, in some cases, a finite-memory controller can be extracted from the solution to a POMDP. We conclude with a discussion of how our approach relates to previous work, the complexity of finding exact solutions to pomdps, and of some possibilities for finding approximate solutions. <s> BIB001 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Partially Observable Markov Decision Process (POMDPs) <s> Wireless body area networks (WBANs) offer many promising new applications in the area of remote health monitoring. An important element in the development of a WBAN is the characterization of the physical layer of the network, including an estimation of the delay spread and the path loss between two nodes on the body. This paper discusses the propagation channel between two half-wavelength dipoles at 2.45 GHz, placed near a human body and presents an application for cross-layer design in order to optimize the energy consumption of different topologies. Propagation measurements are performed on real humans in a multipath environment, considering different parts of the body separately. In addition, path loss has been numerically investigated with an anatomically correct model of the human body in free space using a 3-D electromagnetic solver. Path loss parameters and time-domain channel characteristics are extracted from the measurement and simulation data. A semi-empirical path loss model is presented for an antenna height above the body of 5 mm and antenna separations from 5 cm up to 40 cm. A time-domain analysis is performed and models are presented for the mean excess delay and the delay spread. As a cross-layer application, the proposed path loss models are used to evaluate the energy efficiency of single-hop and multihop network topologies. <s> BIB002

A partially observable Markov decision process (POMDP) is a generalization of Markov decision processes that provide a mathematical framework for modeling decision-making in situations where outcomes are partly random and partly under the control of a decision maker. It provides a rich framework for agent planning under uncertainty BIB001 . A POMDP can be used to model an agent making a real world sequential decision after a set of actions under uncertainty to achieve a goal BIB001 . It is a model for deciding how to act in ``an accessible, stochastic environment with a known transition model'' . A discrete-time POMDP models the relationship between an agent and its environment. A POMDP is described by the following set of parameters:  a set of states S ={s 1 , s 2 , s 3 …, s |s| }  a set of actions A= {a 1 , a 2 , a 3 …a |A|}  a set of observations O= {o 1 , o 2 , o 3 ,…,o |o| }  a set of transition probabilities T= (s i , a,s j )= P(s j | s i , a )  a set of observation probabilities β (si, a, o) = p(oi| sj, a).  a set of rewards R: Formally, a POMDP is specified as a tuple (S, A, O, T, β, R), where S is a set of states, A is a set of actions, and O is a set of observations. In each time step, the agent lies in some state s in S; it takes some action a in A and moves from s i to a new state s j . Due to the uncertainty in action, the end state s i is modeled as a conditional probability function T(s i , a, s j )= P(s j | s i , a ) which models the effect of action and gives the probability that the agent transits to s j , after taking action a from state s. The agent then relates an observation to information on its state. Due to the uncertainty in observation, the observation result o in O is also modeled as a conditional probability function β (s, a, o) = p(o | s, a). At each step, the agent receives a real-value reward model β (s, a), if it takes action a from state s, and the agent's goal is to maximize its expected total reward by choosing a suitable sequence of actions. A reward function R is defined as R: S x A  R . This function is used to influence the agent's characteristics. In POMDP, the objective is to plan, planning means computing an optimal policy that maximizes the expected total discounted or long term reward over an infinite horizon. Since the state of the agent's state is not completely visible, that is, partially observable, this leads us to rely on belief b, which is a probability distribution over S. The initial belief is denoted as b o (s). A POMDP policy π maps a belief b to a prescribed action a in A. The expected reward for policy π starting from belief b is defined as Where γ < 1 is the discount factor. The optimal policy π * is obtained by optimizing the long-term reward. ( ) BIB002 Where b 0 is the initial belief. The optimal policy, noted π * yields the highest expected reward value for each belief state, compactly represented by the optimal value function, noted V * . The value function V * (b) is a solution to Bellman optimality equation as follows. Exact solutions to POMDP problems are intractable, and various approximate solutions techniques were developed.

Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> A review of Solution Techniques for POMDP <s> The thesis develops methods to solve discrete-time finite-state partially observable Markov decision processes. For the infinite horizon problem, only discounted reward case is considered. ::: For the finite horizon problem, two new algorithms are developed. The first algorithm is called the relaxed region algorithm. For each support in the value function, this algorithm determines a region not smaller than its support region and modifies it implicitly in later steps until the exact support region is found. The second algorithm, called linear support algorithm, systematically approximates the value function until all supports in the value function are found. The most important feature of this algorithm is that it can be modified to find an approximate value function. It has been shown that these two algorithms are more efficient than the one-pass algorithm. ::: For the infinite horizon problem, it is first shown that the approximation version of linear support algorithm can be used to substitute the policy improvement step in a standard successive approximation method to obtain an $\epsilon$-optimal value function. Next, an iterative discretization procedure is developed which uses a small number of states to find new supports and improve the value function between two policy improvement steps. Since only a finite number of states are chosen in this process, some techniques developed for finite MDP can be applied here. Finally, we prove that the policy improvement step in iterative discretization procedure can be replaced by the approximation version of linear support algorithm. ::: The last part of the thesis deals with problems with continuous signals. We first show that if the signal processes are uniformly distributed, then the problem can be reformulated as a problem with finite number of signals. Then the result is extended to where the signal processes are step functions. Since step functions can be easily used to approximate most of the probability distributions, this method can be used to approximate most of the problems with continuous signals. Finally, we present some conditions which guarantee that the linear support can be computed for any given state, then the methods developed for finite signal cases can be easily modified and applied to problems for which the conditions hold. <s> BIB001 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> A review of Solution Techniques for POMDP <s> Markov decision processes (MDP''s) are a mathematical formalization of problems in which a decision-maker must choose how to act to maximize its reward over a series of interactions with its environment. Partially observable Markov decision processes (POMDP''s) generalize the MDP framework to the case where the agent must make its decisions in partial ignorance of its current situation. This paper describes the POMDP framework and presents some well-known results from the field. It then presents a novel method called the witness algorithm for solving POMDP problems and analyzes its computational complexity. The paper argues that the witness algorithm is superior to existing algorithms for solving POMDP''s in an important complexity-theoretic sense. <s> BIB002 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> A review of Solution Techniques for POMDP <s> We describe a point-based approximate value iteration algorithm for partially observable Markov decision processes. The algorithm performs value function updates ensuring that in each iteration the new value function is an upper bound to the previous value function, as estimated on a sampled set of belief points. A randomized belief-point selection scheme allows for fast update steps. Results indicate that the proposed algorithm achieves competitive performance, both in terms of solution quality as well as speed. <s> BIB003 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> A review of Solution Techniques for POMDP <s> Hybrid approximate linear programming (HALP) has recently emerged as a promising approach to solving large factored Markov decision processes (MDPs) with discrete and continuous state and action variables. Its central idea is to reformulate initially intractable problem of computing the optimal value function as its linear programming approximation. In this work, we present the HALP framework and discuss several representational and computational issues that make the approach appropriate for large MDPs. We compare three different methods for solving HALP and demonstrate the feasibility of the approach on high-dimensional distributed control problems. <s> BIB004 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> A review of Solution Techniques for POMDP <s> Partially observable Markov decision processes (POMDPs) provide an elegant mathematical framework for modeling complex decision and planning problems in stochastic domains in which states of the system are observable only indirectly, via a set of imperfect or noisy observations. The modeling advantage of POMDPs, however, comes at a price -- exact methods for solving them are computationally very expensive and thus applicable in practice only to very simple problems. We focus on efficient approximation (heuristic) methods that attempt to alleviate the computational problem and trade off accuracy for speed. We have two objectives here. First, we survey various approximation methods, analyze their properties and relations and provide some new insights into their differences. Second, we present a number of new approximation methods and novel refinements of existing techniques. The theoretical results are supported by experiments on a problem from the agent navigation domain. <s> BIB005

In this section, we present an overview of classical algorithms for solving POMDP and investigate the improvements so far in both exact and approximate methods. Solving a POMDP exactly is an intractable problem. Due to computational explosion of results, most efforts are tailored toward the management of the constraints to avoid the exponential complexity increase. The developed methods and algorithms can be classified under various categories such as exact value iteration, policy iteration, witness and greedy solution. Some of the approximate algorithms are either optimal or suboptimal. The entire algorithms would be based on the above classification. Also, most POMDP algorithms depend on dynamic programming; some efforts have been exerted in designing algorithms for updating the dynamic programming. Classical methods for solving POMDP are value iteration , approximate (heuristic) solution techniques in POMDP aimed at reducing the complexity BIB005 . Also, a fast-point based algorithm for POMDP was proposed in BIB003 , it ensures that in each iteration of value function, the new value functions are upper bounded. Another promising technique is approximate Linear programming for solving hybrid factored MDPs BIB004 . The first exact method for solving POMDP is called One pass BIB004 , here, the algorithm chooses an arbitrary belief point and defines a set of constraints and it constructs vectors for each point. In this case, the constraint in the believe space is guaranteed to dominate. The region defines the intersection of three regions. A vector is constructed for that region and the strategy is best for the belief point and some nearby belief point. In , an approximate algorithm called Enumeration algorithm that concentrates on how to generate sets of points by generating all possible sets of vectors by selecting action for each observation in the vectors. In this case, a large number of vectors will be generated and some might be useless. The useless ones can be eliminated to decrease the computation time. An extension to the enumeration algorithm is Incremental pruning BIB004 using interleaved generation and redundancy testing breaking up the dynamic programming update of the value function and the maximizing set of alpha vectors is determined from bottom to top leading to iterative purging of smaller set of alpha vectors. Grid based Algorithm is another example of approximation algorithm in which the value function is determined for sets of points and interpolated to evaluate the optimal action that would be encountered for other belief state that are not in the grid point. Linear Support , is a modification of idea by choosing less strict constraints and not focusing on actions and future course of action . It simply picks a point, generate vectors for the point and check if the region of the vectors is correct for all corners. Corners mean vertices of the region. In BIB001 , the author presented a different idea called Witness algorithm by focusing on the point that the vector is not dominated, it concentrates on the best value function. Instead of finding V and prune it down. It focused on finding V* (optimal policy) directly. In Point based Algorithms BIB002 , a set of belief points are first sampled from the belief simplex (by letting the agent interact with the environment) and then planning is performed on these points only. In , a randomized version of point-based value iteration was presented. Here, the key idea is that, in each value iteration step, it can improve the value of all points in the belief set by only updating the value and its gradient of a subset of the points, and we select these belief points in a randomized greedy manner. In , similar to BIB003 , it selects some representative points in the belief space and iteratively applies value update to those points.

Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Some Variations and Improvements of POMDPs Solution Methods <s> This is a study of finite state discrete time discounted Markovian decision process when the states are probabilistically observed. A model of this process is formulated, and an implicit enumeration algorithm is presented which optimizes the total expected discounted reward given the initial state. Several numerical examples are presented. <s> BIB001 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Some Variations and Improvements of POMDPs Solution Methods <s> Stochastic scheduling problems are difficult stochastic control problems with combinatorial decision spaces. We focus on a class of stochastic scheduling problems, the quiz problem and its variations. We discuss the use of heuristics for their solution, and we propose rollout algorithms based on these heuristics which approximate the stochastic dynamic programming algorithm. We show how the rollout algorithms can be implemented efficiently, with considerable savings in computation over optimal algorithms. We delineate circumstances under which the rollout algorithms are guaranteed to perform better than the heuristics on which they are based. We also show computational results which suggest that the performance of the rollout policies is near-optimal, and is substantially better than the performance of their underlying heuristics. <s> BIB002 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Some Variations and Improvements of POMDPs Solution Methods <s> We propose a new approach to the problem of searching a space of stochastic controllers for a Markov decision process (MDP) or a partially observable Markov decision process (POMDP). Following several other authors, our approach is based on searching in parameterized families of policies (for example, via gradient descent) to optimize solution quality. However, rather than trying to estimate the values and derivatives of a policy directly, we do so indirectly using estimates for the probability densities that the policy induces on states at the different points in time. This enables our algorithms to exploit the many techniques for efficient and robust approximate density propagation in stochastic systems. We show how our techniques can be applied both to deterministic propagation schemes (where the MDP's dynamics are given explicitly in compact form,) and to stochastic propagation schemes (where we have access only to a generative model, or simulator, of the MDP). We present empirical results for both of these variants on complex problems. <s> BIB003 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Some Variations and Improvements of POMDPs Solution Methods <s> We examine the problem of generating state-space compressions of POMDPs in a way that minimally impacts decision quality. We analyze the impact of compressions on decision quality, observing that compressions that allow accurate policy evaluation (prediction of expected future reward) will not affect decision quality We derive a set of sufficient conditions that ensure accurate prediction in this respect, illustrate interesting mathematical properties these confer on lossless linear compressions, and use these to derive an iterative procedure for finding good linear lossy compressions. We also elaborate on how structured representations of a POMDP can be used to find such compressions. <s> BIB004 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Some Variations and Improvements of POMDPs Solution Methods <s> Standard value function approaches to finding policies for Partially Observable Markov Decision Processes (POMDPs) are intractable for large models. The intractability of these algorithms is due to a great extent to their generating an optimal policy over the entire belief space. However, in real POMDP problems most belief states are unlikely, and there is a structured, low-dimensional manifold of plausible beliefs embedded in the high-dimensional belief space. ::: ::: We introduce a new method for solving large-scale POMDPs by taking advantage of belief space sparsity. We reduce the dimensionality of the belief space by exponential family Principal Components Analysis [1], which allows us to turn the sparse, high-dimensional belief space into a compact, low-dimensional representation in terms of learned features of the belief state. We then plan directly on the low-dimensional belief features. By planning in a low-dimensional space, we can find policies for POMDPs that are orders of magnitude larger than can be handled by conventional techniques. We demonstrate the use of this algorithm on a synthetic problem and also on a mobile robot navigation task. <s> BIB005 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Some Variations and Improvements of POMDPs Solution Methods <s> Since the early 1990's, Markov decision processes (MDPs) and their partially observable counterparts (POMDPs) have been widely used by the AI community for planning under uncertainty. POMDPs offer a rich language to describe situations involving uncertainty about the domain, stochastic actions, noisy observations, and a variety of possible objective functions. Even though an optimal solution may be concise, current exact algorithms that use dynamic programming often require an intractable amount of space. POMDP approximation algorithms can operate with a limited amount of memory, but as a consequence they provide very weak theoretical guarantees. In contrast, we describe a new approach that addresses the space requirement of POMDP algorithms while maintaining well-defined optimality guarantees. <s> BIB006

Most of the existing algorithms are based on traditional value function iterations algorithms that try to minimize each point in the belief space search. Researchers, in recent years in the quest for optimal solution continued to explore various approaches to prune down the belief space in order to reduce the computational cost. In this section, we present several of those enhancements and modifications tailored towards specific applications. In , the authors, presented a Monte Carlo based approximation methods to POMDP. Their method does not require analytical tractability; the system is modular and can be treated as plug and play. It is inherently non-myopic allowing tradeoff between short terms for long term reward. This approximation was adopted for adaptive sensing. In , Region-based Dynamic programming reduced the linear programs in cross sum pruning, and hence reducing the computational complexity. This was achieved by dividing the belief space into smaller regions and perform independent pruning in each region. In Parallel roll out , It is a novel approach proposed to generalize the roll out algorithm BIB002 , by rolling out a set of policies rather than single policy. This aimed at the class of problems and forms a system path through which multiple heuristics policies are automatically combined to a new system policy to enhance performance. POMDPs algorithms operate with limited amount of memory which led to weak theoretical guaranty. In BIB006 , they provided approach that addresses the space requirements and guaranty that optimal results can be maintained. Policy search algorithms BIB003 [40] BIB001 [42] have also been devised to concentrate on the computation of belief states, these have succeeded to some large extent but the problem with policy search algorithm is that it could be data inefficient; many policy search algorithms have difficulty using data from its previous computed policies. But for policy compression BIB004 [44] BIB005 focuses on relevant belief space but also allows the use of relevant training data to estimate the effect of any policy.

Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> POMDP and Wireless Body Area Network <s> This is a study of finite state discrete time discounted Markovian decision process when the states are probabilistically observed. A model of this process is formulated, and an implicit enumeration algorithm is presented which optimizes the total expected discounted reward given the initial state. Several numerical examples are presented. <s> BIB001 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> POMDP and Wireless Body Area Network <s> The potential for collaborative, robust networks of microsensors has attracted a great deal of research attention. For the most part, this is due to the compelling applications that will be enabled once wireless microsensor networks are in place; location-sensing, environmental sensing, medical monitoring and similar applications are all gaining interest. However, wireless microsensor networks pose numerous design challenges. For applications requiring long-term, robust sensing, such as military reconnaissance, one important challenge is to design sensor networks that have long system lifetimes. This challenge is especially difficult due to the energy-constrained nature of the devices. In order to design networks that have extremely long lifetimes, we propose a physical layer driven approach to designing protocols and algorithms. We first present a hardware model for our wireless sensor node and then introduce the design of physical layer aware protocols, algorithms, and applications that minimize energy consumption of the system. Our approach prescribes methods that can be used at all levels of the hierarchy to take advantage of the underlying hardware. We also show how to reduce energy consumption of non-ideal hardware through physical layer aware algorithms and protocols. <s> BIB002 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> POMDP and Wireless Body Area Network <s> In wireless sensor networks that consist of a large number of low-power, short-lived, unreliable sensors, one of the main design challenges is to obtain long system lifetime, as well as maintain sufficient sensing coverage and reliability. In this paper, we propose a node-scheduling scheme, which can reduce system overall energy consumption, therefore increasing system lifetime, by turning off some redundant nodes. Our coverage-based off-duty eligibility rule and backoff-based node-scheduling scheme guarantees that the original sensing coverage is maintained after turning off redundant nodes. We implement our proposed scheme in NS-2 as an extension of the LEACH protocol. We compare the energy consumption of LEACH with and without the extension and analyze the effectiveness of our scheme in terms of energy saving. Simulation results show that our scheme can preserve the system coverage to the maximum extent. In addition, after the node-scheduling scheme turns off some nodes, certain redundancy is still guaranteed, which we believe can provide enough sensing reliability in many applications. <s> BIB003 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> POMDP and Wireless Body Area Network <s> A critical aspect of applications with wireless sensor networks is network lifetime. Battery-powered sensors are usable as long as they can communicate captured data to a processing node. Sensing and communications consume energy, therefore judicious power management and scheduling can effectively extend operational time. To monitor a set of targets with known locations when ground access in the monitored area is prohibited, one solution is to deploy the sensors remotely, from an aircraft. The loss of precise sensor placement would then be compensated by a large sensor population density in the drop zone, that would improve the probability of target coverage. The data collected from the sensors is sent to a central node for processing. In this paper we propose an efficient method to extend the sensor network operational time by organizing the sensors into a maximal number of disjoint set covers that are activated successively. Only the sensors from the current active set are responsible for monitoring all targets and for transmitting the collected data, while nodes from all other sets are in a low-energy sleep mode. In this paper we address the maximum disjoint set covers problem and we design a heuristic that computes the sets. Theoretical analysis and performance evaluation results are presented to verify our approach. <s> BIB004 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> POMDP and Wireless Body Area Network <s> Nowadays wireless sensor network have been used in many area for automatic localization and target tracking. However, the large resource demands of sensors for simultaneous running might not be feasible and will lead to loss of functions. In this paper, we consider the problem of dynamical sensor selection based on user-defined objectives, such as maximizing k-coverage and detection probability. We model the problem as a POMDP, which selects k sensors in each time frame to maintain the area coverage quality. In order to fulfill the users chosen objective, we change the system behavior by changing the reward function of POMDP and the parameter of coverage model. The performance of our scheme is compared to three different schemes: RandomSelection, ShortestDistance and POMDP with Boolean model. The POMDP scheme with probability coverage model shows the best performance in terms of coverage and detection probability. <s> BIB005

Generally, sensors deployments require high cost of maintenance, energy conservation; hence, there is a need to optimize power to increase the total number of measurement over a given period of time. In most of these applications, the states are not fully observable, sometimes even are not observable at all. There are many factors affect the performance of sensors such as energy, efficiency, power etc. These factors by nature constitute random processes. Depending on application or user defined performance measure, each state may have different definition. For example, in measuring coverage, one can look at a sensor state and whether it is active or inactive, good or bad, on or off, etc. POMDP provides a rich mathematical structure for planning these types of problems under uncertainty. The cardinal issue in sensor scheduling is how to attain optimal tradeoff between the cost of sensor usage and the sensor performance and the goal of the system is to maximize user defined goals BIB005 . In order to minimize energy consumption, sensor networks require a scheduling scheme to reduce the number of active sensors at any given time while monitoring the quality of network BIB005 . In general, this is an optimization problem . In BIB004 , the sensors are partitioned into disjoint set, with each set expected to cover the target. They provide a heuristic to solve the mixed linear programming problem that arises from their model. But the challenges of these techniques were that it cannot be applied dynamically. Similarly, in BIB002 , Voronoi diagrams were used to determine the hull points to place the sensors in order to maximize coverage and enhance battery life time. But in BIB001 , self-scheduling of duty eligible rules was applied, where a node can turn itself on or off by observing it nearest neighbor. It turns itself off, if its nearest neighbor is active and provides an optimal coverage. In BIB003 , the authors defined a threshold called k-cover (predefined constant) where a set of sensors less than k is determined through neighbor communication. In this approach, a central controller is assigned the duty of monitoring the coverage.

Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Fig. 4. Component of a biosensor [22]. <s> Implanted biological sensors are a special class of wireless sensor networks that are used in-vivo for various medical applications. One of the major challenges of continuous in-vivo sensing is the heat generated by the implanted sensors due to communication radiation and circuitry power consumption. This paper addresses the issues of routing in implanted sensor networks. We propose a thermal-aware routing protocol that routes the data away from high temperature areas (hot spots). With this protocol each node estimates temperature change of its neighbors and routes packets around the hot spot area by a withdraw strategy. The proposed protocol can achieve a better balance of temperature rise and only experience a modest increased delay compared with shortest hop, but thermal-awareness also indicates the capability of load balance, which leads to less packet loss in high load situations. <s> BIB001 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Fig. 4. Component of a biosensor [22]. <s> A network of biosensors can be implanted in a human body for health monitoring, diagnostics, or as a prosthetic device. Biosensors can be organized into clusters where most of the communication takes place within the clusters, and long range transmissions to the base station are performed by the cluster leader to reduce the energy cost. In some applications, the tissues are sensitive to temperature increase and may be damaged by the heat resulting from normal operations and the recharging of sensor nodes. Our work is the first to consider rotating the cluster leadership to minimize the heating effects on human tissues. We explore the factors that lead to temperature increase, and the process for calculating the specific absorption rate (SAR) and temperature increase of implanted biosensors by using the finite-difference time-domain (FDTD) method. We improve performance by rotating the cluster leader based on the leadership history and the sensor locations. We propose a simplified scheme, temperature increase potential, to efficiently predict the temperature increase in tissues surrounding implanted sensors. Finally, a genetic algorithm is proposed to exploit the search for an optimal temperature increase sequence. <s> BIB002 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Fig. 4. Component of a biosensor [22]. <s> Biosensors are tiny wireless medical devices which are attached or implanted into the body of a human being or animal to monitor and control biological processes. They are distinguished from conventional sensors by their biologically derived sensing elements. Biosensors generate heat when they transmit their measurements and their temperature rises when recharged by electromagnetic energy. These phenomena translate to a temperature increase in the tissues surrounding the biosensors. If the temperature increase exceeds a certain threshold, the tissues might be damaged. In this paper, we discuss the problem of finding an optimal operating policy for a rechargable biosensor under a strict maximum temperature increase constraint. This problem can be formulated as a Markov decision process with an average reward criterion. The solution is an optimal policy that maximizes the average number of samples which can be generated by the biosensor while observing the constraint on the maximum safe temperature level. Due to the exponential nature of the problem, a heuristic policy is proposed. The performance of the policies is studied through simulation. A greedy policy is used as a baseline for comparison. <s> BIB003

In BIB001 , the authors exploit the notion of leadership clusters to calculate the optimal temperature in an implanted biosensor network. They used the Penne's heat equation and finite domain time difference method to calculate the temperature increase and specific absorption rate. The scheme is referred to as Temperature Increase Potential (TIP). Genetic algorithm and TIP are combined for the computation of temperature increase. However, they did not incorporate the effect of the wireless channel and the energy in the model. The management problem of rechargeable biosensor in a temperature sensitive environment like human body was considered in BIB002 . Optimal number of sample was considered with the constraint of effect of temperature on tissue and the heat generated while recharging the biosensor. The problem was formulated as Markov Decision Problem (MDP) with numerical and simulation result presented. Though, an interesting result was obtained, however, the formulation of the problem with MDP, with assumption that all states are deterministic does not address the real life situation of noisy sensors and channels. In BIB003 , the authors modeled the dynamic scheduling of biosensors to manage the heat generated. The scheduling problem was modeled as MDP. Though, heat effect was considered in this work, but modeling the scheduling problem as MDP does not incorporate the noisy behavior of the sensor and the channels.

Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Challenges and Open Issues <s> Wireless body area networks (WBANs) offer many promising new applications in the area of remote health monitoring. An important element in the development of a WBAN is the characterization of the physical layer of the network, including an estimation of the delay spread and the path loss between two nodes on the body. This paper discusses the propagation channel between two half-wavelength dipoles at 2.45 GHz, placed near a human body and presents an application for cross-layer design in order to optimize the energy consumption of different topologies. Propagation measurements are performed on real humans in a multipath environment, considering different parts of the body separately. In addition, path loss has been numerically investigated with an anatomically correct model of the human body in free space using a 3-D electromagnetic solver. Path loss parameters and time-domain channel characteristics are extracted from the measurement and simulation data. A semi-empirical path loss model is presented for an antenna height above the body of 5 mm and antenna separations from 5 cm up to 40 cm. A time-domain analysis is performed and models are presented for the mean excess delay and the delay spread. As a cross-layer application, the proposed path loss models are used to evaluate the energy efficiency of single-hop and multihop network topologies. <s> BIB001 </s> Partially Observable Markov Decision Processes (POMDPs) and Wireless Body Area Networks (WBAN): A Survey. <s> Challenges and Open Issues <s> Wireless body area sensing networks have the potential to revolutionize health care in the near term. The coupling of biosensors with a wireless infrastructure enables the real-time monitoring of an individual's health and related behaviors continuously, as well as the provision of realtime feedback with nimble, adaptive, and personalized interventions. The KNOWME platform is reviewed, and lessons learned from system integration, optimization, and in-field deployment are provided. KNOWME is an endto- end body area sensing system that integrates off-the-shelf sensors with a Nokia N95 mobile phone to continuously monitor and analyze the biometric signals of a subject. KNOWME development by an interdisciplinary team and in-laboratory, as well as in-field deployment studies, employing pediatric obesity as a case study condition to monitor and evaluate physical activity, have revealed four major challenges: (1) achieving robustness to highly varying operating environments due to subject-induced variability such as mobility or sensor placement, (2) balancing the tension between acquiring high fidelity data and minimizing network energy consumption, (3) enabling accurate physical activity detection using a modest number of sensors, and (4) designing WBANs to determine physiological quantities of interest such as energy expenditure. The KNOWME platform described in this article directly addresses these challenges. <s> BIB002

In this section, we discuss some of the major challenges facing the application of POMDP in real life WBAN applications. Fully observable MDP have been applied in so many areas as we earlier mentioned in , but not so for POMDP due to numerous challenges, most of the research ended up as published papers in reference journals. Though most real life problems are not fully observable which made POMDP a good candidate but because of some of these challenges, they have not been largely harnessed. The complexity of interaction between the state and information makes the modeling problem a difficult task. POMDP has difficulty in handling some problems with inherited peculiar characteristics. While most of the models assume finite decision point approach, most of real life problems are not. POMDP applications are data intensive and new approaches need to be developed to minimize the collected information as well as retrieving this information. Data mining techniques would help in this domain. In addition, artificial intelligent techniques such as reinforcement learning would help in striking a balance between the accuracy required by the application and the quality of collected information. MDP has been used a lot for solving and approximating the solution of POMDP. However, it ends with simplistic cases which are far away from the real life problems. Hence, new directions should be explored to enhance the formulation models and representation. On the other hand, the limited number of problems with known solutions along with lack of efficient algorithms to solve POMDP problem make the practitioners less enthusiastic in adopting these approaches which also hinder the development is this area. The future WBANs should be dependable where a conjunction of reliability, security and availability is provided BIB001 . Furthermore, building WBANs according to hierarchical configuration BIB002 where sensors can be partitioned into different layers, with each layer providing different granularities can help in achieving better coordination and determination of the state of the monitored system.

Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Introduction <s> Preface Introduction 1. The expression of subjectivity and the sentences of direct and indirect speech 2. The sentence of represented speech and thought 3. Communication and the sentence of discourse 4. The sentences of narration and discourse 5. The sentence representing non-reflective consciousness and the absence of the narrator 6. The historical development of narrative style Conclusion: Narration and representation: the knowledge of the clock and the lens Notes Bibliography Name index Subject index <s> BIB001 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Introduction <s> This paper introduces a methodology for determining polarity of text within a multilingual framework. The method leverages on lexical resources for sentiment analysis available in English (SentiWordNet). First, a document in a different language than English is translated into English using standard translation software. Then, the translated document is classified according to its sentiment into one of the classes "positive" and "negative". For sentiment classification, a document is searched for sentiment bearing words like adjectives. By means of SentiWordNet, scores for positivity and negativity are determined for these words. An interpretation of the scores then leads to the document polarity. The method is tested for German movie reviews selected from Amazon and is compared to a statistical polarity classifier based on n-grams. The results show that working with standard technology and existing sentiment analysis approaches is a viable approach to sentiment analysis within a multilingual framework. <s> BIB002 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Introduction <s> The Internet is frequently used as a medium for exchange of information and opinions, as well as propaganda dissemination. In this study the use of sentiment analysis methodologies is proposed for classification of Web forum opinions in multiple languages. The utility of stylistic and syntactic features is evaluated for sentiment classification of English and Arabic content. Specific feature extraction components are integrated to account for the linguistic characteristics of Arabic. The entropy weighted genetic algorithm (EWGA) is also developed, which is a hybridized genetic algorithm that incorporates the information-gain heuristic for feature selection. EWGA is designed to improve performance and get a better assessment of key features. The proposed features and techniques are evaluated on a benchmark movie review dataset and U.S. and Middle Eastern Web forum postings. The experimental results using EWGA with SVM indicate high performance levels, with accuracies of over 91p on the benchmark dataset as well as the U.S. and Middle Eastern forums. Stylistic features significantly enhanced performance across all testbeds while EWGA also outperformed other feature selection methods, indicating the utility of these features and techniques for document-level classification of sentiments. <s> BIB003 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Introduction <s> We explore the adaptation of English resources and techniques for text sentiment analysis to a new language, Spanish. Our main focus is the modification of an existing English semantic orientation calculator and the building of dictionaries; however we also compare alternate approaches, including machine translation and Support Vector Machine classification. The results indicate that, although languageindependent methods provide a decent baseline performance, there is also a significant cost to automation, and thus the best path to long-term improvement is through the inclusion of language-specific knowledge and resources. <s> BIB004 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Introduction <s> This paper proposes a method that automatically creates a sentiment lexicon in a new language using a sentiment lexicon in a resource–rich language with only a bilingual dictionary. We resolve some of the difficulties in selecting appropriate senses when translating lexicon, and present a framework that sequentially applies an iterative link analysis algorithm to enhance the quality of lexicons of both the source and target languages. The experimental results have empirically shown to improve the sentiment lexicon in the source language as well as create a good quality lexicon in the new language. <s> BIB005 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Introduction <s> Recently, Opinion Mining (OM) is receiving more attention due to the abundance of forums, blogs, ecommerce web sites, news reports and additional web sources where people tend to express their opinions. There are a number of works about Sentiment Analysis (SA) studying the task of identifying the polarity, whether the opinion expressed in a text is positive or negative about a given topic. However, most of research is focused on English texts and there are very few resources for other languages. In this work we present an Opinion Corpus for Arabic (OCA) composed of Arabic reviews extracted from specialized web pages related to movies and films using this language. Moreover, we have translated the OCA corpus into English, generating the EVOCA corpus (English Version of OCA). In the experiments carried out in this work we have used different machine learning algorithms to classify the polarity in these corpora showing that, although the experiments with EVOCA are worse than OCA, the results are comparable with other English experiments, since the loss of precision due to the translation is very slight. <s> BIB006 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Introduction <s> Sentiment analysis is a challenging new task related to text mining and natural language processing. Although there are, at present, several studies related to this theme, most of these focus mainly on English texts. The resources available for opinion mining (OM) in other languages are still limited. In this article, we present a new Arabic corpus for the OM task that has been made available to the scientific community for research purposes. The corpus contains 500 movie reviews collected from different web pages and blogs in Arabic, 250 of them considered as positive reviews, and the other 250 as negative opinions. Furthermore, different experiments have been carried out on this corpus, using machine learning algorithms such as support vector machines and Nave Bayes. The results obtained are very promising and we are encouraged to continue this line of research. © 2011 Wiley Periodicals, Inc. <s> BIB007 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Introduction <s> Opinion mining aims at extracting useful subjective information from reliable amounts of text. Opinion mining holder recognition is a task that has not been considered yet in Arabic Language. This task essentially requires deep understanding of clauses structures. Unfortunately, the lack of a robust, publicly available, Arabic parser further complicates the research. This paper presents a leading research for the opinion holder extraction in Arabic news independent from any lexical parsers. We investigate constructing a comprehensive feature set to compensate the lack of parsing structural outcomes. The proposed feature set is tuned from English previous works coupled with our proposed semantic field and named entities features. Our feature analysis is based on Conditional Random Fields (CRF) and semi-supervised pattern recognition techniques. Different research models are evaluated via cross-validation experiments achieving 54.03 F-measure. We publicly release our own research outcome corpus and lexicon for opinion mining community to encourage further research. <s> BIB008

The web has become a read-write platform where users are no longer strictly consumers of information but also producers. User-generated content, in the form of unstructured free text, is becoming an integral part of the web mainly because of the dramatic increase of social network sites, video sharing sites, online news, online reviews sites, online forums and blogs. Because of this proliferation of user-generated content, Web content mining is gaining considerable attention due to its importance for many businesses, governmental agencies and institutions. Subjectivity and sentiment analysis (SSA) is an important sub-area of Web content mining. In natural language, subjectivity refers to aspects of language used to express opinions, feelings, evaluations, and speculations BIB001 , including sentiment. The process of subjectivity classification refers to the task of classifying texts as either objective (e.g., The new iPad was released) or subjective (e.g., The new iPad is cool). Subjective text can further be classified by its sentiment and polarity. For sentiment classification, the task consists of identifying whether a subjective text is positive (e.g., Egyptians inspired the world with their revolution!), negative (e.g., The bloodbaths in Syria are horrifying!), neutral (e.g., Obama may sign the bill), or mixed (e.g., The iPad is cool, but way too expensive). These various types of SSA are gaining increasing attention because they provide an automated way to summarize vast quanitities of text (including reviewers, blogs, Twitter feeds, etc.) into the opinions that they express. This data is of use to businesses and institutions who want to monitor feelings about their products and services. Private citizens can also benefit from this information to be able to compare sentiments about competing products, for example. Following a considerable body of related literature, we will henceforth use the terms subjectivity and sentiment analysis and sentiment analysis interchangeably. Sentiment analysis can thus be viewed as a classification process that aims to determine whether a certain document or text was written to express a positive or a negative opinion about a certain object (e.g., a topic, product, or person). This process regards each document as a basic information unit. The process has been referred to as 'document level-sentiment classification' where the document is seen as an opinionated artifact. The more fine-grained problem of identifying the sentiment of every sentence has also been studied . Sentiment analysis is typically performed using one of two basic approaches: rule-based classifiers, in which rules derived from linguistic study of a language are applied to sentiment analysis BIB004 BIB002 BIB008 BIB005 , and machine learning classifiers in which statistical machine learning algorithms are used to learn signals of sentiment automatically BIB003 BIB006 BIB007 . Currently, most of the systems built for sentiment analysis are tailored for the English language , but there has been some work on other languages. This paper reviews efforts to build SSA systems for Arabic, a Semitic language. After a brief discussion of the properties of Arabic in Section 2, we review resources and test corpora for Arabic in Section 3. Section 4 then reviews various approaches that have been employed for Arabic SSA.

Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Arabic <s> This general introduction to the Arabic Language, now available in paperback, places special emphasis on the history and variation of the language. Concentrating on the difference between the two types of Arabic - the Classical standard language and the dialects - Kees Versteegh charts the history and development of the Arabic language from the earliest beginnings to modern times. The reader is offered a solid grounding in the structure of the language, its historical context and its use in various literary and non-literary genres, as well as an understanding of the role of Arabic as a cultural, religious and political world language. Intended as an introductory guide for students of Arabic, it will also be a useful tool for discussions both from a historical linguistic and from a socio-linguistic perspective. Coverage includes all aspects of the history of Arabic, the Arabic linguistic tradition, Arabic dialects and Arabic as a world language. Links are made between linguistic history and cultural history, while the author emphasises the role of contacts between Arabic and other languages. This important book will be an ideal text for all those wishing to acquire an understanding or develop their knowledge of the Arabic language. Key Features: * A general introduction to the Arabic language * Accessible and effective communication of information * Impeccably documented * Updated guide to further reading * User-friendly index <s> BIB001 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Arabic <s> In this work, we investigate sentiment mining of Arabic text at both the sentence level and the document level. Existing research in Arabic sentiment mining remains very limited. For sentence-level classification, we investigate two approaches. The first is a novel grammatical approach that employs the use of a general structure for the Arabic sentence. The second approach is based on the semantic orientation of words and their corresponding frequencies, to do this we built an interactive learning semantic dictionary which stores the polarities of the roots of different words and identifies new polarities based on these roots. For document-level classification, we use sentences of known classes to classify whole documents, using a novel approach whereby documents are divided dynamically into chunks and classification is based on the semantic contributions of different chunks in the document. This dynamic chunking approach can also be investigated for sentiment mining in other languages. Finally, we propose a hierarchical classification scheme that uses the results of the sentence-level classifier as input to the document-level classifier, an approach which has not been investigated previously for Arabic documents. We also pinpoint the various challenges that are faced by sentiment mining for Arabic texts and propose suggestions for its development. We demonstrate promising results with our sentence-level approach, and our document-level experiments show, with high accuracy, that it is feasible to extract the sentiment of an Arabic document based on the classes of its sentences. <s> BIB002

As the official language of 22 countries, Arabic is spoken by more than 300 million people, and is the fastest-growing language on the web (with an annual growth rate of 2,501.2% in the number of Internet users as of 2010, compared to 1,825.8% for Russian, 1,478.7% for Chinese and 301.4% for English) [1] . There are about 65 million Arabic-speaking users online, or about 18.8% of the global Internet population [1] . Arabic is a Semitic language BIB001 and consists of many different regional dialects. While these dialects are true native language forms, they are typically used only in informal daily communication and are not standarized or taught in schools . There is one formal written standard that is commonly used in written media and education throughout the Arab world called Modern Standard Arabic (MSA). There is a large degree of difference between MSA and most Arabic dialects, and, interestingly, MSA is not actually the native language of any Arabic country or group. MSA is syntactically, morphologically, and phonologically based on Classical Arabic (CA) , which is the language of the Qur'an (Islam's Holy Book). Arabic has a very rich inflectional system and is considered one of the richest languages in terms of morphology . Arabic sentential forms are divided into two types, nominal and verbal constructions BIB002 . In the verbal domain, Arabic has two word order patterns (i.e., Subject-VerbObject and Verb-Subject-Object). In the nominal domain, a normal pattern would consist of two consecutive words, a noun (i.e., subject) then an adjective (subject descriptor).

Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Corpora and lexicons <s> Opinion mining (OM) is a recent subdiscipline at the crossroads of information retrieval and computational linguistics which is concerned not with the topic a document is about, but with the opinion it expresses. OM has a rich set of applications, ranging from tracking users’ opinions about products or about political candidates as expressed in online forums, to customer relationship management. In order to aid the extraction of opinions from text, recent research has tried to automatically determine the “PN-polarity” of subjective terms, i.e. identify whether a term that is a marker of opinionated content has a positive or a negative connotation. Research on determining whether a term is indeed a marker of opinionated content (a subjective term) or not (an objective term) has been, instead, much more scarce. In this work we describe SENTIWORDNET, a lexical resource in which each WORDNET synset s is associated to three numerical scoresObj(s), Pos(s) and Neg(s), describing how objective, positive, and negative the terms contained in the synset are. The method used to develop SENTIWORDNET is based on the quantitative analysis of the glosses associated to synsets, and on the use of the resulting vectorial term representations for semi-supervised synset classification. The three scores are derived by combining the results produced by a committee of eight ternary classifiers, all characterized by similar accuracy levels but different classification behaviour. SENTIWORDNET is freely available for research purposes, and is endowed with a Web-based graphical user interface. <s> BIB001 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Corpora and lexicons <s> Named Entity Recognition (NER) task has become essential to improve the performance of many NLP tasks. Its aim is to endeavor a solution to boost accurately the identification of extracted named entities. This paper presents a novel solution for Arabic Named Entity Recognition (ANER) problem. The solution is an integration approach between two machine learning techniques, namely bootstrapping semi-supervised pattern recognition and Conditional Random Fields (CRF) classifier as a supervised technique. The paper solution contributions are the exploit of pattern and word semantic fields as CRF features, the adventure of utilizing bootstrapping semisupervised pattern recognition technique in Arabic Language, and the integration success to improve the performance of its components. Moreover, as per to our knowledge, this proposed integration has not been utilized for NER task of other natural languages. Using 6-fold cross-validation experimental tests, the solution is proved that it outperforms previous CRF sole work <s> BIB002 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Corpora and lexicons <s> For languages with rich content over the web, business reviews are easily accessible via many known websites, e.g., Yelp.com. For languages with poor content over the web like Arabic, there are very few websites (we are actually aware of only one that is indeed unpopular) that provide business reviews. However, this does not mean that such reviews do not exist. They indeed exist unstructured in websites not originally intended for reviews, e.g., Forums and Blogs. Hence, there is a need to mine for those Arabic reviews from the web in order to provide them in the search results when a user searches for a business or a category of businesses. In this paper, we show how to extract the business reviews scattered on the web written in the Arabic language. The mined reviews are analyzed to also provide their sentiments (positive, negative or neutral). This way, we provide our users the information they need about the local businesses in the language they understand, and therefore provide a better search experience for the Middle East region, which mostly speaks Arabic. <s> BIB003 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Corpora and lexicons <s> Subjectivity and sentiment analysis (SSA) is an area that has been witnessing a flurry of novel research. However, only few attempts have been made to build SSA systems for morphologically-rich languages (MRL). In the current study, we report efforts to partially bridge this gap. We present a newly labeled corpus of Modern Standard Arabic (MSA) from the news domain manually annotated for subjectivity and domain at the sentence level. We summarize our linguistically-motivated annotation guidelines and provide examples from our corpus exemplifying the different phenomena. Throughout the paper, we discuss expression of subjectivity in natural language, combining various previously scattered insights belonging to many branches of linguistics. <s> BIB004 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Corpora and lexicons <s> The area of Subjectivity and sentiment analysis (SSA) has been witnessing a flurry of novel research. However, only few attempts have been made to build SSA systems for the health domain. In the current study, we report efforts to partially bridge this gap. We present a new labeled corpus of professional articles collected from major Websites focused on the Obama health reform plan (OHRP). We introduce a new annotation scheme that incorporates subjectivity as well as topics directly related to the OHRP and describe a highlysuccessful SSA system that exploits the annotation. In the process, we introduce a number of novel features and a wide-coverage polarity lexicon for the health domain. <s> BIB005 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Corpora and lexicons <s> Sentiment analysis is a challenging new task related to text mining and natural language processing. Although there are, at present, several studies related to this theme, most of these focus mainly on English texts. The resources available for opinion mining (OM) in other languages are still limited. In this article, we present a new Arabic corpus for the OM task that has been made available to the scientific community for research purposes. The corpus contains 500 movie reviews collected from different web pages and blogs in Arabic, 250 of them considered as positive reviews, and the other 250 as negative opinions. Furthermore, different experiments have been carried out on this corpus, using machine learning algorithms such as support vector machines and Nave Bayes. The results obtained are very promising and we are encouraged to continue this line of research. © 2011 Wiley Periodicals, Inc. <s> BIB006 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Corpora and lexicons <s> In this paper, we present a combined approach that automatically extracts opinions from Arabic documents. Most research efforts in the area of opinion mining deal with English texts and little work with Arabic text. Unlike English, from our experiments, we found that using only one method on Arabic opinioned documents produce a poor performance. So, we used a combined approach that consists of three methods. At the beginning, lexicon based method is used to classify as much documents as possible. The resultant classified documents used as training set for maximum entropy method which subsequently classifies some other documents. Finally, k-nearest method used the classified documents from lexicon based method and maximum entropy as training set and classifies the rest of the documents. Our experiments showed that in average, the accuracy moved (almost) from 50% when using only lexicon based method to 60% when used lexicon based method and maximum entropy together, to 80% when using the three combined methods. <s> BIB007 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Corpora and lexicons <s> We present AWATIF, a multi-genre corpus of Modern Standard Arabic (MSA) labeled for subjectivity and sentiment analysis (SSA) at the sentence level. The corpus is labeled using both regular as well as crowd sourcing methods under three different conditions with two types of annotation guidelines. We describe the sub-corpora constituting the corpus and provide examples from the various SSA categories. In the process, we present our linguistically-motivated and genre-nuanced annotation guidelines and provide evidence showing their impact on the labeling task. <s> BIB008 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Corpora and lexicons <s> Opinion mining aims at extracting useful subjective information from reliable amounts of text. Opinion mining holder recognition is a task that has not been considered yet in Arabic Language. This task essentially requires deep understanding of clauses structures. Unfortunately, the lack of a robust, publicly available, Arabic parser further complicates the research. This paper presents a leading research for the opinion holder extraction in Arabic news independent from any lexical parsers. We investigate constructing a comprehensive feature set to compensate the lack of parsing structural outcomes. The proposed feature set is tuned from English previous works coupled with our proposed semantic field and named entities features. Our feature analysis is based on Conditional Random Fields (CRF) and semi-supervised pattern recognition techniques. Different research models are evaluated via cross-validation experiments achieving 54.03 F-measure. We publicly release our own research outcome corpus and lexicon for opinion mining community to encourage further research. <s> BIB009

The availability of annotated corpora for training and testing is very important to enable progress on sentiment recognition systems. Collecting this data (and particularly the annotations) can be very labor-intensive. Fortunately, a number of research groups have developed and released Arabic sentiment analysis corpora, and we review this work here. AWATIF is a multi-genre, multi-dialect corpus for Arabic SSA built by Abdul-Mageed and Diab BIB008 BIB004 . AWATIF is extracted from three different resources: the Penn Arabic Treebank (PATB), which is an existing collection of news wire stories in different domains (e.g. sports, politics, finance, etc.), Wikipedia user talk pages covering a variety of topics, and conversation threads from web forums on seven different sites. In annotating the corpus they used two different procedures, one that used untrained annotators via crowd sourcing technologies to give a coarse sentiment label (positive, negative, or neutral) to each sentence, and then one that used annotators trained with some linguistic background to label each sentence. The authors also manually created an adjective polarity lexicon, giving coarse labels (again positive, negative or neutral) to each of nearly 4,000 Arabic adjectives. In a related work, Abdul-Mageed and Diab use a machine translation procedure to translate available English lexicons (e.g., from BIB001 and BIB005 ) into Arabic. They retrieve 229,452 entries, including expressions commonly used in social media. Opinion corpus for Arabic (OCA) is a corpus of text from movie review sites by RushdiSaleh et al. BIB006 , and includes a parallel English version called EVOCA. The corpus consists of 500 reviews, half negative and half positive. The raw reviews contained a number of challenges which the authors attempted to fix manually, including filtering out spurious and unrelated comments, romanization of Arabic, multi-language reviews, differing spellings of proper names, and movie reviews that were more opinions of the cultural and political themes of a movie than the film itself. (As an example of the latter issue, the movie "Antichrist" has a rating of 6.7 in IMDB but a rating of 1 in the reviews on the Arabic blog.) OCA and EVOCA performed standard preprocessing on the corpus, including correcting spelling mistakes and deleting special characters, and also have made available unigram, bigram, and trigrams for the dataset. MPQA subjective lexicon & Arabic opinion holder corpus: Another corpus for Arabic opinion holder and subjectivity lexicon is proposed by Elaranoty et al. BIB009 , who created an Arabic news corpus. They crawled 150 MB of Arabic news and manually annotated 1 MB of the corpus for opinion holder. The opinion holder corpus was annotated by three different people. Any conflict emerging because of different annotations was solved using majority voting. For preprocessing the corpus, the Research and Development International (RDI) tool (http://www.rdieg.com) was used to handle the morphological analysis of Arabic sentences and assign parts of speech (POS) tags. Finally, semantic analysis of the words were done. Arabic Named Entity Recognition (ANER) BIB002 was used for extracting names from documents. The proposed Arabic subjectivity lexicon contains strong as well as weak subjective clues by manually translating the MPQA lexicon . An Arabic Lexicon for Business Reviews was proposed by Elhawary and Elfeky BIB003 . Elhawary and Elfeky BIB003 used the similarity graph to build an Arabic lexicon. The similarity graph is a type of graph where the two words or phrases have an edge if they are similar on polarity or meaning. The weight of the edge represents the degree of similarity between two nodes. Usually, this graph is built in an unsupervised manner based on lexical co-occurrences from large Web corpora. Here, the researchers initially used a small set of seeds and then performed label propagation on an Arabic similarity graph. For building the Arabic similarity graph, a list of seeds for positive, negative and neutral are used. The Arabic lexicon created from the similarity graph consists of two columns where the first column is the word or phrase and the second column represent the score of the word which is the sum of the scores of all edges connected to this node (word/phrase). They applied filtering rules to avoid both the sparseness of the data and garbage nodes. Garbage nodes caused the top 200 positive words to be non-positive. They removed nodes with a high number of weighted edges and kept the 25 top ranked synonyms of the word. The top 25 synonyms of positive words are 90% positive. This ratio became 50-60% when considering all synonyms. The sentiment of the review is computed based on the sentiment of the sentences. That is, sentence boundary detection is used, and negation is also used, to flip the sentiment score from positive to negative and vice versa. There are around 20 Arabic words for negation. Sentences greater than 120 character are ignored. The results show that the created Arabic lexicon has high precision but has low recall. Another subjectivity lexicon is proposed by El-Halees BIB007 . This lexicon is built manually based on two resources, the SentiStrength project and an online dictionary. They translated the English list from SentiStrength project and then manually filtered it. Common Arabic words were added to the lexicon.

Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Language-Independent Feature Selection and Extraction <s> We consider the problem of classifying documents not by topic, but by overall sentiment, e.g., determining whether a review is positive or negative. Using movie reviews as data, we flnd that standard machine learning techniques deflnitively outperform human-produced baselines. However, the three machine learning methods we employed (Naive Bayes, maximum entropy classiflcation, and support vector machines) do not perform as well on sentiment classiflcation as on traditional topic-based categorization. We conclude by examining factors that make the sentiment classiflcation problem more challenging. <s> BIB001 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Language-Independent Feature Selection and Extraction <s> Sentiment analysis seeks to identify the viewpoint(s) underlying a text span; an example application is classifying a movie review as "thumbs up" or "thumbs down". To determine this sentiment polarity, we propose a novel machine-learning method that applies text-categorization techniques to just the subjective portions of the document. Extracting these portions can be implemented using efficient techniques for finding minimum cuts in graphs; this greatly facilitates incorporation of cross-sentence contextual constraints. <s> BIB002 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Language-Independent Feature Selection and Extraction <s> Little work to date in sentiment analysis (classifying texts by `positive' or `negative' orientation) has attempted to use fine-grained semantic distinctions in features used for classification. We present a new method for sentiment classification based on extracting and analyzing appraisal groups such as ``very good'' or ``not terribly funny''. An appraisal group is represented as a set of attribute values in several task-independent semantic taxonomies, based on Appraisal Theory. Semi-automated methods were used to build a lexicon of appraising adjectives and their modifiers. We classify movie reviews using features based upon these taxonomies combined with standard ``bag-of-words'' features, and report state-of-the-art accuracy of 90.2%. In addition, we find that some types of appraisal appear to be more significant for sentiment classification than others. <s> BIB003 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Language-Independent Feature Selection and Extraction <s> Department of Computing, University of Surrey Guildford, Surrey, GU2 7XH, United Kingdom E-mail: [email protected] Encouraged by the feasibility demonstration that a relatively low-cost grid environment can speed up the processing of continuous text streams of financial news in English, we have attempted to replicate our methods for automatic sentiment analysis in two major languages of the world – Arabic and Chinese. We show that our local grammar approach, developed on an archive of English (Indo-European language) texts, works equally on the typologically different Chinese (Sino-Asiatic) and Arabic (Semitic) languages. Grid Technology for Financial Modeling and Simulation February 3/4, 2006 – Palermo, (Italy) <s> BIB004 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Language-Independent Feature Selection and Extraction <s> The Internet is frequently used as a medium for exchange of information and opinions, as well as propaganda dissemination. In this study the use of sentiment analysis methodologies is proposed for classification of Web forum opinions in multiple languages. The utility of stylistic and syntactic features is evaluated for sentiment classification of English and Arabic content. Specific feature extraction components are integrated to account for the linguistic characteristics of Arabic. The entropy weighted genetic algorithm (EWGA) is also developed, which is a hybridized genetic algorithm that incorporates the information-gain heuristic for feature selection. EWGA is designed to improve performance and get a better assessment of key features. The proposed features and techniques are evaluated on a benchmark movie review dataset and U.S. and Middle Eastern Web forum postings. The experimental results using EWGA with SVM indicate high performance levels, with accuracies of over 91p on the benchmark dataset as well as the U.S. and Middle Eastern forums. Stylistic features significantly enhanced performance across all testbeds while EWGA also outperformed other feature selection methods, indicating the utility of these features and techniques for document-level classification of sentiments. <s> BIB005 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Language-Independent Feature Selection and Extraction <s> An open problem in machine learning-based sentiment classification is how to extract complex features that outperform simple features; figuring out which types of features are most valuable is another Most of the studies focus primarily on character or word Ngrams features, but substring-group features have never been considered in sentiment classification area before In this study, the substring-group features are extracted and selected for sentiment classification by means of transductive learning-based algorithm To demonstrate generality, experiments have been conducted on three open datasets in three different languages: Chinese, English and Spanish The experimental results show that the proposed algorithm's performance is usually superior to the best performance in related work, and the proposed feature subsumption algorithm for sentiment classification is multilingual Compared to the inductive learning-based algorithm, the experimental results also illustrate that the transductive learning-based algorithm can significantly improve the performance of sentiment classification As for term weighting, the experiments show that the “tfidf-c” outperforms all other term weighting approaches in the proposed algorithm. <s> BIB006 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Language-Independent Feature Selection and Extraction <s> The paper presents results of an experiment dealing with sentiment analysis of Croatian text from the domain of finance. The goal of the experiment was to design a system model for automatic detection of general sentiment and polarity phrases in these texts. We have assembled a document collection from web sources writing on the financial market in Croatia and manually annotated articles from a subset of that collection for general sentiment. Additionally, we have manually annotated a number of these articles for phrases encoding positive or negative sentiment within a text. In the paper, we provide an analysis of the compiled resources. We show a statistically significant correspondence (1) between the overall market trend on the Zagreb Stock Exchange and the number of positively and negatively accented articles within periods of trend and (2) between the general sentiment of articles and the number of polarity phrases within those articles. We use this analysis as an input for designing a rule-based local grammar system for automatic detection of polarity phrases and evaluate it on held out data. The system achieves F1-scores of 0.61 (P: 0.94, R: 0.45) and 0.63 (P: 0.97, R: 0.47) on positive and negative polarity phrases. <s> BIB007 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Language-Independent Feature Selection and Extraction <s> Many methods for cross-lingual processing tasks are resource-dependent, which will not work without machine translation system or bilingual lexicon. In this paper, we propose a novel approach for multilingual sentiment classification just by few seed words. For a given language, the proposed approach learns a sentiment classifier from the initial seed words instead of any labeled data. We employ our method both in supervised learning and unsupervised learning. Experimental results demonstrate that our method relies less on external resource but performs as well as or better than the baseline. <s> BIB008 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Language-Independent Feature Selection and Extraction <s> We describe a novel language-independent approach to the task of determining the polarity, positive or negative, of the author's opinion on a specific topic in natural language text. In particular, weights are assigned to attributes, individual words or word bi-grams, based on their position and on their likelihood of being subjective. The subjectivity of each attribute is estimated in a two-step process, where first the probability of being subjective is calculated for each sentence containing the attribute, and then these probabilities are used to alter the attribute's weights for polarity classification. The evaluation results on a standard dataset of movie reviews shows 89.85% classification accuracy, which rivals the best previously published results for this dataset for systems that use no additional linguistic information nor external resources. <s> BIB009

One way of building sentiment analyses for languages other than English or building systems that work for multiple languages is to extract and select features that do not depend on the language itself. Different approaches have been followed to select and extract these features, including: (1) Entropy Weighted Genetic algorithms BIB005 , (2) Feature Subsumption BIB006 , (3) Local grammar-based methods BIB007 BIB004 , (4) Positional features BIB009 and (5) Common seeds word methods BIB008 . Among these methods, we are aware of only two that have been applied to Arabic: entropy weighted genetic algorithms and local grammar methods. Abbasi et al. BIB005 used genetic algorithms to select language features for both Arabic and English. Genetic Algorithms (GA) are a general class of techniques that apply the concept of evolution to general optimization problems. Entropy Weighted Genetic Algorithms (EWGA) combine genetic algorithms with information gain (IG) to perform the feature selection. In particular, IG is used to select the initial set of features for the initial stage of the GA, and is also applied during the cross-over and mutation stages. EWGA is applied to select features for sentiment analysis in a corpus of Web forum data containing multiple languages BIB005 . They used two types of features, stylistic features and lexical features. They avoided semantic features because they are language dependent and need lexicon resources, while the limitation of their data prevent using of linking features. The paper evaluates the proposed system on a benchmark testbed consisting of 1000 positive and 1000 negative movie reviews. Their system that uses feature selection outperforms several previous systems BIB002 BIB001 BIB003 . Using this system, they achieved an accuracy rate of 91% while other systems achieved accuracy rates between 87-90% on the movie review data set. They were also able to achieve 92% accuracy on Middle Eastern forums and 90% on US forums using the EWGA feature selection method. Use of Local Grammar is another method that can be used to extract sentiment features BIB007 . Ahmed et al. BIB004 applied this approach to documents from the financial news domain. They identified domain-specific key words by looking for words that occurred often in a corpus of financial news but relatively infrequently in a general corpus. Using the context around these words they built a local grammar to extract sentiment-bearing phrases. They applied their approach to Arabic, English and Chinese. They evaluated the system manually and achieved accuracy rates between 60-75% for extracting the sentiment bearing phrases. Importantly, the proposed system could be used to extract the sentiment phrases in financial domain for any language.

Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Standard IR and Hybrid Classifiers <s> For languages with rich content over the web, business reviews are easily accessible via many known websites, e.g., Yelp.com. For languages with poor content over the web like Arabic, there are very few websites (we are actually aware of only one that is indeed unpopular) that provide business reviews. However, this does not mean that such reviews do not exist. They indeed exist unstructured in websites not originally intended for reviews, e.g., Forums and Blogs. Hence, there is a need to mine for those Arabic reviews from the web in order to provide them in the search results when a user searches for a business or a category of businesses. In this paper, we show how to extract the business reviews scattered on the web written in the Arabic language. The mined reviews are analyzed to also provide their sentiments (positive, negative or neutral). This way, we provide our users the information they need about the local businesses in the language they understand, and therefore provide a better search experience for the Middle East region, which mostly speaks Arabic. <s> BIB001 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Standard IR and Hybrid Classifiers <s> Recently, Opinion Mining (OM) is receiving more attention due to the abundance of forums, blogs, ecommerce web sites, news reports and additional web sources where people tend to express their opinions. There are a number of works about Sentiment Analysis (SA) studying the task of identifying the polarity, whether the opinion expressed in a text is positive or negative about a given topic. However, most of research is focused on English texts and there are very few resources for other languages. In this work we present an Opinion Corpus for Arabic (OCA) composed of Arabic reviews extracted from specialized web pages related to movies and films using this language. Moreover, we have translated the OCA corpus into English, generating the EVOCA corpus (English Version of OCA). In the experiments carried out in this work we have used different machine learning algorithms to classify the polarity in these corpora showing that, although the experiments with EVOCA are worse than OCA, the results are comparable with other English experiments, since the loss of precision due to the translation is very slight. <s> BIB002 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Standard IR and Hybrid Classifiers <s> Sentiment analysis is a challenging new task related to text mining and natural language processing. Although there are, at present, several studies related to this theme, most of these focus mainly on English texts. The resources available for opinion mining (OM) in other languages are still limited. In this article, we present a new Arabic corpus for the OM task that has been made available to the scientific community for research purposes. The corpus contains 500 movie reviews collected from different web pages and blogs in Arabic, 250 of them considered as positive reviews, and the other 250 as negative opinions. Furthermore, different experiments have been carried out on this corpus, using machine learning algorithms such as support vector machines and Nave Bayes. The results obtained are very promising and we are encouraged to continue this line of research. © 2011 Wiley Periodicals, Inc. <s> BIB003 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Standard IR and Hybrid Classifiers <s> In this paper, we present a combined approach that automatically extracts opinions from Arabic documents. Most research efforts in the area of opinion mining deal with English texts and little work with Arabic text. Unlike English, from our experiments, we found that using only one method on Arabic opinioned documents produce a poor performance. So, we used a combined approach that consists of three methods. At the beginning, lexicon based method is used to classify as much documents as possible. The resultant classified documents used as training set for maximum entropy method which subsequently classifies some other documents. Finally, k-nearest method used the classified documents from lexicon based method and maximum entropy as training set and classifies the rest of the documents. Our experiments showed that in average, the accuracy moved (almost) from 50% when using only lexicon based method to 60% when used lexicon based method and maximum entropy together, to 80% when using the three combined methods. <s> BIB004

Here we describe systems employing standard IR methods (e.g., TF*IDF) and hybrid classifiers. The work of Rushdi et al. BIB002 BIB003 builds machine learning classifiers using both the OCA and EVOCA corpora. They use both Support Vector Machines (SVMs) and Naive Bayes (NB) classifiers, reporting 90% F-measure on OCA and 86.9% on EVOCA using SVMs. They show that SVMs outperform the NB classifier, which is common in text classification tasks. Results also show that there is no difference between using term frequency (tf) and the slightly more complicated term frequency-inverse document frequency (tf-idf) as weighting schemes. Experiments also show that stemming words before feature extraction and classification nearly always degrades the results. Elhawary and Elfeky BIB001 present a system for sentiment analysis on Arabic business reviews, with the specific goal of building a web search engine that would automatically annotate returned pages with sentiment scores. The system has several components. The first component classifies whether an Internet page is a review or not. For this component, they extend an in-house multilabel classifier to work for Arabic such that its task is to assign a tag from the set {review, forum, blog, news, shopping store} to a document. To build an Arabic review classifier data set, 2000 URLs were collected and more than 40% of them were found to be reviews through manual labeling. This data set was collected by searching the web using keywords that usually exist in reviews (such as "the camera is very bad"). The authors translated the lists of keywords collected and add to them a list of Arabic keywords that usually appear in opinionated Arabic text. The final list contained 1500 features and was used to build an AdaBoost classifier, using 80% of the data for training and the rest for testing. After a document is classified as belonging to the Arabic review class or not, a second component of the system analyzes the document for its sentiment. They build an Arabic lexicon based on a similarity graph for use with the sentiment component. The final component of the system is designed to provide the search engine with an estimate of the sentiment score assigned to a document during the search. A combined classification approach is proposed by El-Halees BIB004 for document level sentiment classification. He applied different classifiers in a sequential manner. A lexicon-based classifier is first used to estimate the sentiment of a document based on an aggregation of all the opinion words and phrases in the document. However, because some documents lack enough opinion words to use this lexicon-based classifier, a second phase uses a maximum entropy classifier. All classified documents from first classifier are used as the training set for this maximum entropy classifier, which is then used to compute the probability that a given document belongs to a certain sentiment class. In particular, if the probability is greater than a threshold of 0.75, then the document is assigned a class, and otherwise the document is passed to the next stage. The final stage is a k-nearest neighbors (KNN) classifier that finds the nearest neighbors for the unannotated document using the training set coming from the previous two classifiers. The corpus used for evaluation consisted of 1134 documents collected from different domains (e.g., education, politics, and sports), with 635 positive documents (with 4375 positive sentences) and 508 negative documents (with 4118 negative sentences). Preprocessing is applied to filter HTML tags and non-textual contents are removed. Alphabets are normalized and some misspelled words are corrected. Sentences are tokenized, stop words are removed, and an Arabic light stemmer is used for stemming the words, and TF-IDF is used for term weighting. The paper reports an f-measure of 81.70% averaged over all domains for positive documents and 78.09% F-measure for negative documents. The best F-measure is obtained in the education domain (85.57% for the positive class and 82.86% for the negative class).

Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Arabic-Specific, Social Media, and Genre-Specific Features <s> In this work, we investigate sentiment mining of Arabic text at both the sentence level and the document level. Existing research in Arabic sentiment mining remains very limited. For sentence-level classification, we investigate two approaches. The first is a novel grammatical approach that employs the use of a general structure for the Arabic sentence. The second approach is based on the semantic orientation of words and their corresponding frequencies, to do this we built an interactive learning semantic dictionary which stores the polarities of the roots of different words and identifies new polarities based on these roots. For document-level classification, we use sentences of known classes to classify whole documents, using a novel approach whereby documents are divided dynamically into chunks and classification is based on the semantic contributions of different chunks in the document. This dynamic chunking approach can also be investigated for sentiment mining in other languages. Finally, we propose a hierarchical classification scheme that uses the results of the sentence-level classifier as input to the document-level classifier, an approach which has not been investigated previously for Arabic documents. We also pinpoint the various challenges that are faced by sentiment mining for Arabic texts and propose suggestions for its development. We demonstrate promising results with our sentence-level approach, and our document-level experiments show, with high accuracy, that it is feasible to extract the sentiment of an Arabic document based on the classes of its sentences. <s> BIB001 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Arabic-Specific, Social Media, and Genre-Specific Features <s> Although Subjectivity and Sentiment Analysis (SSA) has been witnessing a flurry of novel research, there are few attempts to build SSA systems for Morphologically-Rich Languages (MRL). In the current study, we report efforts to partially fill this gap. We present a newly developed manually annotated corpus of Modern Standard Arabic (MSA) together with a new polarity lexicon. The corpus is a collection of newswire documents annotated on the sentence level. We also describe an automatic SSA tagging system that exploits the annotated data. We investigate the impact of different levels of preprocessing settings on the SSA classification task. We show that by explicitly accounting for the rich morphology the system is able to achieve significantly higher levels of performance. <s> BIB002 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Arabic-Specific, Social Media, and Genre-Specific Features <s> Subjectivity and sentiment analysis (SSA) is an area that has been witnessing a flurry of novel research. However, only few attempts have been made to build SSA systems for morphologically-rich languages (MRL). In the current study, we report efforts to partially bridge this gap. We present a newly labeled corpus of Modern Standard Arabic (MSA) from the news domain manually annotated for subjectivity and domain at the sentence level. We summarize our linguistically-motivated annotation guidelines and provide examples from our corpus exemplifying the different phenomena. Throughout the paper, we discuss expression of subjectivity in natural language, combining various previously scattered insights belonging to many branches of linguistics. <s> BIB003 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Arabic-Specific, Social Media, and Genre-Specific Features <s> We present AWATIF, a multi-genre corpus of Modern Standard Arabic (MSA) labeled for subjectivity and sentiment analysis (SSA) at the sentence level. The corpus is labeled using both regular as well as crowd sourcing methods under three different conditions with two types of annotation guidelines. We describe the sub-corpora constituting the corpus and provide examples from the various SSA categories. In the process, we present our linguistically-motivated and genre-nuanced annotation guidelines and provide evidence showing their impact on the labeling task. <s> BIB004 </s> Subjectivity and Sentiment Analysis of Arabic: A Survey <s> Arabic-Specific, Social Media, and Genre-Specific Features <s> In this work, we present SAMAR, a system for Subjectivity and Sentiment Analysis (SSA) for Arabic social media genres. We investigate: how to best represent lexical information; whether standard features are useful; how to treat Arabic dialects; and, whether genre specific features have a measurable impact on performance. Our results suggest that we need individualized solutions for each domain and task, but that lemmatization is a feature in all the best approaches. <s> BIB005

Other techniques use the linguistic features of Arabic in order to perform sentiment analysis, by analyzing the grammatical structure of Arabic BIB001 and Arabic-specific morphological features BIB004 BIB002 BIB003 . Farra et al. BIB001 proposed Arabic sentence-level sentiment classification, considering two different approaches: a grammatical approach and a semantic approach. The grammatical approach is based on Arabic grammatical structure and combines the verbal and nominal sentence structures in one general form based on the idea of actor/action. In this approach, the subjects in verbal and nominal sentences are actors and verbs are actions. They manually label action and actor tags in sentences and used these tags as features. Their feature vector constitutes the following dimensions: sentence type (verbal or nominal), actor, action, object, adjective, type of pronoun and noun, transition (the type of word linking the current sentence with the previous sentence), word polarity (positive, negative, neutral) and sentence class. The second approach proposed by Farra et al. BIB001 combined syntactic and semantic features by extracting some features like the frequency of positive, negative, and neutral words, the frequency of special characters (e.g., "!"), the frequency of emphasis words (e.g., "really" and "especially"), the frequency of conclusive and contradiction words, etc. For extracting the semantics of the words, the paper builds a semantic interactive learning dictionary which stores the semantic polarity of word roots extracted by stemmer. The system asks the user for the polarity of a word if it has not yet been learned. For evaluation of the grammatical approach, only 29 sentences are annotated manually with part-of-speech tags. They report 89.3% accuracy using an SVM classifier with 10-fold cross validation. Sentences from 44 random documents are used for evaluating the semantic and syntactic approach using a J48 decision tree classifier. They report 80% accuracy when the semantic orientation of the words extracted and assigned manually is used, and 62% when the dictionary is used. They also classified the documents by using all sentence features and chunking the document into different parts, reporting 87% accuracy with an SVM classifier when documents divided into 4 chunks and neutral class excluded. Abdul-Mageed et al. BIB004 BIB002 BIB003 created sentence-level annotated Arabic corpora and built subjectivity and sentiment analysis systems exploiting them. In their systems these authors use various types of features, including language independent features, Arabic-specific morphological features, and genre-specific features. In BIB002 , they classify MSA news data at the sentence level for both subjectivity and sentiment. They use a two-stage SVM classifier, where a subjectivity classifier is first used to separate subjective from objective sentence. In a second stage, subjective sentences are classified into positive and negative cases, with an assumption that neutral sentences will be treated in a future system. These authors make use of both language-independent and Arabic-specific features. The language independent features include a feature indicating the domain of the document (e.g., politics, sports) from which a sentence is derived, a unique feature where all words occuring less than four times are replaced by the token "UNIQUE", N-gram features including all N-grams of frequency less than 4, and an adjective feature for adjectives indicating the occurrence of a polarized adjective based on a pre-developed polarity lexicon of 3982 entries. Classification accuracy of 95.52% are reported. The results showed that the adjective feature is very important, as it improved the accuracy by more than 20% and the unique and domain features are also helpful. Other work by Abdul-Mageed et al. BIB005 presents SAMAR, an SVM-based system for Subjectivity and Sentiment Analysis (SSA) for Arabic social media genres. They tackle a number of research questions, including how to best represent lexical information, whether standard features are useful, how to treat Arabic dialects, and whether genre specific features have a measurable impact on performance. The authors exploit data from four social media genres: Wikipedia Talk Pages, Web forums, chat, and Twitter tweets. The corpus includes data in both Modern Standard Arabic and dialectal Arabic. These authors break down their data into 80% training, 10% development, and 10% testing and exploit standard SSA features (e.g., the "unique" feature, a wide coverage polarity lexicon), social and genre features (e.g., the gender of a user), and a binary feature indicating whether a sentence is in MSA or dialectal Arabic. They are able to significantly beat their majority class baselines with most data sets. The results suggest that they need indi-vidualized solutions for each domain and task, but that lemmatization is a feature in all the best approaches. Table 1 summarizes the SSA systems which are described above.

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> <s> One of the greatest challenges facing municipal engineers is the condition rating of buried infrastructure assets, particularly water mains. This is because water mains are typically underground, operated under pressure, and usually inaccessible. Condition rating is a mandatory process to establish and employ management strategies for any asset. To assess the condition of water mains, current research considers physical, environmental, and operational factors and their effect on different types of mains (i.e., cast iron, ductile iron, and asbestos). A condition rating model is developed to assess and set up rehabilitation priority for water mains using the artificial neural network (ANN) approach. Data are collected from different municipalities to train the developed model. The ANN input factors incorporate pipe type, size, age, breakage rate, Hazen-Williams factor, excavation depth, soil type, and top road surface; however, the output is pipe condition. The trained ANN shows robust performance (learning... <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> <s> Abstract Gas leaks can cause major incidents resulting in both human injuries and financial losses. To avoid such situations, a considerable amount of effort has been devoted to the development of reliable techniques for detecting gas leakage. As knowing about the existence of a leak is not always enough to launch a corrective action, some of the leak detection techniques were designed to allow the possibility of locating the leak. The main purpose of this paper is to identify the state-of-the-art in leak detection and localization methods. Additionally we evaluate the capabilities of these techniques in order to identify the advantages and disadvantages of using each leak detection solution. <s> BIB002

includes acoustic and non-acoustic methods, whereas continuous monitoring includes measurement and model-based methods. R e f e r e n c e classified technologies based on the area of inspection, such as internal pipe surface, pipe wall integrity, and pipe bedding/void conditions. R e f e r e n c e s BIB002 , , and classified leak detection systems into non-technical and hardware-and software-based methods. Nontechnical methods do not involve any devices and use only natural senses, such as hearing and smelling, to identify a leak, whereas the technical methods use special devices to identify leaks; in the hardware methods, these devices include liquid sensing cables, vapor sampling, etc., and in the software methods, these devices include negative pressure waves, pressure point analysis, etc. Re f er ence BIB001 divided the leak detection systems into visual, physical, acoustic, ultra-spectrum, and electromagnetic. A similar classification by divides LDS into visual, acoustic, and Electromagnetic-Radio Frequency (EM-RF) techniques. Figure 1 depicts the LDS classification. LDS can be broadly classified into continuous or non-continuous monitoring systems. In non-continuous monitoring systems, the inspection is performed at regular intervals. Depending on the mode of inspection, pipeline operations can either continue or need to stop. For example, visual inspection or a helicopter survey does not require pipeline operations to be stopped, whereas an intelligent pigging system may require the operations to be stopped. The remote sensing of liquid hydrocarbons u s i n g aircraft mounted gas remote sensing is given in . This system detects evaporative plumes from pools of oil, gasoline, condensate, or pentane. Continuous monitoring systems monitor pipelines around the clock and are based on a physical principle. This approach can further be classified into external and internal systems.

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> External Systems <s> Abstract When a gas or liquid pipeline leaks, it will generate broadband acoustic signal. This acoustic pressure will induce an optical phase signal of the optical fiber fixed on the surface of pipes. In this paper, we propose a hybrid configuration of Mach-Zehnder and Sagnac interferometer as sensing frame. In this interferometer, there are two light paths that have the same optical length but travel different sequence paths. Because the propagation lights of the two light paths pass through the leaking point at different times, the resulting phase signals differ respectively. After interference, we demodulate the sensing phase signal by a broadband phase generated carrier (PGC) circuit and then the leaking point can be acquired from null frequency of the output spectrum. This system has many advantages, such as its in-line configuration, all fiber structural design and polarization-insensitive for sensing fiber. This system also has very wide dynamic range which can be greater than 76 dB with the minimum detectable phase signal about 3.3 × 10 − 4 ( rad / Hz ) . <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> External Systems <s> Pipelines have been a vital component of the energy supply chain in India; have to be laid in harsh surroundings; crossing mountain ranges characterized by unstable grounds; where seasonal soil texture changes increase the probability of hazards and uncertainties. Therefore, pipeline monitoring systems for leakage, ground movement, and intrusion detection are part of new pipeline projects. Leakage detection using distributed fibre-optic sensors can be a comprehensive solution for continuous, in-line, real-time monitoring of various pipelines. The monitoring of temperature profiles over long distance by means of optical fibres represents a highly efficient way to perform leakage detection along pipelines. Different techniques have been developed taking advantages of the fibre geometry and of optical time domain analysis for the localization of the information. Raman-based systems have been envisaged for one of the very first projects of India where leakage detection using Distributed Temperature Sensing has been envisaged. The paper presents and discusses the possibility to actively and automatically monitor leakages using distributed fibre optics sensing techniques. The second part of the paper focuses on the monitoring of leakage and third party intrusion detection of petroleum product pipelines. The key features and performances of the technology are reviewed in this paper. <s> BIB002

External systems use local sensors to detect fluids escaping from pipes. Impedance methods use cables with fiber optic or electro-chemical detection to sense liquids. Sniffing methods depend on vapor sensing through tubes. Acoustic methods depend on sensing noises induced by leaks. These systems are highly sensitive to leaks and c a n accurately locate them , . However, due to th e high costs, these methods are employed only in sensitive locations or for short pipeline segments . Vapor sensing cables : In the vapor sensing method, a highly permeable, pressure-tight air tube is fitted along the entire length of the pipe. When a leak occurs, the leaked material diffuses into the tube due to the concentration gradient. After a certain time, an accurate image of the substance surrounding the tube is obtained. A column of air that is pumped at constant speed passes through a gas sensor, and the substance produces a peak, indicating a leak. The increase in the gas concentration produces a leak peak, and the height of the peak is proportional to the concentration (Figure 3 ). Initially, a test gas is injected from one end to pass through the entire length of the pipe. The detector unit marks the start peak and the end peak to calculate the length of the pipe. When a leak occurs, the ratio of leaked distance to the overall distance is calculated to identify the location of the leak. Fiber optic sensing cables: In this method, a fiber op-tic cable is installed along the entire length of the pipeline. When a liquid comes into contact with the cable, the transmission characteristics of the fiber change. While a pulsed laser propagates through the fiber, any changes to the density or composition of the fiber cause the light to scatter backwards. Spectral analysis reveals the temperature profile, leading to leak detection and localization , BIB002 . The process is depicted in Figure 4 . Cable replacement may be required after a leak. Recent advances in fiber optic sensors include quasi-distributed sensing, e.g., integrated Bragg gratings, and distributed sensing, e.g., interferometry and fiber sheath sensors ( , BIB001 ). Figure 4 : Leak detection using a fiber optic cable . When a pipeline leak occurs, the liquid moves from a high-pressure area to a low-pressure area and a turbulent flow is generated. This flow generates a characteristic sound that can be picked up by a specially designed hydrophone. Using sophisticated software, Leak ACO detects this signal, analyzes it, and evaluates the measurement results, thus identifying and providing the location of the leak. Acoustic Emission (AE) testing: When a pipeline leak occurs, turbulent liquid flow occurs in a high-pressure to a low-pressure area, creating a low frequency sound signal. Acoustic sensors affixed to the outside of the pipe, e.g., accelerometers, hydrophones, piezoelectric transducers, etc., pick up these signals. The deviation of the sound signals from a baseline fingerprint triggers a leak alarm. The received signal is stronger near the leak site, enabling localization of t h e leak. For pipes, such as PCCP, wire breaks release energy and cause a series of discrete events. AE monitoring of wire breaks is limited only to on-going wire events and cannot be used to detect already broken wires.

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Pipeline Monitoring using WSN <s> Acoustic emission (AE) was used to measure energy associated with fracture of standard concrete test specimens. The goal of the work was to identify ways in which AE could be used to quantify damage in generic laboratory structures for the purpose of tuning damage models. A series of mortar and concrete specimens of different compositions were tested for fracture energy G-sub-f, while simultaneously being monitored for acoustic emission energy release. Reasonable correlation between the 2 quantities was observed for fine-grained specimens; however, the relationship was not as good for coarse-grained specimens. Toughening mechanisms such as friction are suggested as being responsible for the poor relationship observed in the course-grained materials. It is further suggested that AE energy release can be related to actual crack formation energy but not to friction and other internal energy dissipation or toughening mechanisms. <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Pipeline Monitoring using WSN <s> This research reveals the realization of 5.4 GHz radio frequency bandpass filter utilizing film bulk acoustic wave resonator (FBAR) technology. It can be applied to wireless sensor network systems. Furthermore, it can also be application for wireless 802.11n and 4G-WiMAX system. In this paper, the 5.4 GHz filter has been demonstrated, which has central frequency of 5406 MHz, bandwidth of 70 MHz, insertion loss of −10 dB, return loss of −18 dB, and stopband attenuation of −36 dB, and size of 0.74 mm × 0.42 mm. <s> BIB002 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Pipeline Monitoring using WSN <s> This paper reports on the application of Wigner-Ville Distribution (WVD) method in leak detection of gas transport pipelines. WVD of the acoustic signal inside a pipeline is used to detect the change of the signal caused by leak. In order to reduce the influence of cross term disturbances, the time average of WVD results is calculated and used as the residual to detect leak faults. Experimental results based on real data show that the method can detect the leak faults of long gas transport pipelines effectively. <s> BIB003 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Pipeline Monitoring using WSN <s> This paper describes results of an experimental study that used sensing methods for monitoring damage along segmental concrete pipelines resulting from permanent ground displacement across a simulated earthquake fault. The literature contains examples of such damage occurring during actual earthquakes, significantly impacting the functionality of the pipelines. Detecting the location of the damage and the extent of the damage in pipelines can significantly accelerate post-earthquake repair efforts. In this paper, electrical sensing methods, magnetic sensing, and acoustic emission are used to monitor structural damage in a segmental concrete pipeline during a large-scale test. In this test, the segmental concrete pipeline was subjected to a concentrated transverse permanent ground displacements (PGDs). The majority of the damage to the pipe segments was localized at the joints, especially the bell sections while the damage to the spigots was minimal. The damage extended away from the joints in the pipe segments in the immediate vicinity of the fault line. Telescoping (i.e., crushing of the bell-and-spigot) was a primary mode of failure that was observed. The results of this study indicate that electrical sensing methods (including the use of conductive grout), magnetic sensing, and acoustic emission, employed alone or in combination, can detect and quantify the damage in segmental concrete pipelines. <s> BIB004 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Pipeline Monitoring using WSN <s> This paper describes the results of an experimental study that uses radio frequency identification (RFID) technology to detect cracking in concrete elements. A RFID-based sensor is used to monitor the change in electrical resistance that occurs in conductive materials applied to the surface of the concrete. When the concrete substrate is strained, the conductive material at the surface is stretched, and its electrical resistance increases. If the concrete substrate is strained to the point where it cracks, the conductive material at the surface also cracks, causing its electrical resistance to increase by orders of magnitude. This paper describes how this increase in electrical resistance attributable to cracking can be detected wirelessly by RFID technology. To experimentally illustrate the application of this technology, an RFID-based sensor and conductive surface materials are used to detect cracking in the restrained ring test. The experimental results indicate that this technology can be easily implemented and successfully used for wireless crack detection in concrete and reinforced concrete members. The technology that is described in this paper is not limited to the laboratory environment and can be easily extended to field applications. <s> BIB005

A sensor node in Wireless Sensor Network (WSN) typically consists of transducers (to determine variations in temperature, pressure, strain, etc.), analog-to-digital converters, signal processing, power sources, memory, etc. Typical sensors used in pipeline monitoring are discussed in this section. Acoustic sensors: Acoustic sensors are based on the principle that a liquid escaping through a hole in a pipeline produces a detectable sound. Acoustic sensors are easy to install and maintain and can continuously monitor a long pipeline. An important drawback is their high susceptibility to noise sources, such as system noises, environment noises, radio chatter, wind, Doppler effects, etc. To eliminate system noises, various techniques, such as band pass filtering BIB002 , Fast Fourier Transform (FFT) and time-averaging Wigner-Ville distribution BIB003 , can be used. Acoustic sensors can be used along with other sensors to overcome these limitations. I n t h e y used piezoelectric sensors along with acoustic sensors to identify leaks and other pipe defects. Piezoelectric sensors: Piezoelectric sensors (PZT) or lead zirconate titanate sensors can monitor t h e physical properties of pipelines, such as pressure, acceleration, vibration, acoustic waves, etc., and convert them into electric signals. The strength of the signal is determined by factors, such as the amount of energy released, distance from the monitored event, orientation of the sensor, transmission media, etc. BIB004 , BIB005 , BIB001 . Piezoelectric sensors are a suitable candidate for pipeline monitoring because they exhibit high tolerance to harsh conditions and are not sensitive to electromagnetic radiation. However, piezoelectric sensors are not free from generating false alarms because a sensor deployed to detect one physical quantity may be affected by another, e.g., a pressure sensor may be affected by vibrations in the pipeline. To compensate for this limitation, secondary sensors can be used where the pressure sensors are used along with accelerometer piezoelectric sensors to detect pressure transients.

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Classification and Architecture Models <s> We maintain that the key to intelligent fusion of disparate sensory information is to provide an effective model of sensor capabilities. A sensor model is an abstraction of the actual sensing process. It describes the information a sensor is able to provide, how this information is limited by the environment, how it can be enhanced by information obtained from other sensors, and how it may be improved by active use of the physical sensing device. The importance of having a model of sensor performance is that capabilities can be estimated a priori and, thus, sensor strategies developed in line with information requirements. <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Classification and Architecture Models <s> A formal framework is proposed for defining data fusion processes. Particularly the notion of qualified gain is proposed: gain related to representation, completeness, accuracy and certainty. These notions are applied to a medical monitoring and diagnosis problem where a dynamic Bayesian network is used to model time series of observations and evolving states. The model aims at giving a daily diagnosis. Experiments are under way using data of an already existing system collected on kidney disease patients. Results are be characterized using our notion of qualified gains. <s> BIB002

The fusion of data from multiple sensors, called multi-sensor data fusion, provides more information than a single sensor. Multi-sensor data fusion can also include fusing overlapping measurements from the same sensor obtained at different times. Data fusion improves performance i n at least four ways: representation, accuracy, certainty, and completeness BIB002 . Durrant-Whyte classified data fusion based on the relationship among the sources, such as complementary, competitive, and co-operative BIB001 . Complementary: Non-redundant data from different sensors can be fused to provide a complete view.

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Redundant(competitive): <s> In this paper, we consider two important problems for distributed fault-tolerant detection in wireless sensor networks: 1) how to address both the noise-related measurement error and sensor fault simultaneously in fault-tolerant detection and 2) how to choose a proper neighborhood size n for a sensor node in fault correction such that the energy could be conserved. We propose a fault-tolerant detection scheme that explicitly introduces the sensor fault probability into the optimal event detection process. We mathematically show that the optimal detection error decreases exponentially with the increase of the neighborhood size. Experiments with both Bayesian and Neyman-Pearson approaches in simulated sensor networks demonstrate that the proposed algorithm is able to achieve better detection and better balance between detection accuracy and energy usage. Our work makes it possible to perform energy-efficient fault-tolerant detection in a wireless sensor network. <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Redundant(competitive): <s> As a safe and economical method the pipeline is widely used to transport oil and gas. At present, most of the serious pipeline accidents are the artificial damages. The traditional leakage detection methods only infer the leakages by analyzing the parameters. The sensitivities are low and the leakage warning exclusively arrives after the leakage. The urgent problem is to detect the intrusions and locate the threat position. For this reason, the authors studied and developed a novel pipeline security monitoring and pre-warning system based on multi-seismic sensors. The system consists of many seismic geophones and data acquisition modules which are laid along the pipeline. Events such as footsteps, car moving and digging generate seismic waves. These sensors and modules acquire the seismic signals. The empirical mode decomposition (EMD) is used to acquire the intrinsic mode functions (IMFs). Then the IMF kurtosis features are normalized and used to recognize the different targets. In order to improve the individual target recognition accuracy, it is proposed to fuse the identification results by using the Dempster-Shafer evidence theory and obtain the final judgment. If the target belongs to a threat, the location will be finished. The time-frequency analysis method is used to process the seismic signals. The EMD outcomes are transformed by Hilbert-Huang Transform (HHT) and then the signal's instantaneous properties such as instantaneous frequency, phase and amplitude can be acquired. In this paper the authors also discuss the localization procedures and put forward a new passive localization method based on HHT and time difference of arrival (TDOA). The field trials show the proposed methods work well. <s> BIB002 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Redundant(competitive): <s> Multi-Sensor Data Fusion (MSDF) is very rapidly growing as an independent discipline to be considered with and finds applications in many areas. Surplus and complementary sensor data can be fused using multi-sensor fusion techniques to enhance system competence and consistency. The objective of this work is to evaluate multi sensor data fusion algorithms for target tracking. Target tracking using observations from several sensors can achieve improved estimation performance than a single sensor. In this work, three data fusion algorithms based on Kalman filter namely State Vector Fusion (SVF), Measurement Fusion (MF) and Gain fusion (GF) are implemented in a tracking system. Using MATLAB, these three methods are compared and performance metrics are computed for the evaluation of algorithms. The results show that State Vector Fusion estimates the states well when compared to Measurement Fusion and Gain Fusion. <s> BIB003

The same pieces of data from a single sensor or multiple sensors can be fused to increase the associated confidence. Co-operative: Different data can be fused to provide a realistic view. The abstraction levels of the input and output in the fusion process, including the measurement, signal, feature, and decision, can also form a basis for classification. Re f e r e nc e BIB001 applied these levels to classify fusion into signal fusion, pixel fusion, feature-level fusion, and symbol fusion. Boudjemaa Figure 7 : Signal Processing using EMD, HHT and Kurtosis BIB002 . Reference provided a data centric taxonomy of data fusion methodologies and discussed the fusion of imperfect data, the fusion of correlated data, and the fusion of inconsistent data. Imperfect data can be fused using probabilistic, evidential, fuzzy reasoning, possibility theory, rough set theory, random set, and hybridization approaches. The fusion of correlated data can be achieved using correlation elimination and correlation presence. The fusion of inconsistent data focuses on removing the outliers, disorders, and conflicts. Reference BIB003 evaluated the performance of data fusion algorithms based on the Kalman filter, i.e., state vector fusion, measurement fusion, and gain fusion.

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Existing Data Fusion (DF) Approaches <s> This paper presents a technique that can be used to fuse data from multiple sensors that are employed in nondestructive evaluation (NDE) applications, specifically for the in-line inspection of gas transmission pipelines. A radial basis function artificial neural network is used to perform geometric transformations on data obtained from multiple sources. The technique allows the user to define the redundant and complementary information present in the data sets. The efficacy of the algorithm is demonstrated using experimental images obtained from the NDE of a test specimen suite using magnetic flux leakage (MFL), ultrasonic (UT) and thermal imaging methods. The results presented in this paper indicate that neural network based geometric transformation algorithms show considerable promise in multi-sensor data fusion applications. <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Existing Data Fusion (DF) Approaches <s> As a safe and economical method the pipeline is widely used to transport oil and gas. At present, most of the serious pipeline accidents are the artificial damages. The traditional leakage detection methods only infer the leakages by analyzing the parameters. The sensitivities are low and the leakage warning exclusively arrives after the leakage. The urgent problem is to detect the intrusions and locate the threat position. For this reason, the authors studied and developed a novel pipeline security monitoring and pre-warning system based on multi-seismic sensors. The system consists of many seismic geophones and data acquisition modules which are laid along the pipeline. Events such as footsteps, car moving and digging generate seismic waves. These sensors and modules acquire the seismic signals. The empirical mode decomposition (EMD) is used to acquire the intrinsic mode functions (IMFs). Then the IMF kurtosis features are normalized and used to recognize the different targets. In order to improve the individual target recognition accuracy, it is proposed to fuse the identification results by using the Dempster-Shafer evidence theory and obtain the final judgment. If the target belongs to a threat, the location will be finished. The time-frequency analysis method is used to process the seismic signals. The EMD outcomes are transformed by Hilbert-Huang Transform (HHT) and then the signal's instantaneous properties such as instantaneous frequency, phase and amplitude can be acquired. In this paper the authors also discuss the localization procedures and put forward a new passive localization method based on HHT and time difference of arrival (TDOA). The field trials show the proposed methods work well. <s> BIB002 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Existing Data Fusion (DF) Approaches <s> AbstractA multilayer data-driven framework for robust structural health monitoring based on a comprehensive application of machine learning and signal processing techniques is introduced. This paper focuses on demonstrating the effectiveness of the framework for damage detection in a steel pipe under environmental and operational variations. The pipe was instrumented with piezoelectric wafers that can generate and sense ultrasonic waves. Damage was simulated physically by a mass scatterer grease-coupled to the surface of the pipe. Benign variations included variable internal air pressure and ambient temperature over time. Ultrasonic measurements were taken on three different days with the scatterer placed at different locations on the pipe. The wave patterns are complex and difficult to interpret, and it is even more difficult to differentiate the changes produced by the scatterer from the changes produced by benign variations. The sensed data were characterized by 365 features extracted from a variety of... <s> BIB003

Some of the data fusion schemes applied in pipeline monitoring are discussed as follows: Homogeneous DF of seismic pulses: Reference BIB002 used Dempster-Shafer (D-S) method to fuse data from multiple seismic sensors in a proactive pipeline monitoring system. The data fusion increased the accuracy of the decisions by 8-25%. To detect the seismic pulses, geophones were deployed along the length of the pipeline at a depth of approximately half a meter, with varied sensor spacing. Different sources, such as people walking, driving a car, manual digging, etc., generated seismic signals with different frequencies, as shown in Figure 7a . The signals were then amplified, filtered, and A/D converted, followed by the extraction of the features. Figure 7b shows the decomposition of the original time series data into intrinsic oscillation mode functions (IMF) using empirical mode decomposition (EMD). Each IMF component was subjected to a Hilbert-Huang Transform (HHT) to obtain the amplitude and frequency. Normalized Kurtosis gives the feature vectors of different signals, as shown in Figure 7c , and extracts the features of each target. Data fusion was applied to the inputs from multiple seismic sensors with the D-S method using the rule given in Eq. 1. Data driven framework using DF: Reference BIB003 proposed a data driven framework that used piezoelectric wafers to generate and sense ultrasonic waves. Multiple signal processing techniques were applied to extract as many as 365 features. The wave patterns were then checked u s i n g an adaptive boosting algorithm and five machine learning classifiers for damage detection. The system was shown to exhibit an average ac c ura c y of 84.2-89%. As shown in Figure 8 , the received signals were preprocessed to remove low/high frequency vibrations via a band pass filter of t h e range f r o m 190 to 450 kHz. By normalizing the signals, the ambient effects were contained to a certain degree. Various signal processing methods, such as the wavelet transform, Hilbert transform, Mellin transform, etc., were then applied to extract features. A voting strategy-based data fusion in a spatially distributed sensor network is given in BIB001 . Certain features, such as the normalized mean squared error (MSE), correlation coefficient, curve length, loss of local coherence, etc., were used to detect the damage. For data fusion, the independent decisions were fused to arrive at the outcome, i.e., decision-level data fusion was employed. The features for all monitored signals were compared against the threshold, as shown in Figure 9 . A lower false alarm rate means t h a t the result is highly accurate. For each transducer pair, a voting strategy was used to increase the credibility. The system seemed to increase the detection probability to more than 90% and reduced the false alarms to under 5%. Figure 9 : Sensor and Feature fusion at the decision level BIB001 .

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Fusion of GPR and EMI for buried pipes: <s> Sudden pipe breaks occur in water transmission pipelines and distribution mains. The consequences of these breaks can be very expensive because of the service interruption, the cost of repair, and damage to surrounding property and infrastructure. The costs associated with the pipeline breaks can be reduced by minimizing the break detection and location time. This paper presents a new continuous monitoring approach for detecting and locating breaks in pipelines. A sudden pipe break creates a negative pressure wave that travels in both directions away from the break point and is reflected at the pipeline boundaries. Using the pressure data measured at one location along the pipeline, the timing of the initial and reflected transient waves induced by the break determines the location of the break. The magnitude of the transient wave provides an estimate of the break size. The continuous monitoring technique uses a modified two-sided cumulative sum (CUSUM) algorithm to detect abrupt break-induced changes in the pressure data. The adaptive tuning of CUSUM parameters is implemented to detect breaks of differing sizes and opening times. The continuous monitoring technique is verified by using results from both laboratory and field experiments and shows potential for detecting and locating sudden breaks in real pipelines. <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Fusion of GPR and EMI for buried pipes: <s> Underground water and waste-water pipelines are essential arteries and veins of developed cities and countries. Being hidden under the surface of the earth, these pipelines are at risk of being abandoned or damaged by “blind” construction machines digging the ground to place new utilities or undertake road or other construction works. Moreover, due to the high costs associated with the inspection and condition assessment of water pipelines, which can require the pipeline to be drained, water utility companies are often reluctant to inspect them. With the introduction of new and increasingly strict rules and regulations imposed by governments, the safety, security, structural integrity, and sustainability in operation of water pipelines are becoming strategic factors for utility companies in their quest to offer low failure probabilities under controlled and optimized costs. This paper, therefore, reviews the state-of-the-art techniques and sensor technologies currently used for either the location or the condition assessment of underground pipelines. It also introduces a new multisensor tool, typically employing ground penetrating radar and electromagnetic induction sensors, intended to simultaneously locate and assess the condition of underground pipelines from above the ground, hence minimizing the costs and reducing the problems associated with in-line inspections. Moreover, a novel multisensor data fusion system and architecture will also be introduced to fuse the information acquired from the utilized sensors. This novel architecture is designed within a heuristic framework, typically employing an artificial neural network at its heart. <s> BIB002

A multisensory system was used to fuse data from seismic, GPR, and EMI sensors to reduce false alarms in landmine detection BIB001 . The responses of the three types of sensors to the soil condition differed from each other. EMI sensors were sensitive to soil conductivity, while the seismic sensor was sensitive to the difference between the mechanical properties of the soil and t h e landmine; GPR was sensitive to dielectric properties. A good use of the complementary features from sensors reduced the false alarm rate significantly. Reference BIB002 proposed a multisensory data fusion architecture to assess the locations and structural conditions of the buried pipes. Using a combination of ground penetrating radar and electromagnetic waves, inspection can be performed without draining the pipelines. GPR can detect the presence and depth information of buried pipes but cannot distinguish metallic and plastic pipes. Electromagnetic sensors can detect the condition of metallic pipes but cannot measure the depth. A data fusion algorithm that is based on artificial neural networks and uses a combination of inputs from GPR and EMI can detect and classify various defects, such as major cracks and leaks in pipelines. However, this architecture is only conceptual, and the implementation results are n o t yet known. Redundancy increased the reliability of defect characterization by identifying the common information in different NDE methods. Complementarity improved the accuracy of defect characterization by identifying the defect characteristics unique to each inspection method.

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Expensive <s> This research reveals the realization of 5.4 GHz radio frequency bandpass filter utilizing film bulk acoustic wave resonator (FBAR) technology. It can be applied to wireless sensor network systems. Furthermore, it can also be application for wireless 802.11n and 4G-WiMAX system. In this paper, the 5.4 GHz filter has been demonstrated, which has central frequency of 5406 MHz, bandwidth of 70 MHz, insertion loss of −10 dB, return loss of −18 dB, and stopband attenuation of −36 dB, and size of 0.74 mm × 0.42 mm. <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Expensive <s> Strongly reverberating diffuse-like ultrasonic waves can interrogate large areas of complex structures that do not support more easily interpreted guided waves. However, sensitivity to environmental changes such as temperature and surface wetting can degrade the performance of a structural health monitoring system using these types of waves. Surface wetting is investigated here with a simplified experiment where controlled amounts of water are applied to the surface of a specimen in conjunction with incrementally introduced artificial damage. A feature-based approach is taken whereby differential features between a signal and a baseline are defined that are sensitive to damage but less sensitive to surface wetting, and multiple features obtained from a spatially distributed sensor array are combined via a voting strategy. In addition, the features considered are insensitive to moderate temperature changes, which are unavoidable even in the laboratory. Experimental results show a probability of detection greater than 90% when detecting damage in the presence of modest surface wetting while maintaining a false alarm rate under 5%. <s> BIB002 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Expensive <s> Wireless Sensor Networks (WSN) are ad-hoc networks in which small independent sensor nodes have limited energy, computational resources and wireless communication capabilities. Recently, both academia and industry have shown great interest in the area of Wireless Sensor Networks. This paper focuses on the practical applications in commerce and feasible future employment of WSNs. Continued advances of wireless communication technologies have led to the following fields of applications: habitat and environmental monitoring, security of buildings and property, oil and gas installations, mines and tunnels, emergency medical care, military applications. In the near future WSNs will certainly enter our homes and offices changing the way we monitor our nearest surrounding. <s> BIB003 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Expensive <s> Underground water and waste-water pipelines are essential arteries and veins of developed cities and countries. Being hidden under the surface of the earth, these pipelines are at risk of being abandoned or damaged by “blind” construction machines digging the ground to place new utilities or undertake road or other construction works. Moreover, due to the high costs associated with the inspection and condition assessment of water pipelines, which can require the pipeline to be drained, water utility companies are often reluctant to inspect them. With the introduction of new and increasingly strict rules and regulations imposed by governments, the safety, security, structural integrity, and sustainability in operation of water pipelines are becoming strategic factors for utility companies in their quest to offer low failure probabilities under controlled and optimized costs. This paper, therefore, reviews the state-of-the-art techniques and sensor technologies currently used for either the location or the condition assessment of underground pipelines. It also introduces a new multisensor tool, typically employing ground penetrating radar and electromagnetic induction sensors, intended to simultaneously locate and assess the condition of underground pipelines from above the ground, hence minimizing the costs and reducing the problems associated with in-line inspections. Moreover, a novel multisensor data fusion system and architecture will also be introduced to fuse the information acquired from the utilized sensors. This novel architecture is designed within a heuristic framework, typically employing an artificial neural network at its heart. <s> BIB004 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Expensive <s> Every day, millions of people depend on the safe and reliable transportation of oil and gas through pipelines. Pipelines are historically a very safe means of transporting large quantities of oil, natural gas, fuels, and other hazardous materials. However, accidents of pipeline leakage have occurred frequently because of pipelines age, rot and third-party damage and destruction. Waste and pollution occur and result in such serious disasters as fire and explosions. A dependable leak detection system is important to promptly identify the occurrence of a leak in order to shut down the line, isolate the leak, initiate response actions, reduce the volume of the spill, and mitigate safety, environmental, and economic consequences of the release. Leak detection methods cover a wide spectrum of technologies and are based on a number of different detection principles. They vary from intermittent aerial inspections to hydrocarbon sensors to Supervisory Control and Data Acquisition (SCADA) based, real-time monitoring. Each approach has its strengths and weaknesses. This paper describes the capabilities and limitations of current leak detection systems. The operational principle, data and equipment requirements, strengths, weaknesses, and realistic performance limits for the leak detection methods are addressed in this paper. <s> BIB005

Easy to break but difficult to repair , BIB001 Acoustic-Inline Applicable for all Pipe size and diameter Tethered accurately pinpoint leaks, non-tethered can survey long distances Risk of losing free swimming hydrophones Tethered hydrophone needs flow rate to flow along the pipe BIB004 , BIB002 , BIB005 , BIB003 Broadband Electromagnetic(BEM)

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Independent of frequency so independent of electromagnetic interference Detect cracks and other anomalies <s> Wireless Sensor Networks (WSN) are ad-hoc networks in which small independent sensor nodes have limited energy, computational resources and wireless communication capabilities. Recently, both academia and industry have shown great interest in the area of Wireless Sensor Networks. This paper focuses on the practical applications in commerce and feasible future employment of WSNs. Continued advances of wireless communication technologies have led to the following fields of applications: habitat and environmental monitoring, security of buildings and property, oil and gas installations, mines and tunnels, emergency medical care, military applications. In the near future WSNs will certainly enter our homes and offices changing the way we monitor our nearest surrounding. <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Independent of frequency so independent of electromagnetic interference Detect cracks and other anomalies <s> Underground water and waste-water pipelines are essential arteries and veins of developed cities and countries. Being hidden under the surface of the earth, these pipelines are at risk of being abandoned or damaged by “blind” construction machines digging the ground to place new utilities or undertake road or other construction works. Moreover, due to the high costs associated with the inspection and condition assessment of water pipelines, which can require the pipeline to be drained, water utility companies are often reluctant to inspect them. With the introduction of new and increasingly strict rules and regulations imposed by governments, the safety, security, structural integrity, and sustainability in operation of water pipelines are becoming strategic factors for utility companies in their quest to offer low failure probabilities under controlled and optimized costs. This paper, therefore, reviews the state-of-the-art techniques and sensor technologies currently used for either the location or the condition assessment of underground pipelines. It also introduces a new multisensor tool, typically employing ground penetrating radar and electromagnetic induction sensors, intended to simultaneously locate and assess the condition of underground pipelines from above the ground, hence minimizing the costs and reducing the problems associated with in-line inspections. Moreover, a novel multisensor data fusion system and architecture will also be introduced to fuse the information acquired from the utilized sensors. This novel architecture is designed within a heuristic framework, typically employing an artificial neural network at its heart. <s> BIB002

Works only for ferrous pipes, Intensive postprocessing Pipe must be drained, exposed and opened Exorbitant amount of data to be processed Manual, time consuming and labor intensive Subjective as it depends heavily on expert judgment BIB001 , BIB002 Closed Circuit Television Inspection (CCTV) Examine pipe wall surface for defects

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Lack of consistency and reliability <s> Sensor fusion models have been characterized in the literature in a number of distinctly different ways: in terms of information levels at which the fusion is accomplished; the objectives of the fusion process, the application domain; the types of sensors employed, the sensor suite configuration and so on. The characterization most commonly encountered in the rapidly growing sensor fusion literature based on level of detail in the information is that of the now well known triplet: data level, feature level, and decision level. We consider here a generalized input-output (I/O) descriptor pair based characterization of the sensor fusion process that can be looked upon as a natural out growth of the trilevel characterization. The fusion system design philosophy expounded here is that an exhaustive exploitation of the sensor fusion potential should explore fusion under all of the different I/O-based fusion modes conceivable under such a characterization. Fusion system architectures designed to permit such exploitation offer the requisite flexibility for developing the most effective fusion system designs for a given application. A second facet of this exploitation is aimed at exploring the new concept of self-improving multisensor fusion system architectures wherein the central (fusion system) and focal (individual sensor subsystems) decision makers mutually enhance the other's performance by providing reinforced learning. A third facet is that of investigating fusion system architectures for environments wherein the different local decision makers may only be capable of narrower decisions that span only a subset of decision choices. The paper discusses these flexible fusion system architectures along with related issues and illustrates them with examples of their application to real-world scenarios. <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Lack of consistency and reliability <s> A major UK initiative, entitled Mapping the Underworld (MTU), is seeking to address the serious social, environmental and economic consequences arising from an inability to locate accurately and completely the buried utility service infrastructure without resorting to excavations. One of the four MTU projects aims to develop and prove the efficacy of a multi-sensor device for accurate remote buried utility service detection, location and, where possible, utility identification. This paper aims to introduce the MTU programme followed by a state-of-the-art review of the three essential technologies that are to be combined in the device – ground penetrating radar (GPR), low-frequency quasi-static electromagnetic fields and acoustics – and a summary of the influence of different soil types and states on the transmission of the various signals, and therefore how the techniques might be optimised from a knowledge of the ground instead of using very broad simplifying assumptions. The latest developments in impulse GPR, frequency modulated continuous waveform (FMCW) GPR and stepped frequency continuous waveform (SFCW) GPR are described and previous attempts to combine GPR with other sensing technologies are introduced. The work on quasi-static fields explores the ‘fields-of-opportunity’ related to the 50 Hz currents flowing in existing underground power circuits and the electric field variations when low-frequency current in actively induced into the ground. Acoustic techniques have been primarily used for leak detection and the review focuses on the potential for their application to buried utility service location. The paper concludes with a discussion of the facilities required, and currently available, for comprehensive assessment and independent verification of the performance of both existing devices/technologies and of the multi-sensor device under development. <s> BIB002 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Lack of consistency and reliability <s> This research reveals the realization of 5.4 GHz radio frequency bandpass filter utilizing film bulk acoustic wave resonator (FBAR) technology. It can be applied to wireless sensor network systems. Furthermore, it can also be application for wireless 802.11n and 4G-WiMAX system. In this paper, the 5.4 GHz filter has been demonstrated, which has central frequency of 5406 MHz, bandwidth of 70 MHz, insertion loss of −10 dB, return loss of −18 dB, and stopband attenuation of −36 dB, and size of 0.74 mm × 0.42 mm. <s> BIB003 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Lack of consistency and reliability <s> Wireless Sensor Networks (WSN) are ad-hoc networks in which small independent sensor nodes have limited energy, computational resources and wireless communication capabilities. Recently, both academia and industry have shown great interest in the area of Wireless Sensor Networks. This paper focuses on the practical applications in commerce and feasible future employment of WSNs. Continued advances of wireless communication technologies have led to the following fields of applications: habitat and environmental monitoring, security of buildings and property, oil and gas installations, mines and tunnels, emergency medical care, military applications. In the near future WSNs will certainly enter our homes and offices changing the way we monitor our nearest surrounding. <s> BIB004 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Lack of consistency and reliability <s> Underground water and waste-water pipelines are essential arteries and veins of developed cities and countries. Being hidden under the surface of the earth, these pipelines are at risk of being abandoned or damaged by “blind” construction machines digging the ground to place new utilities or undertake road or other construction works. Moreover, due to the high costs associated with the inspection and condition assessment of water pipelines, which can require the pipeline to be drained, water utility companies are often reluctant to inspect them. With the introduction of new and increasingly strict rules and regulations imposed by governments, the safety, security, structural integrity, and sustainability in operation of water pipelines are becoming strategic factors for utility companies in their quest to offer low failure probabilities under controlled and optimized costs. This paper, therefore, reviews the state-of-the-art techniques and sensor technologies currently used for either the location or the condition assessment of underground pipelines. It also introduces a new multisensor tool, typically employing ground penetrating radar and electromagnetic induction sensors, intended to simultaneously locate and assess the condition of underground pipelines from above the ground, hence minimizing the costs and reducing the problems associated with in-line inspections. Moreover, a novel multisensor data fusion system and architecture will also be introduced to fuse the information acquired from the utilized sensors. This novel architecture is designed within a heuristic framework, typically employing an artificial neural network at its heart. <s> BIB005 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Lack of consistency and reliability <s> Every day, millions of people depend on the safe and reliable transportation of oil and gas through pipelines. Pipelines are historically a very safe means of transporting large quantities of oil, natural gas, fuels, and other hazardous materials. However, accidents of pipeline leakage have occurred frequently because of pipelines age, rot and third-party damage and destruction. Waste and pollution occur and result in such serious disasters as fire and explosions. A dependable leak detection system is important to promptly identify the occurrence of a leak in order to shut down the line, isolate the leak, initiate response actions, reduce the volume of the spill, and mitigate safety, environmental, and economic consequences of the release. Leak detection methods cover a wide spectrum of technologies and are based on a number of different detection principles. They vary from intermittent aerial inspections to hydrocarbon sensors to Supervisory Control and Data Acquisition (SCADA) based, real-time monitoring. Each approach has its strengths and weaknesses. This paper describes the capabilities and limitations of current leak detection systems. The operational principle, data and equipment requirements, strengths, weaknesses, and realistic performance limits for the leak detection methods are addressed in this paper. <s> BIB006

Surveys only above the waterline Real time assessment needed-though some new automated processing techniques BIB001 , BIB002 , , BIB003 Eddy Current Good for small metallic pipes Access to pipe required Skin effect limits testing only on the surface near to the probe , BIB003 Ground penetrating radars (GPR) Used from the surface Independent of pipe materials Hard to interpret, highly skilled personnel needed to interpret results Need to choose a right frequency for different Soils Metal objects in ground can raise false alarms BIB006 , BIB004 , , Impact Echo or Spectral Analysis of Surface waves detects voids, cracks and overall condition can detect entire length of pipe investigate both pipe and soil conditions Thorough cleaning needed Access to pipe needed to excite the pipe Presence of tuberculation in pipe mains will render the echo dysfunctional Cannot detect extent of cracks Infrared Thermography Used from the surface, non-invasive Can cover large areas without excavation Accurately determine geometry and defects Can scan entire length of pipe Useful only with liquids and gas having higher temperature than surroundings Weather restrictions such as wind speed and ground cover can influence results Expensive Significant Training and Experience needed Unable to measure depth BIB003 , BIB004 , , BIB005 Laser Scanning Reduce the cost of testing considerably Can be coupled with algorithms to classify the detects Inspect only dry portions of pipe Time consuming BIB006 , BIB001 , BIB003 Listening Sticks

Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Simple and cheap <s> A formal framework is proposed for defining data fusion processes. Particularly the notion of qualified gain is proposed: gain related to representation, completeness, accuracy and certainty. These notions are applied to a medical monitoring and diagnosis problem where a dynamic Bayesian network is used to model time series of observations and evolving states. The model aims at giving a daily diagnosis. Experiments are under way using data of an already existing system collected on kidney disease patients. Results are be characterized using our notion of qualified gains. <s> BIB001 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Simple and cheap <s> Current analysis of Acoustic Emission (AE) signals is dominated by waveform parameter assessment, such as amplitude, rise-time, absolute energy, initiation frequency etc. whilst the actual waveform collected is often ignored. This paper explores the dispersive behaviour of Lamb Wave propagation as a means of estimating source to sensor distance and source orientation from collected waveforms. <s> BIB002 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Simple and cheap <s> A major UK initiative, entitled Mapping the Underworld (MTU), is seeking to address the serious social, environmental and economic consequences arising from an inability to locate accurately and completely the buried utility service infrastructure without resorting to excavations. One of the four MTU projects aims to develop and prove the efficacy of a multi-sensor device for accurate remote buried utility service detection, location and, where possible, utility identification. This paper aims to introduce the MTU programme followed by a state-of-the-art review of the three essential technologies that are to be combined in the device – ground penetrating radar (GPR), low-frequency quasi-static electromagnetic fields and acoustics – and a summary of the influence of different soil types and states on the transmission of the various signals, and therefore how the techniques might be optimised from a knowledge of the ground instead of using very broad simplifying assumptions. The latest developments in impulse GPR, frequency modulated continuous waveform (FMCW) GPR and stepped frequency continuous waveform (SFCW) GPR are described and previous attempts to combine GPR with other sensing technologies are introduced. The work on quasi-static fields explores the ‘fields-of-opportunity’ related to the 50 Hz currents flowing in existing underground power circuits and the electric field variations when low-frequency current in actively induced into the ground. Acoustic techniques have been primarily used for leak detection and the review focuses on the potential for their application to buried utility service location. The paper concludes with a discussion of the facilities required, and currently available, for comprehensive assessment and independent verification of the performance of both existing devices/technologies and of the multi-sensor device under development. <s> BIB003 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Simple and cheap <s> This research reveals the realization of 5.4 GHz radio frequency bandpass filter utilizing film bulk acoustic wave resonator (FBAR) technology. It can be applied to wireless sensor network systems. Furthermore, it can also be application for wireless 802.11n and 4G-WiMAX system. In this paper, the 5.4 GHz filter has been demonstrated, which has central frequency of 5406 MHz, bandwidth of 70 MHz, insertion loss of −10 dB, return loss of −18 dB, and stopband attenuation of −36 dB, and size of 0.74 mm × 0.42 mm. <s> BIB004 </s> Pipeline Leak Detection Systems and Data Fusion: A Survey <s> Simple and cheap <s> Dense networks of low-cost wireless sensors have the potential to facilitate prolific data collection in large and complex infrastructure at costs lower than those historically associated with tethered counterparts. While wireless telemetry has been previously proposed for structural monitoring, comparatively less research has focused on the creation of a complete and scalable data management system that manages the storage and interrogation of wireless sensor data. This paper reports on the development of a novel wireless structural monitoring system specifically tailored for large-scale civil infrastructure systems by architecturally combining dense wireless sensor networks with a suite of information technologies remotely accessible by the Internet. The architectural overview of the proposed Internet-enabled wireless structural monitoring system is presented including a description of its functional elements (for example, wireless sensors, database server, and application programming interfaces). The monitoring-system architecture proposed is validated on the New Carquinez (Alfred Zampa Memorial) Bridge in Vallejo, California. A permanent wireless monitoring system is installed consisting of 28 wireless sensor nodes collecting data from over 80 channels. The bridge sensor data are transferred by a wireless cellular connection to a remote database server where it is stored and available for interrogation by software clients granted access to the data. To illustrate the ability to autonomously process the bridge response data, the stochastic subspace identification method is used to extract accurate modal characteristics of the bridge that are used to update high-fidelity finite-element models of the bridge. The Internet-enabled wireless structural monitoring system proved to be scalable to a large number of nodes and has thus far proven stable and reliable over long-term use. <s> BIB005

Success depends on experience of user Background noise can cause erroneous detection Can only detect area of the leak, not the number and positions of leaks BIB004 , , In-line MFL has size limitation, external MFL requires costly excavation of pipes BIB005 , BIB002 , BIB003 , , BIB004 , BIB001 Remote

Spatiotemporal Data Clustering: A Survey of Methods <s> Introduction <s> Finding useful patterns in large datasets has attracted considerable interest recently, and one of the most widely studied problems in this area is the identification of clusters, or densely populated regions, in a multi-dimensional dataset. Prior work does not adequately address the problem of large datasets and minimization of I/O costs.This paper presents a data clustering method named BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies), and demonstrates that it is especially suitable for very large databases. BIRCH incrementally and dynamically clusters incoming multi-dimensional metric data points to try to produce the best quality clustering with the available resources (i.e., available memory and time constraints). BIRCH can typically find a good clustering with a single scan of the data, and improve the quality further with a few additional scans. BIRCH is also the first clustering algorithm proposed in the database area to handle "noise" (data points that are not part of the underlying pattern) effectively.We evaluate BIRCH's time/space efficiency, data input order sensitivity, and clustering quality through several experiments. We also present a performance comparisons of BIRCH versus CLARANS, a clustering method proposed recently for large datasets, and show that BIRCH is consistently superior. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Introduction <s> Cluster analysis is a primary method for database mining. It is either used as a stand-alone tool to get insight into the distribution of a data set, e.g. to focus further analysis and data processing, or as a preprocessing step for other algorithms operating on the detected clusters. Almost all of the well-known clustering algorithms require input parameters which are hard to determine but have a significant influence on the clustering result. Furthermore, for many real-data sets there does not even exist a global parameter setting for which the result of the clustering algorithm describes the intrinsic clustering structure accurately. We introduce a new algorithm for the purpose of cluster analysis which does not produce a clustering of a data set explicitly; but instead creates an augmented ordering of the database representing its density-based clustering structure. This cluster-ordering contains information which is equivalent to the density-based clusterings corresponding to a broad range of parameter settings. It is a versatile basis for both automatic and interactive cluster analysis. We show how to automatically and efficiently extract not only 'traditional' clustering information (e.g. representative points, arbitrary shaped clusters), but also the intrinsic clustering structure. For medium sized data sets, the cluster-ordering can be represented graphically and for very large data sets, we introduce an appropriate visualization technique. Both are suitable for interactive exploration of the intrinsic clustering structure offering additional insights into the distribution and correlation of the data. <s> BIB002 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Introduction <s> When clusters with different densities and noise lie in a spatial point set, the major obstacle to classifying these data is the determination of the thresholds for classification, which may form a series of bins for allocating each point to different clusters. Much of the previous work has adopted a model-based approach, but is either incapable of estimating the thresholds in an automatic way, or limited to only two point processes, i.e. noise and clusters with the same density. In this paper, we present a new density-based cluster method (DECODE), in which a spatial data set is presumed to consist of different point processes and clusters with different densities belong to different point processes. DECODE is based upon a reversible jump Markov Chain Monte Carlo (MCMC) strategy and divided into three steps. The first step is to map each point in the data to its mth nearest distance, which is referred to as the distance between a point and its mth nearest neighbor. In the second step, classification thresholds are determined via a reversible jump MCMC strategy. In the third step, clusters are formed by spatially connecting the points whose mth nearest distances fall into a particular bin defined by the thresholds. Four experiments, including two simulated data sets and two seismic data sets, are used to evaluate the algorithm. Results on simulated data show that our approach is capable of discovering the clusters automatically. Results on seismic data suggest that the clustered earthquakes, identified by DECODE, either imply the epicenters of forthcoming strong earthquakes or indicate the areas with the most intensive seismicity, this is consistent with the tectonic states and estimated stress distribution in the associated areas. The comparison between DECODE and other state-of-the-art methods, such as DBSCAN, OPTICS and Wavelet Cluster, illustrates the contribution of our approach: although DECODE can be computationally expensive, it is capable of identifying the number of point processes and simultaneously estimating the classification thresholds with little prior knowledge. <s> BIB003 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Introduction <s> Understand the need for analyses of large, complex, information-rich data sets. Identify the goals and primary tasks of the data-mining process. Describe the roots of data-mining technology. Recognize the iterative character of a data-mining process and specify its basic steps. Explain the influence of data quality on a data-mining process. Establish the relation between data warehousing and data mining. Data mining is an iterative process within which progress is defined by discovery, through either automatic or manual methods. Data mining is most useful in an exploratory analysis scenario in which there are no predetermined notions about what will constitute an "interesting" outcome. Data mining is the search for new, valuable, and nontrivial information in large volumes of data. It is a cooperative effort of humans and computers. Best results are achieved by balancing the knowledge of human experts in describing problems and goals with the search capabilities of computers. In practice, the two primary goals of data mining tend to be prediction and description. Prediction involves using some variables or fields in the data set to predict unknown or future values of other variables of interest. Description, on the other hand, focuses on finding patterns describing the data that can be interpreted by humans. Therefore, it is possible to put data-mining activities into one of two categories: Predictive data mining, which produces the model of the system described by the given data set, or Descriptive data mining, which produces new, nontrivial information based on the available data set. <s> BIB004 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Introduction <s> The advances in location-acquisition and mobile computing techniques have generated massive spatial trajectory data, which represent the mobility of a diversity of moving objects, such as people, vehicles, and animals. Many techniques have been proposed for processing, managing, and mining trajectory data in the past decade, fostering a broad range of applications. In this article, we conduct a systematic survey on the major research into trajectory data mining, providing a panorama of the field as well as the scope of its research topics. Following a road map from the derivation of trajectory data, to trajectory data preprocessing, to trajectory data management, and to a variety of mining tasks (such as trajectory pattern mining, outlier detection, and trajectory classification), the survey explores the connections, correlations, and differences among these existing techniques. This survey also introduces the methods that transform trajectories into other data formats, such as graphs, matrices, and tensors, to which more data mining and machine learning techniques can be applied. Finally, some public trajectory datasets are presented. This survey can help shape the field of trajectory data mining, providing a quick understanding of this field to the community. <s> BIB005 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Introduction <s> As the number, volume and resolution of spatio-temporal datasets increases, traditional statistical methods for dealing with such data are becoming overwhelmed. Nevertheless, the spatio-temporal data are rich sources of information and knowledge, waiting to be discovered. The field of spatio-temporal data mining (STDM) emerged out of a need to create effective and efficient techniques in order to turn the massive data into meaningful information and knowledge. This chapter reviews the state of the art in STDM research and applications, with emphasis placed on three key areas, including spatio-temporal prediction and forecasting, spatio-temporal clustering and spatio-temporal visualization. The future direction and research challenges of STDM are discussed at the end of this chapter. <s> BIB006 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Introduction <s> Large volumes of spatio-temporal data are increasingly collected and studied in diverse domains, including climate science, social sciences, neuroscience, epidemiology, transportation, mobile health, and Earth sciences. Spatio-temporal data differ from relational data for which computational approaches are developed in the data-mining community for multiple decades in that both spatial and temporal attributes are available in addition to the actual measurements/attributes. The presence of these attributes introduces additional challenges that needs to be dealt with. Approaches for mining spatio-temporal data have been studied for over a decade in the data-mining community. In this article, we present a broad survey of this relatively young field of spatio-temporal data mining. We discuss different types of spatio-temporal data and the relevant data-mining questions that arise in the context of analyzing each of these datasets. Based on the nature of the data-mining problem studied, we classify literature on spatio-temporal data mining into six major categories: clustering, predictive learning, change detection, frequent pattern mining, anomaly detection, and relationship mining. We discuss the various forms of spatio-temporal data-mining problems in each of these categories. <s> BIB007

Large-scale data mining brings new opportunities and challenges for discovering hidden valuable information from enormous data sets. In particular, with the rapid development of positioning technologies as well as the emergence of a large number of positioning devices, a vast amount of data could be easily collected from different sources. These sources could come from broad domains, including government documentary and decades of collected data, transportation BIB005 , and social media [2] . For example, governments conduct censuses and own large datasets containing information about population change, human movement, and economic characteristics during different periods for planning and policy making. Many floating cars such as taxi and truck installing GPS receivers can monitor running state and record spatial and temporal information every second. Social media like Facebook and Twitter can post users' experiences at a given place and time. All this spatiotemporal information is useful for pattern analysis in space and time. Space can be represented by an address, geographical coordinates of latitudes and longitude, or local (X, Y) coordinates. Time can be shown by year, month, and day and sometimes as detailed as hour, minute, or second. Spatiotemporal (ST) data types can be divided into five categories containing events, geo-referenced variables, geo-referenced time series, moving points, and trajectories [3] (Figure 1 ). The collected datasets, regardless of if they are in tabular or graphical forms, are often too complex to be understood. An efficient spatiotemporal analysis method is important to mine meaningful patterns for better understanding or visualization . A good approach is to put data with similar characteristics together to find interesting and useful features. Clustering is one popular unsupervised method for discovering potential patterns and is widely used in data analysis, especially for geographical data . It aims to group events according to neighboring occurrence and/or similar attributes. Most clustering algorithms should measure the distance between each pair. Various distance functions are adopted in the clustering methods, such as the Euclidean and Manhattan distance functions. A famous application of clustering occurred in 1854, when Dr John Snow found that clusters of cholera cases occurred around a public water pump, which was the source of the spread of cholera. Clustering is a high-performance tool for detecting hot spot patterns in spatial/ST data analysis BIB004 . ST data analysis methods can be classified into six categories-clustering, prediction, change detection, frequent pattern mining, anomaly detection, and relationship mining BIB007 . Clustering has been used in many applications BIB006 . In some cases, spatiotemporal clustering methods are not all that different from two-dimensional spatial clustering BIB002 BIB003 BIB001 . Figure 2 shows the procedure of clustering. For raw spatiotemporal data, the first step is cleaning and reorganization. Incorrect and missing data should be identified and deleted before applying an appropriate clustering algorithm. However, different parameters can affect the clustering results. It is necessary to adjust parameters for a better understanding of cluster results and interpreting potential information. A good approach is to put data with similar characteristics together to find interesting and useful features. Clustering is one popular unsupervised method for discovering potential patterns and is widely used in data analysis, especially for geographical data . It aims to group events according to neighboring occurrence and/or similar attributes. Most clustering algorithms should measure the distance between each pair. Various distance functions are adopted in the clustering methods, such as the Euclidean and Manhattan distance functions. A famous application of clustering occurred in 1854, when Dr John Snow found that clusters of cholera cases occurred around a public water pump, which was the source of the spread of cholera. Clustering is a high-performance tool for detecting hot spot patterns in spatial/ST data analysis BIB004 . ST data analysis methods can be classified into six categories-clustering, prediction, change detection, frequent pattern mining, anomaly detection, and relationship mining BIB007 . Clustering has been used in many applications BIB006 . In some cases, spatiotemporal clustering methods are not all that different from two-dimensional spatial clustering BIB002 BIB003 BIB001 . Figure 2 shows the procedure of clustering. For raw spatiotemporal data, the first step is cleaning and reorganization. Incorrect and missing data should be identified and deleted before applying an appropriate clustering algorithm. However, different parameters can affect the clustering results. It is necessary to adjust parameters for a better understanding of cluster results and interpreting potential information. There are still many challenges for extracting useful ST patterns due to complex data types. Many methods simply treat the temporal dimension of spatiotemporal data as an additional dimension. With different units of time and space, clustering results could have big differences when considering the scale of time. Multiple scales effect is another challenge as the clustering results depend on the various spatial and temporal scales. Different space regions and temporal periods could form distinguished clusters. In this paper, we only focus on the clustering methods of the events ST data type. In our view, these could be divided into two categories, the hypothesis testing-based methods and the partition-based methods. The former one mainly uses a probability model and statistical hypothesis testing to find significant clusters. In general, the null hypothesis is that the distribution of events is random; if it is rejected, a cluster could be formed. The partitional clustering methods mostly utilize distance functions to compute the closeness of events to distinguish cluster and noise. Some popular There are still many challenges for extracting useful ST patterns due to complex data types. Many methods simply treat the temporal dimension of spatiotemporal data as an additional dimension. With different units of time and space, clustering results could have big differences when considering the scale of time. Multiple scales effect is another challenge as the clustering results depend on the various spatial and temporal scales. Different space regions and temporal periods could form distinguished clusters. In this paper, we only focus on the clustering methods of the events ST data type. In our view, these could be divided into two categories, the hypothesis testing-based methods and the partitionbased methods. The former one mainly uses a probability model and statistical hypothesis testing to find significant clusters. In general, the null hypothesis is that the distribution of events is random; if it is rejected, a cluster could be formed. The partitional clustering methods mostly utilize distance functions to compute the closeness of events to distinguish cluster and noise. Some popular spatiotemporal clustering methods are introduced in the following sections. This will help to understand the evolution of techniques in the past decades and explore future research trends.

Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> 1. Introduction. 2. Partitioning Around Medoids (Program PAM). 3. Clustering large Applications (Program CLARA). 4. Fuzzy Analysis. 5. Agglomerative Nesting (Program AGNES). 6. Divisive Analysis (Program DIANA). 7. Monothetic Analysis (Program MONA). Appendix 1. Implementation and Structure of the Programs. Appendix 2. Running the Programs. Appendix 3. Adapting the Programs to Your Needs. Appendix 4. The Program CLUSPLOT. References. Author Index. Subject Index. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Finding useful patterns in large datasets has attracted considerable interest recently, and one of the most widely studied problems in this area is the identification of clusters, or densely populated regions, in a multi-dimensional dataset. Prior work does not adequately address the problem of large datasets and minimization of I/O costs.This paper presents a data clustering method named BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies), and demonstrates that it is especially suitable for very large databases. BIRCH incrementally and dynamically clusters incoming multi-dimensional metric data points to try to produce the best quality clustering with the available resources (i.e., available memory and time constraints). BIRCH can typically find a good clustering with a single scan of the data, and improve the quality further with a few additional scans. BIRCH is also the first clustering algorithm proposed in the database area to handle "noise" (data points that are not part of the underlying pattern) effectively.We evaluate BIRCH's time/space efficiency, data input order sensitivity, and clustering quality through several experiments. We also present a performance comparisons of BIRCH versus CLARANS, a clustering method proposed recently for large datasets, and show that BIRCH is consistently superior. <s> BIB002 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Clustering algorithms are attractive for the task of class identification in spatial databases. However, the application to large spatial databases rises the following requirements for clustering algorithms: minimal requirements of domain knowledge to determine the input parameters, discovery of clusters with arbitrary shape and good efficiency on large databases. The well-known clustering algorithms offer no solution to the combination of these requirements. In this paper, we present the new clustering algorithm DBSCAN relying on a density-based notion of clusters which is designed to discover clusters of arbitrary shape. DBSCAN requires only one input parameter and supports the user in determining an appropriate value for it. We performed an experimental evaluation of the effectiveness and efficiency of DBSCAN using synthetic data and real data of the SEQUOIA 2000 benchmark. The results of our experiments demonstrate that (1) DBSCAN is significantly more effective in discovering clusters of arbitrary shape than the well-known algorithm CLAR-ANS, and that (2) DBSCAN outperforms CLARANS by a factor of more than 100 in terms of efficiency. <s> BIB003 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Spatial data mining, i.e., discovery of interesting characteristics and patterns that may implicitly exist in spatial databases, is a challenging task due to the huge amounts of spatial data and to the new conceptual nature of the problems which must account for spatial distance. Clustering and region oriented queries are common problems in this domain. Several approaches have been presented in recent years, all of which require at least one scan of all individual objects (points). Consequently, the computational complexity is at least linearly proportional to the number of objects to answer each query. In this paper, we propose a hierarchical statistical information grid based approach for spatial data mining to reduce the cost further. The idea is to capture statistical information associated with spatial cells in such a manner that whole classes of queries and clustering problems can be answered without recourse to the individual objects. In theory, and confirmed by empirical studies, this approach outperforms the best previous method by at least an order of magnitude, especially when the data set is very large. <s> BIB004 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Clustering, in data mining, is useful for discovering groups and identifying interesting distributions in the underlying data. Traditional clustering algorithms either favor clusters with spherical shapes and similar sizes, or are very fragile in the presence of outliers. We propose a new clustering algorithm called CURE that is more robust to outliers, and identifies clusters having non-spherical shapes and wide variances in size. CURE achieves this by representing each cluster by a certain fixed number of points that are generated by selecting well scattered points from the cluster and then shrinking them toward the center of the cluster by a specified fraction. Having more than one representative point per cluster allows CURE to adjust well to the geometry of non-spherical shapes and the shrinking helps to dampen the effects of outliers. To handle large databases, CURE employs a combination of random sampling and partitioning . A random sample drawn from the data set is first partitioned and each partition is partially clustered. The partial clusters are then clustered in a second pass to yield the desired clusters. Our experimental results confirm that the quality of clusters produced by CURE is much better than those found by existing algorithms. Furthermore, they demonstrate that random sampling and partitioning enable CURE to not only outperform existing algorithms but also to scale well for large databases without sacrificing clustering quality. <s> BIB005 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Many applications require the management of spatial data. Clustering large spatial databases is an important problem which tries to find the densely populated regions in the feature space to be used in data mining, knowledge discovery, or efficient information retrieval. A good clustering approach should be efficient and detect clusters of arbitrary shape. It must be insensitive to the outliers (noise) and the order of input data. We propose WaveCluster, a novel clustering approach based on wavelet transforms, which satisfies all the above requirements. Using multiresolution property of wavelet transforms, we can effectively identify arbitrary shape clusters at different degrees of accuracy. We also demonstrate that WaveCluster is highly efficient in terms of time complexity. Experimental results on very large data sets are presented which show the efficiency and effectiveness of the proposed approach compared to the other recent clustering methods. This research is supported by Xerox Corporation. Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the VLDB copyright notice and the title of the publication and its date appear, and notice is given that copying is by permission of the Very Large Data Base Endowment. To copy otherwise, or to republish, requires a fee and/or special permission from the Endowment. Proceedings of the 24th VLDB Conference New York, USA, 1998 <s> BIB006 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Cluster analysis is a primary method for database mining. It is either used as a stand-alone tool to get insight into the distribution of a data set, e.g. to focus further analysis and data processing, or as a preprocessing step for other algorithms operating on the detected clusters. Almost all of the well-known clustering algorithms require input parameters which are hard to determine but have a significant influence on the clustering result. Furthermore, for many real-data sets there does not even exist a global parameter setting for which the result of the clustering algorithm describes the intrinsic clustering structure accurately. We introduce a new algorithm for the purpose of cluster analysis which does not produce a clustering of a data set explicitly; but instead creates an augmented ordering of the database representing its density-based clustering structure. This cluster-ordering contains information which is equivalent to the density-based clusterings corresponding to a broad range of parameter settings. It is a versatile basis for both automatic and interactive cluster analysis. We show how to automatically and efficiently extract not only 'traditional' clustering information (e.g. representative points, arbitrary shaped clusters), but also the intrinsic clustering structure. For medium sized data sets, the cluster-ordering can be represented graphically and for very large data sets, we introduce an appropriate visualization technique. Both are suitable for interactive exploration of the intrinsic clustering structure offering additional insights into the distribution and correlation of the data. <s> BIB007 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> The self-organizing map (SOM) is an excellent tool in exploratory phase of data mining. It projects input space on prototypes of a low-dimensional regular grid that can be effectively utilized to visualize and explore properties of the data. When the number of SOM units is large, to facilitate quantitative analysis of the map and the data, similar units need to be grouped, i.e., clustered. In this paper, different approaches to clustering of the SOM are considered. In particular, the use of hierarchical agglomerative clustering and partitive clustering using K-means are investigated. The two-stage procedure-first using SOM to produce the prototypes that are then clustered in the second stage-is found to perform well when compared with direct clustering of the data and to reduce the computation time. <s> BIB008 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Conceptual clustering is an important way of summarizing and explaining data. However, the recent formulation of this paradigm has allowed little exploration of conceptual clustering as a means of improving performance. Furthermore, previous work in conceptual clustering has not explicitly dealt with constraints imposed by real world environments. This article presents COBWEB, a conceptual clustering system that organizes data so as to maximize inference ability. Additionally, COBWEB is incremental and computationally economical, and thus can be flexibly applied in a variety of domains. <s> BIB009 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Spatial data mining is the discovery of interesting relationships and characteristics that may exist implicitly in spatial databases. To this end, this paper has three main contributions. First, it proposes a new clustering method called CLARANS, whose aim is to identify spatial structures that may be present in the data. Experimental results indicate that, when compared with existing clustering methods, CLARANS is very efficient and effective. Second, the paper investigates how CLARANS can handle not only point objects, but also polygon objects efficiently. One of the methods considered, called the IR-approximation, is very efficient in clustering convex and nonconvex polygon objects. Third, building on top of CLARANS, the paper develops two spatial data mining algorithms that aim to discover relationships between spatial and nonspatial attributes. Both algorithms can discover knowledge that is difficult to find with existing spatial data mining algorithms. <s> BIB010 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Clustering analysis is a primary method for data mining. Density clustering has such advantages as: its clusters are easy to understand and it does not limit itself to shapes of clusters. But existing density-based algorithms have trouble in finding out all the meaningful clusters for datasets with varied densities. This paper introduces a new algorithm called VDBSCAN for the purpose of varied-density datasets analysis. The basic idea of VDBSCAN is that, before adopting traditional DBSCAN algorithm, some methods are used to select several values of parameter Eps for different densities according to a k-dist plot. With different values of Eps, it is possible to find out clusters with varied densities simultaneity. For each value of Eps, DBSCAN algorithm is adopted in order to make sure that all the clusters with respect to corresponding density are clustered. And for the next process, the points that have been clustered are ignored, which avoids marking both denser areas and sparser ones as one cluster. Finally, a synthetic database with 2-dimension data is used for demonstration, and experiments show that VDBSCAN is efficient in successfully clustering uneven datasets. <s> BIB011 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Understand the need for analyses of large, complex, information-rich data sets. Identify the goals and primary tasks of the data-mining process. Describe the roots of data-mining technology. Recognize the iterative character of a data-mining process and specify its basic steps. Explain the influence of data quality on a data-mining process. Establish the relation between data warehousing and data mining. Data mining is an iterative process within which progress is defined by discovery, through either automatic or manual methods. Data mining is most useful in an exploratory analysis scenario in which there are no predetermined notions about what will constitute an "interesting" outcome. Data mining is the search for new, valuable, and nontrivial information in large volumes of data. It is a cooperative effort of humans and computers. Best results are achieved by balancing the knowledge of human experts in describing problems and goals with the search capabilities of computers. In practice, the two primary goals of data mining tend to be prediction and description. Prediction involves using some variables or fields in the data set to predict unknown or future values of other variables of interest. Description, on the other hand, focuses on finding patterns describing the data that can be interpreted by humans. Therefore, it is possible to put data-mining activities into one of two categories: Predictive data mining, which produces the model of the system described by the given data set, or Descriptive data mining, which produces new, nontrivial information based on the available data set. <s> BIB012 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> As a major analytical method for outbreak detection, Kulldorff's space-time scan statistic (2001, Journal of the Royal Statistical Society, Series A 164, 61-72) has been implemented in many syndromic surveillance systems. Since, however, it is based on circular windows in space, it has difficulty correctly detecting actual noncircular clusters. Takahashi et al. (2008, International Journal of Health Geographics 7, 14) proposed a flexible space-time scan statistic with the capability of detecting noncircular areas. It seems to us, however, that the detection of the most likely cluster defined in these space-time scan statistics is not the same as the detection of localized emerging disease outbreaks because the former compares the observed number of cases with the conditional expected number of cases. In this article, we propose a new space-time scan statistic which compares the observed number of cases with the unconditional expected number of cases, takes a time-to-time variation of Poisson mean into account, and implements an outbreak model to capture localized emerging disease outbreaks more timely and correctly. The proposed models are illustrated with data from weekly surveillance of the number of absentees in primary schools in Kitakyushu-shi, Japan, 2006. <s> BIB013 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Density Density-based clustering methods are an important category of clustering methods that are able to identify areas with dense clusters of any shape and size. One of the basic and simple methods in this group is DBSCAN. This algorithm clusters dataset based on two received parameters from the user. one of the disadvantages of DBSCAN is its inability in identifing clusters with different densities in a dataset. In this paper, we propose a DBSCAN-based method to cover multi-density datasets, called EXDBSCAN. This method only get one parameter from the user and in addition of detecting clusters with different densities, can detect outlier correctly. The results of comparing final clusters of our method with two other clustering methods on some multi-density data sets shows our method's performance in such datasets. <s> BIB014 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> This paper presents a comprehensive study on clustering: exiting methods and developments made at various times. Clustering is defined as an unsupervised learning where the objects are grouped on the basis of some similarity inherent among them. There are different methods for clustering the objects such as hierarchical, partitional, grid, density based and model based. The approaches used in these methods are discussed with their respective states of art and applicability. The measures of similarity as well as the evaluation criteria, which are the central components of clustering, are also presented in the paper. The applications of clustering in some fields like image segmentation, object and character recognition and data mining are highlighted. <s> BIB015 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> ABSTRACTWe examined three different ways to integrate spatial and temporal data in kernel density estimation methods (KDE) to identify space–time clusters of geographic events. Spatial data and time data are typically measured in different units along respective dimensions. Therefore, spatial KDE methods require special extensions when incorporating temporal data to detect spatiotemporal clusters of geographical event. In addition to a real-world data set, we applied the proposed methods to simulated data that were generated through random and normal processes to compare results of different kernel functions. The comparison is based on hit rates and values of a compactness index with considerations of both spatial and temporal attributes of the data. The results show that the spatiotemporal KDE (STKDE) can reach higher hit rates while keeping identified hotspots compact. The implementation of these STKDE methods is tested using the 2012 crime event data in Akron, Ohio, as an example. The results show that... <s> BIB016 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> From Spatial to ST Clustering <s> Abstract Predictive hotspot mapping plays a critical role in hotspot policing. Existing methods such as the popular kernel density estimation (KDE) do not consider the temporal dimension of crime. Building upon recent works in related fields, this article proposes a spatio-temporal framework for predictive hotspot mapping and evaluation. Comparing to existing work in this scope, the proposed framework has four major features: (1) a spatio-temporal kernel density estimation (STKDE) method is applied to include the temporal component in predictive hotspot mapping, (2) a data-driven optimization technique, the likelihood cross-validation, is used to select the most appropriate bandwidths, (3) a statistical significance test is designed to filter out false positives in the density estimates, and (4) a new metric, the predictive accuracy index (PAI) curve, is proposed to evaluate predictive hotspots at multiple areal scales. The framework is illustrated in a case study of residential burglaries in Baton Rouge, Louisiana in 2011, and the results validate its utility. <s> BIB017

In this section, we will first briefly discuss spatial clustering. There is no clear definition of clustering BIB015 and different categories have overlap such that an algorithm could contain more than one feature of categories. Therefore, a lot of methods have been proposed according to diverse principles. Han, et al. BIB012 divided major clustering methods into four categories. which were partitioning methods, hierarchical methods, grid-based methods, and density-based methods. Partitioning methods divide the entire dataset into several groups. For example, K-means is the most popular clustering method in the partitioning methods. It is an iteration process to find the cluster and its center. Based on this theory, Kaufman and Rousseeuw BIB001 proposed partitioning around medoids (PAM) and clustering large application (CLARA) to improve the efficiency of clustering. Ng and Han BIB010 proposed clustering large applications based upon randomized search (CLARANS) to investigate not only detect points but also polygon objects. Hierarchical methods can separate a dataset into multiple levels based on distance or density functions. For example, Balanced iterative reducing and clustering using hierarchies (BIRCH) use a tree structure to form clusters with speed and efficiency BIB002 . Chameleon finds the clusters by measuring the similarity of data and grouping them . Clustering using Representatives (CURE) BIB005 can identify non-spherical shapes of clusters within a large database. Density-based methods have the ability to discover different shapes of clusters. For example, density based spatial clustering of applications with noise(DBSCAN) BIB003 is a well-known algorithm for detecting an arbitrary shape of clusters, and many people have proposed improved methods to overcome any drawbacks and promote efficiency BIB014 BIB011 . DBSCAN is sensitive In this section, we will first briefly discuss spatial clustering. There is no clear definition of clustering BIB015 and different categories have overlap such that an algorithm could contain more than one feature of categories. Therefore, a lot of methods have been proposed according to diverse principles. Han, et al. BIB012 divided major clustering methods into four categories, which were partitioning methods, hierarchical methods, grid-based methods, and density-based methods. Partitioning methods divide the entire dataset into several groups. For example, K-means is the most popular clustering method in the partitioning methods. It is an iteration process to find the cluster and its center. Based on this theory, Kaufman and Rousseeuw BIB001 proposed partitioning around medoids (PAM) and clustering large application (CLARA) to improve the efficiency of clustering. Ng and Han BIB010 proposed clustering large applications based upon randomized search (CLARANS) to investigate not only detect points but also polygon objects. Hierarchical methods can separate a dataset into multiple levels based on distance or density functions. For example, Balanced iterative reducing and clustering using hierarchies (BIRCH) use a tree structure to form clusters with speed and efficiency BIB002 . Chameleon finds the clusters by measuring the similarity of data and grouping them . Clustering using Representatives (CURE) BIB005 can identify non-spherical shapes of clusters within a large database. Density-based methods have the ability to discover different shapes of clusters. For example, density based spatial clustering of applications with noise(DBSCAN) BIB003 is a well-known algorithm for detecting an arbitrary shape of clusters, and many people have proposed improved methods to overcome any drawbacks and promote efficiency BIB014 BIB011 . DBSCAN is sensitive to input parameters, however, ordering points to identify the clustering structure (OPTICS) BIB007 could prevent this problem from affecting the clustering results. However, it cannot get accurate cluster results. A method called DENCLUE (Density-based clustering) uses a kernel density estimation model to identify the high density of clusters with an arbitrary shape. Grid-based methods build a grid structure for storing the dataset and each grid is the basic unit to form a cluster BIB006 BIB004 . Asides from the four categories, many other methods have also been proposed, such as model based methods BIB009 BIB008 . The major difference between spatial and ST clustering is the 'time' element, which is treated as either another dimension or an attribute. By space, it can be at least 2-dimensional (X,Y) or 3-dimensional (X,Y,Z) in which events or attributes are clustered. Most socio-economic information, such as population and traffic, is considered as variations in 2-dimensions [(X,Y) + attribute] only; whereas natural phenomena, such as temperature and pressure, vary with space and height [(X,Y,Z) + attribute]. When 'time' is added, it may be treated as merely an attribute to 2-dimensional or 3-dimensional space, for example, a date when a certain event occurs or a record is created; but this does not allow clustering in terms of time. An alternative common method is to model 'time' as a third dimension in addition to the 2-dimensional [(X,Y, T) + attribute] space. Therefore, some ST clustering methods have been developed from spatial clustering methods BIB017 BIB016 BIB013 . The addition of a time dimension to the 3-dimensional [(X,Y, T, Z) + attribute] space is still a challenging issue to model and to visualize. There is a need in many applications to integrate spatial and temporal information together for more detailed and accurate analyses. For example, in the study of human mobility, there is a need to identify at what time and where people cluster instead of just relying on census data or a generalized pattern of population distribution. This applies in the same way to crime patterns, traffic patterns etc. In the following sections, we will discuss the different categories of ST clustering.

Spatiotemporal Data Clustering: A Survey of Methods <s> Hypothesis testing-based methods <s> Motivated by the current debate on possible raised incidence of certain types of cancers near nuclear installations, this paper develops a methodology for fitting a class of inhomogeneous Poisson point process models to data consisting of the locations of all occurrences of some phenomenon of interest within a designated planar region. The model is based on a multiplicative decomposition of the intensity function, with separate terms to describe natural spatial variation in intensity and possible raised incidence around a prespecified point. A nonparametric kernel smoothing approach, based on data from a related phenomenon, is used to describe natural spatial variation, while a parametric maximum likelihood approach is used to describe raised incidence near the prespecified point. The methodology is applied to data on the spatial distribution of cancers of the larynx and of the lung in the Chorley‐Ribble area of Lancashire, England. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Hypothesis testing-based methods <s> In spatial analysis buffer impact areas are called hotspots and are determined by means of density clustering methods. In a previous work, we found these hotspots in the context of a Geographic Information System (GIS) by using the extended fuzzy C-means (EFCM). Here we show how the spatial distribution of the hotspots can evolve temporally and like applicational example, we present the spatial-temporal evolution in the period 2000-2006 of the fire point-events data of the Santa Fe district (NM) (downloaded from URL: www.fs.fed.us/r3/gis/sfe_gis.shtml). <s> BIB002

In the field of statistics, some existing fundamental research has been studied , including ST point pattern detection and analysis BIB001 . Hypothesis testing is used to determine the probability of a given hypothesis being true or not. The advantage of this method is it considers space and time information together. It is a new research direction that could allow some traditional spatial statistics to be extended for ST data analysis. For example, Di Martino and Sessa BIB002 proposed an extended algorithm of fuzzy c-means to find circular clusters from ST data. This method could reduce the noise and outliers influencing clustering results. Detailed processes of some famous algorithms are described below.

Spatiotemporal Data Clustering: A Survey of Methods <s> Space-time interaction methods <s> The "Encyclopaedia Americana" defines an epi demic as a "sudden excessive prevalence of a disease in a population". This is a typical definition, and although suddenness is not always considered an essential feature of an epidemic, most would agree that it refers to disease, and that it denotes an in creased prevalence. But there are two serious difficul ties in this concept which limit its usefulness and unless they are recognized they may prove a serious hindrance to the detection of epidemic properties in several classes of disease. The first is found when the occurrences are few in number, sparsely scattered in time, and widely sep arated in relationship to the frequency of any supposed fluctuations. This problem arises with uncommon illnesses such as congenital malformations when an apparent crop of cases poses the question how many cases can make an epidemic. The true difficulty seems to be the breakdown of the concept of prevalence upon which in turn the concept of an epidemic depends. The situation is analogous to the breakdown of the gas laws when very few molecules are involved and the inappropriateness there of the ordinary con cept of pressure; or the breakdown of the theory of wave propagation in the study of low intensity radia tion. Pressure, wave motion, prevalence, and epide micity, are terms appropriate to the study of high concentrations and when we are compelled to deal with low intensities special methods of examination must be devised. The second difficulty arises from the assumption implicit in the above definition, that an epidemic is recognized on a time base. However, it is clear that space must sometimes be taken into account as well as time. A feature of many epidemics is that the geographical region of maximum prevalence moves with change in time, or stated conversely, that the time of maximum prevalence changes with change of place. The result is that a disease which is manifestly epidemic on the time scale in each of a number of small regions, may over a sum of those areas show only equivocal evidence of epidemicity in the terms defined above. Conversely, a disease of which the incidence is geographically uneven during a short period may appear more evenly distributed when a longer period is considered. These difficulties are not serious with a frequently occurring disease; either the time scale or the geo graphical area may be divided into sufficiently small classes to demonstrate variations of prevalence in terms of the other. But with uncommon diseases such as leukaemia or congenital malformations the pro cedure of subdivision leads back into the problems of small numbers. It may be impossible to examine either the space or the time variations in prevalence, without at the same time impairing the applicability of the concept of prevalence. This may not be for lack of large numbers of events for analysis, but because they occur sparsely over a large area and over a long period of time, and accumulation of further data may do little towards solving the problem. For these reasons, no doubt, most of the statements to date of low intensity space-time clustering in particular disea ses have been based upon impressions and anecdotes rather than upon rigorous analysis. A previous attempt (Knox, 1959) has been made to analyse low intensity data on time base alone and the present paper is an attempt to extend this to space and time jointly. The two purposes of the present study are to suggest a method of analysing this class of data in a more discriminating manner than has so far been employed and to employ the technique in the investigation of a particular problem. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Space-time interaction methods <s> The problem of identifying subtle time-space clustering of disease, as may be occurring in leukemia, is described and reviewed. Published approaches, generally associated with studies of leukemia, not dependent on knowledge of the underlying population for their validity, are directed towards identifying clustering by establishing a relationship between the temporal and the spatial separations for the n ( n - 1)/2 possible pairs which can be formed from the n observed cases of disease. Here it is proposed that statistical power can be improved by applying a reciprocal transform to these separations. While a permutational approach can give valid probability levels for any observed association, for reasons of practicability, it is suggested that the observed association be tested relative to its permutational variance. Formulas and computational procedures for doing so are given. ::: ::: While the distance measures between points represent symmetric relationships subject to mathematical and geometric regularities, the variance formula developed is appropriate for arbitrary relationships. Simplified procedures are given for the case of symmetric and skew-symmetric relationships. The general procedure is indicated as being potentially useful in other situations as, for example, the study of interpersonal relationships. Viewing the procedure as a regression approach, the possibility for extending it to nonlinear and multivariate situations is suggested. ::: ::: Other aspects of the problem and of the procedure developed are discussed. ::: ::: Similarly, pure temporal clustering can be identified by a study of incidence rates in periods of widespread epidemics. In point of fact, many epidemics of communicable diseases are somewhat local in nature and so these do actually constitute temporal-spatial clusters. For leukemia and similar diseases in which cases seem to arise substantially at random rather than as clear-cut epidemics, it is necessary to devise sensitive and efficient procedures for detecting any nonrandom component of disease occurrence. ::: ::: Various ingenious procedures which statisticians have developed for the detection of disease clustering are reviewed here. These procedures can be generalized so as to increase their statistical validity and efficiency. The technic to be given below for imparting statistical validity to the procedures already in vogue can be viewed as a generalized form of regression with possible useful application to problems arising in quite different contexts. <s> BIB002 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Space-time interaction methods <s> The Knox method, as well as other tests for space-time interaction, are biased when there are geographical population shifts, i.e., when there are different percent population growths in different regions. In this paper, the size of the population shift bias is investigated for the Knox test, and it is shown that it can be a considerable problem. A Monte Carlo method for constructing unbiased space-time interaction tests is then presented and illustrated on the Knox test as well as for a combined Knox test. Practical implications are discussed in terms of the interpretation of past results and the design of future studies. <s> BIB003

A number of methods have been explored for detecting ST clustering. The core essence of a cluster is that objects should be close to each other in the space or time dimension. Knox and Bartlett proposed a test to quantify a space and time interaction of disease. Low-intensity disease detection by joining space and time analysis was conducted in Reference BIB001 . Improvements to existing drawbacks were proposed by others BIB003 . In this method, critical space distance α and time distance β should be manually defined first. Pairs of cases less than the critical space distance and time distance separately were regarded as near in space and time. The test statistics equation was: where K was the total number of paired cases smaller than the critical space and time distance, N was the total number of data. d ij was space adjacency, if the distance between i and j was less than α, it was equal to 1, otherwise equal to 0. t ij was time adjacency, if the distance between i and j was less than β, it was equal to 1, otherwise equal to 0. The Monte Carlo method was used for the significant test of K and a predefined number of runs was identified. The probability value of K being larger than the test statistic should belong to right hand tail of null distribution. The disadvantage of this method was critical space and time distances values may be assigned subjectively. A modification was proposed by Mantel BIB002 who multiplied the sum of time distances by the sum of spatial distances. The test statistic of Mantel's test was similar to Knox's test. It focused on the problem of selecting the critical distances of Knox's test. It is based on a simple cross-product term: where d s ij is the distance between data i and j in space. d t ij is the distance between data i and j in time. Then, it is normalized: where M is the standardized Mantel statistic and N is the number of data. d s ij is the distance between data i and j in space. d t ij is the distance between data i and j in time. d s is the average distance of all data in space. d t is the average distance of all data in time. s s and s t are the standard deviations of data in space and time, respectively. This equation allowed for different units of space and time in the same framework, and multiple scale problems could be solved by limiting the range of correlation coefficient values into [−1,1].

Spatiotemporal Data Clustering: A Survey of Methods <s> Scan Statistics <s> The scan statistic is commonly used to test if a one dimensional point process is purely random, or if any clusters can be detected. Here it is simultaneously extended in three directions:(i) a spatial scan statistic for the detection of clusters in a multi-dimensional point process is proposed, (ii) the area of the scanning window is allowed to vary, and (iii) the baseline process may be any inhomogeneous Poisson process or Bernoulli process with intensity pro-portional to some known function. The main interest is in detecting clusters not explained by the baseline process. These methods are illustrated on an epidemiological data set, but there are other potential areas of application as well. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Scan Statistics <s> Most disease registries are updated at least yearly. If a geographically localized health hazard suddenly occurs, we would like to have a surveillance system in place that can pick up a new geographical disease cluster as quickly as possible, irrespective of its location and size. At the same time, we want to minimize the number of false alarms. By using a space–time scan statistic, we propose and illustrate a system for regular time periodic disease surveillance to detect any currently ‘active’ geographical clusters of disease and which tests the statistical significance of such clusters adjusting for the multitude of possible geographical locations and sizes, time intervals and time periodic analyses. The method is illustrated on thyroid cancer among men in New Mexico 1973–1992. <s> BIB002 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Scan Statistics <s> BACKGROUND ::: The ability to detect disease outbreaks early is important in order to minimize morbidity and mortality through timely implementation of disease prevention and control measures. Many national, state, and local health departments are launching disease surveillance systems with daily analyses of hospital emergency department visits, ambulance dispatch calls, or pharmacy sales for which population-at-risk information is unavailable or irrelevant. ::: ::: ::: METHODS AND FINDINGS ::: We propose a prospective space-time permutation scan statistic for the early detection of disease outbreaks that uses only case numbers, with no need for population-at-risk data. It makes minimal assumptions about the time, geographical location, or size of the outbreak, and it adjusts for natural purely spatial and purely temporal variation. The new method was evaluated using daily analyses of hospital emergency department visits in New York City. Four of the five strongest signals were likely local precursors to citywide outbreaks due to rotavirus, norovirus, and influenza. The number of false signals was at most modest. ::: ::: ::: CONCLUSION ::: If such results hold up over longer study times and in other locations, the space-time permutation scan statistic will be an important tool for local and national health departments that are setting up early disease detection surveillance systems. <s> BIB003 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Scan Statistics <s> BackgroundThe spatial scan statistic proposed by Kulldorff has been applied to a wide variety of epidemiological studies for cluster detection. This scan statistic, however, uses a circular window to define the potential cluster areas and thus has difficulty in correctly detecting actual noncircular clusters. A recent proposal by Duczmal and Assunção for detecting noncircular clusters is shown to detect a cluster of very irregular shape that is much larger than the true cluster in our experiences.MethodsWe propose a flexibly shaped spatial scan statistic that can detect irregular shaped clusters within relatively small neighborhoods of each region. The performance of the proposed spatial scan statistic is compared to that of Kulldorff's circular spatial scan statistic with Monte Carlo simulation by considering several circular and noncircular hot-spot cluster models. For comparison, we also propose a new bivariate power distribution classified by the number of regions detected as the most likely cluster and the number of hot-spot regions included in the most likely cluster.ResultsThe circular spatial scan statistics shows a high level of accuracy in detecting circular clusters exactly. The proposed spatial scan statistic is shown to have good usual powers plus the ability to detect the noncircular hot-spot clusters more accurately than the circular one.ConclusionThe proposed spatial scan statistic is shown to work well for small to moderate cluster size, up to say 30. For larger cluster sizes, the method is not practically feasible and a more efficient algorithm is needed. <s> BIB004 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Scan Statistics <s> BackgroundEarly detection of disease outbreaks enables public health officials to implement disease control and prevention measures at the earliest possible time. A time periodic geographical disease surveillance system based on a cylindrical space-time scan statistic has been used extensively for disease surveillance along with the SaTScan software. In the purely spatial setting, many different methods have been proposed to detect spatial disease clusters. In particular, some spatial scan statistics are aimed at detecting irregularly shaped clusters which may not be detected by the circular spatial scan statistic.ResultsBased on the flexible purely spatial scan statistic, we propose a flexibly shaped space-time scan statistic for early detection of disease outbreaks. The performance of the proposed space-time scan statistic is compared with that of the cylindrical scan statistic using benchmark data. In order to compare their performances, we have developed a space-time power distribution by extending the purely spatial bivariate power distribution. Daily syndromic surveillance data in Massachusetts, USA, are used to illustrate the proposed test statistic.ConclusionThe flexible space-time scan statistic is well suited for detecting and monitoring disease outbreaks in irregularly shaped areas. <s> BIB005

Scan statistics is a popular method and software can implement scan statistics for detecting clusters. Joseph Naus [41] has been called the father of scan statistics as his method has helped to solve many research problems. The space scan statistic was developed from an original scan statistics method based on the scanning window process BIB001 . A circular scan window with different radii is used to find circular clusters of two-dimensional spatial data with a statistical significance test. An appropriate radius is important to avoid too large or too small clusters, otherwise the results could be meaningless and hard to interpret. Normally, the upper limit of the circle should not include more than 50 percent of all the dataset. Each point could be the center of a circle that contains different numbers of other points. Space and space-time scan statistics have many similar calculation processes. Space-time scan statistics was extended from space scan statistics to detect clusters with the highest likelihood ratio by moving a cylinder as a scan window to scan ST data BIB002 BIB003 . Figure 3 shows the difference between the two methods. The left graph uses space scan statistics to detect clusters, the red center is the core point and the larger circle is the scan window for detection. The right graph uses space-time scan statistics to find clusters, it adopts a red cylinder as the scan window. Space-time scan statistics considers the time dimension and is an extension of space scan statistics in that a three-dimensional cylinder instead of a two-dimensional circle is used. The time interval between events is the height of cylinder. numbers of other points. Space and space-time scan statistics have many similar calculation processes. Space-time scan statistics was extended from space scan statistics to detect clusters with the highest likelihood ratio by moving a cylinder as a scan window to scan ST data BIB002 BIB003 . Figure 3 shows the difference between the two methods. The left graph uses space scan statistics to detect clusters, the red center is the core point and the larger circle is the scan window for detection. The right graph uses space-time scan statistics to find clusters, it adopts a red cylinder as the scan window. Space-time scan statistics considers the time dimension and is an extension of space scan statistics in that a three-dimensional cylinder instead of a two-dimensional circle is used. The time interval between events is the height of cylinder. As with the space scan statistic, the null hypothesis is that the spatiotemporal distribution of events is random. The scan window of the cylinder was changed with different radii and height, looking for the maximum value of log likelihood ratio of all the circles as the cluster region. The formulation was: where was the log likelihood of cylinder, and were the observed and expected number of points, respectively, was the total number of observed points, and was the indicator function. If the left side was larger than right side, was equal to 1, otherwise equal to 0. Many distribution functions could be used, one of which was the Poisson distribution. To obtain the simulated distribution for significance testing of clusters, Monte Carlo replications of data were used to obtain likelihood ratio statistics . It was necessary to obtain values by generating replications such as 999 or even higher to calculate the probability of a random appearance of an observed high-density cluster in a cylindrical window. The likely clusters could be based on the lowest value, which was defined by the cylindrical window. However, similar to space scan, the disadvantage of this method was that it could not discover the arbitrary shape of ST data. To overcome this problem, flexible spatial scan statistic BIB004 and flexibly shaped space-time scan statistic BIB005 were proposed in 2005 and 2008, respectively. FleXScan is the software that was developed to analyze spatial data by using flexible spatial scan statistics. Compare with spatial and space-time scan statistics that can only detect circular or cylinder clusters with variable size, these two methods have the ability to detect noncircular and non-cylinder clusters with high accuracy. For example, Tango and Takahashi BIB004 proposed a flexible spatial scan statistics method that was illustrated using simulated disease maps in the Tokyo Metropolitan area. First, they divided the entire area into many small regions and the location of each region was the administrative population centroid. Next, the set of irregularly shaped windows were consisted K concentric circles and connected regions, where K is a pre-specified As with the space scan statistic, the null hypothesis is that the spatiotemporal distribution of events is random. The scan window of the cylinder was changed with different radii and height, looking for the maximum value of log likelihood ratio of all the circles as the cluster region. The formulation was: where S was the log likelihood of cylinder, n z and u z were the observed and expected number of points, respectively, N was the total number of observed points, and I was the indicator function. If the left side was larger than right side, I was equal to 1, otherwise equal to 0. Many distribution functions could be used, one of which was the Poisson distribution. To obtain the simulated distribution for significance testing of clusters, Monte Carlo replications of data were used to obtain likelihood ratio statistics S. It was necessary to obtain p values by generating replications such as 999 or even higher to calculate the probability of a random appearance of an observed high-density cluster in a cylindrical window. The likely clusters could be based on the lowest p value, which was defined by the cylindrical window. However, similar to space scan, the disadvantage of this method was that it could not discover the arbitrary shape of ST data. To overcome this problem, flexible spatial scan statistic BIB004 and flexibly shaped space-time scan statistic BIB005 were proposed in 2005 and 2008, respectively. FleXScan is the software that was developed to analyze spatial data by using flexible spatial scan statistics. Compare with spatial and space-time scan statistics that can only detect circular or cylinder clusters with variable size, these two methods have the ability to detect non-circular and non-cylinder clusters with high accuracy. For example, Tango and Takahashi BIB004 proposed a flexible spatial scan statistics method that was illustrated using simulated disease maps in the Tokyo Metropolitan area. First, they divided the entire area into many small regions and the location of each region was the administrative population centroid. Next, the set of irregularly shaped windows were consisted K concentric circles and connected regions, where K is a pre-specified maximum length of cluster. The idea was also used in the flexible space-time scan statistic. However, both of these were fitted to a small cluster size. Neill gave a very comprehensive account of spatial and ST clustering methods, especially in the area of scan statistics methods and Bayesian clustering methods. They proposed a statistical framework for detecting clusters in detail. The results of case studies show it has good performance compared to previous studies. However, they are still subject to the limitations of statistical methods.

Spatiotemporal Data Clustering: A Survey of Methods <s> DBSCAN <s> Clustering algorithms are attractive for the task of class identification in spatial databases. However, the application to large spatial databases rises the following requirements for clustering algorithms: minimal requirements of domain knowledge to determine the input parameters, discovery of clusters with arbitrary shape and good efficiency on large databases. The well-known clustering algorithms offer no solution to the combination of these requirements. In this paper, we present the new clustering algorithm DBSCAN relying on a density-based notion of clusters which is designed to discover clusters of arbitrary shape. DBSCAN requires only one input parameter and supports the user in determining an appropriate value for it. We performed an experimental evaluation of the effectiveness and efficiency of DBSCAN using synthetic data and real data of the SEQUOIA 2000 benchmark. The results of our experiments demonstrate that (1) DBSCAN is significantly more effective in discovering clusters of arbitrary shape than the well-known algorithm CLAR-ANS, and that (2) DBSCAN outperforms CLARANS by a factor of more than 100 in terms of efficiency. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> DBSCAN <s> In order to mine spatial-temporal clusters from geo-databases, two clustering methods with close relationships are proposed, which are both based on neighborhood searching strategy, and rely on the sorted k-dist graph to automatically specify their respective algorithm arguments. We declare the most distinguishing advantage of our clustering methods is they avoid calculating the spatial-temporal distance between patterns which is a tough job. Our methods are validated with the successful extraction of seismic sequence from seismic databases, which is a typical example of spatial–temporal clusters. <s> BIB002 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> DBSCAN <s> Clustering is one of the major data mining methods to obtain a number of clues about how the physical properties of the water are distributed in a marine environment. It is a difficult problem, especially when we consider the task for spatial–temporal marine data. This study introduces a new clustering algorithm to discover regions that have similar physical seawater characteristics. In contrast to the existing density-based clustering algorithms, our algorithm has the ability of discovering clusters according to the nonspatial, spatial, and temporal values of the objects. Our algorithm also overcomes three drawbacks of existing clustering algorithms: problems in the identification of core objects, noise objects, and adjacent clusters. This paper also presents a spatial–temporal marine data warehouse system designed for storing and clustering physical data from Turkish seas. Special functions were developed for data integration, data conversion, querying, visualization, analysis, and management. User-friendly interfaces were also developed, allowing relatively inexperienced users to operate the system. As a case study, we show the spatial–temporal distributions of sea surface temperature, sea surface height residual, and significant wave height values in Turkish seas to demonstrate our algorithm. <s> BIB003 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> DBSCAN <s> Understand the need for analyses of large, complex, information-rich data sets. Identify the goals and primary tasks of the data-mining process. Describe the roots of data-mining technology. Recognize the iterative character of a data-mining process and specify its basic steps. Explain the influence of data quality on a data-mining process. Establish the relation between data warehousing and data mining. Data mining is an iterative process within which progress is defined by discovery, through either automatic or manual methods. Data mining is most useful in an exploratory analysis scenario in which there are no predetermined notions about what will constitute an "interesting" outcome. Data mining is the search for new, valuable, and nontrivial information in large volumes of data. It is a cooperative effort of humans and computers. Best results are achieved by balancing the knowledge of human experts in describing problems and goals with the search capabilities of computers. In practice, the two primary goals of data mining tend to be prediction and description. Prediction involves using some variables or fields in the data set to predict unknown or future values of other variables of interest. Description, on the other hand, focuses on finding patterns describing the data that can be interpreted by humans. Therefore, it is possible to put data-mining activities into one of two categories: Predictive data mining, which produces the model of the system described by the given data set, or Descriptive data mining, which produces new, nontrivial information based on the available data set. <s> BIB004

DBSCAN is a very popular method, especially in the data mining community BIB004 . It has been extended for many different types of data. The biggest advantages of this method is that it can find clusters with arbitrary shape and noise points BIB001 . The key idea is that each cluster should include at least a minimum number of points with a fixed radius. Similar to kernel density estimation (KDE), DBSCAN can also be extended for spatiotemporal data. ST-DBSCAN BIB002 was proposed to cluster spatiotemporal data. Wang, et al. BIB002 added another radius r t which is the temporal neighborhood radius. The core points should satisfy directly the density reachable in both spatial radius r s and temporal radius r t . To define an appropriate spatial and temporal radius, k-dist graph was used to decide values. Generally speaking, cluster data should be clearly separated from noise data. To do this, the distance of each point to its k nearest neighbor, called the k value, was calculated. As depicted in Figure 4 , the left graph shows the distribution of point sample, clearly indicating three similar density clusters surrounded by noise points. The right graph was drawn based on a descending order of k values. The smooth red line on the right part of the graph highlights cluster points that have a low k value, but the left part of the red line indicates noise points that have high values. An appropriate threshold could be selected from the graph with an obvious and abrupt change from high value of small number of points to low value of large number of points. maximum length of cluster. The idea was also used in the flexible space-time scan statistic. However, both of these were fitted to a small cluster size. Neill gave a very comprehensive account of spatial and ST clustering methods, especially in the area of scan statistics methods and Bayesian clustering methods. They proposed a statistical framework for detecting clusters in detail. The results of case studies show it has good performance compared to previous studies. However, they are still subject to the limitations of statistical methods. DBSCAN is a very popular method, especially in the data mining community BIB004 . It has been extended for many different types of data. The biggest advantages of this method is that it can find clusters with arbitrary shape and noise points BIB001 . The key idea is that each cluster should include at least a minimum number of points with a fixed radius. Similar to kernel density estimation(KDE), DBSCAN can also be extended for spatiotemporal data. ST-DBSCAN BIB002 was proposed to cluster spatiotemporal data. Wang, et al. BIB002 added another radius which is the temporal neighborhood radius. The core points should satisfy directly the density reachable in both spatial radius and temporal radius . To define an appropriate spatial and temporal radius, -dist graph was used to decide values. Generally speaking, cluster data should be clearly separated from noise data. To do this, the distance of each point to its nearest neighbor, called the value, was calculated. As depicted in Figure 4 , the left graph shows the distribution of point sample, clearly indicating three similar density clusters surrounded by noise points. The right graph was drawn based on a descending order of values. The smooth red line on the right part of the graph highlights cluster points that have a low value, but the left part of the red line indicates noise points that have high values. An appropriate threshold could be selected from the graph with an obvious and abrupt change from high value of small number of points to low value of large number of points. Another method was called ST-GRID. The core idea was that a three-dimensional grid covers the entire dataset followed by merging the dense neighboring cells. First, the above k-dist graph could be used to define the border length of the grid and put all the data into a multi-dimension grid. Second, the number of points in each cell was counted. Those equal or larger than + 1 were merged with neighbor cells as a cluster. The process was repeated until no additional cells could be merged. Compared with the above method, more detailed data such as non-spatial data should be considered when extending DBSCAN BIB003 . A new method called ST-DBSCAN was proposed for Another method was called ST-GRID. The core idea was that a three-dimensional grid covers the entire dataset followed by merging the dense neighboring cells. First, the above k-dist graph could be used to define the border length of the grid and put all the data into a multi-dimension grid. Second, the number of points in each cell was counted. Those equal or larger than k + 1 were merged with neighbor cells as a cluster. The process was repeated until no additional cells could be merged. Compared with the above method, more detailed data such as non-spatial data should be considered when extending DBSCAN BIB003 . A new method called ST-DBSCAN was proposed for discovering clusters based on three attributes; non-spatial, spatial, and temporal attributes of data. Basic concepts were the same as conventional DBSCAN except for three modifications. When DBSCAN only considers one distance parameter to find similar data, ST-DBSCAN used two distance parameters for two-dimensional data. One distance measured two points distance in spatial scale. Another distance measured non-spatial attributes. Euclidean distance was adopted to calculate the two distances. where x and y represented spatial information. DBSCAN algorithm' result could be affected by selecting a different radius. If the dataset included different densities of clusters, a single radius could not clearly identify each cluster. To solve the problem, they proposed a concept called the density factor. Each cluster has their own density factor. To calculate it, three concepts of distances are introduced, which are density_distance_max, density_distance_min and density_distance. Density_distance_max was the maximum distance between object p and its neighbor objects within the radius Eps. Density_distance_min was the minimum distance of each cluster. The density_distance of object p was defined as density_distance_max (p)/density_distance_min (p). The density_factor was defined as follows. The density_factor C denoted the degree of each cluster. If the points of a cluster were close to each other, density_diatance_min would decrease, the density_distance would be quite large, and the density_factor would be close to 0. Otherwise, if points were a little further away from each other, the density_distance would be quite small and the density_factory would be close to 1. For non-spatial values of objects, this added value could change the average value of existing points when clustering. To solve the problem, ST-DBCSAN compared the average value of a cluster with every other point. If the absolute difference between the average value and object value was larger than a threshold, that point should not to be contained in the cluster.

Spatiotemporal Data Clustering: A Survey of Methods <s> Authors Methods <s> Abstract Previous research exploring space–time patterns has focused on the relative merits and drawbacks of the effectiveness of static maps vis-a-vis interactive dynamic visualisation techniques. In particular, they have tended to concentrate on the role of animation in interpretation of patterns and the understanding of underlying factors influencing such patterns. The aim of this paper is to broaden this debate out to consider the effectiveness of a wider range of visualisation techniques in permitting an understanding of spatio-temporal trends. The merits of three visualisation techniques, (map animation, the comap and the isosurface) are evaluated on their ability to assist in the exploration of space–time patterns of crime disturbance data. We conclude that each technique has some merit for crime analysts charged with studying such trends but that further research is needed to apply the techniques to other sources of crime data (and to other sectors such as health) to permit a comprehensive evaluation of their respective strengths and limitations as exploratory visualisation tools. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Authors Methods <s> For an effective interpretation of spatio-temporal patterns of crime clusters/hotspots, we explore the possibility of three-dimensional mapping of crime events in a space-time cube with the aid of space-time variants of kernel density estimation and scan statistics. Using the crime occurrence dataset of snatch-and-run offences in Kyoto City from 2003 to 2004, we confirm that the proposed methodology enables simultaneous visualisation of the geographical extent and duration of crime clusters, by which stable and transient space-time crime clusters can be intuitively differentiated. Also, the combined use of the two statistical techniques revealed temporal inter-cluster associations showing that transient clusters alternatively appeared in a pair of hotspot regions, suggesting a new type of “displacement” phenomenon of crime. Highlighting the complementary aspects of the two space-time statistical approaches, we conclude that combining these approaches in a space-time cube display is particularly valuable for a spatio-temporal exploratory data analysis of clusters to extract new knowledge of crime epidemiology from a data set of space-time crime events. <s> BIB002 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Authors Methods <s> ABSTRACTWe examined three different ways to integrate spatial and temporal data in kernel density estimation methods (KDE) to identify space–time clusters of geographic events. Spatial data and time data are typically measured in different units along respective dimensions. Therefore, spatial KDE methods require special extensions when incorporating temporal data to detect spatiotemporal clusters of geographical event. In addition to a real-world data set, we applied the proposed methods to simulated data that were generated through random and normal processes to compare results of different kernel functions. The comparison is based on hit rates and values of a compactness index with considerations of both spatial and temporal attributes of the data. The results show that the spatiotemporal KDE (STKDE) can reach higher hit rates while keeping identified hotspots compact. The implementation of these STKDE methods is tested using the 2012 crime event data in Akron, Ohio, as an example. The results show that... <s> BIB003 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Authors Methods <s> With the growing interest in studying the characteristics of people’s access to the food environment and its influence upon individual health, there has been a focus on assessing individual food exposure based on GPS trajectories. However, existing studies have largely focused on the overall activity space using short-period trajectories, which ignores the complexity of human movements and the heterogeneity of the spaces that are experienced by the individual over daily life schedules. In this study, we propose a novel framework to extract the exposure areas consisting of the localized activity spaces around daily life centers and non-motorized commuting routes from long-term GPS trajectories. The newly proposed framework is individual-specific and can incorporate the internal heterogeneity of individual activities (spatial extent, stay duration, and timing) in different places as well as the dynamics of the context. A pilot study of the GeoLife dataset suggests that there are significant variations in the magnitude as well as the composition of the food environment in different parts of the individual exposure area, and residential environment is not representative of the overall foodscape exposure. <s> BIB004

Brunsdon, et al. BIB001 Nakaya and Yano BIB002 Wei, et al. BIB004 Temporal attribute is regarded as another dimension, calculate space and time kernel density estimations separately. Lee, et al. BIB003 1. Setting a threshold to filter inappropriate space and time distances. 2. Standardization of space and time data for integrating them with same kernel function. Conventional KDE should be extended by adjusting the parameters for spatiotemporal data. Brunsdon, et al. BIB001 extended the two-dimensional KDE into three-dimensional for space and time data analysis. It helped to visualize and understand the trend of spatiotemporal data. The three-dimensional spatiotemporal KDE formula was: where the notation was the same as Equation (5), k t was the kernel function for time, h t was the bandwidth parameter of time kernel. Spatial and temporal information were treated separately, each of which had its own bandwidths and kernel functions. Nakaya and Yano BIB002 adopted this method for visualizing high-density crime events during a specific time interval in Kyoto. A threshold was set to filter data beyond a defined range. For most data, the longer space/time distance between two datasets, the lower possibility of their correlation. For example, if the time distance of two adjacent data was larger than a threshold, there was no need to calculate kernel density. The advantage of this method was no requirement to define a density function of time, but time was regarded as a constant. The formula was: In this formula, only kernel density of space needs to be calculated. However, it is difficult to define an appropriate method for filtering time. In order to directly integrated space and time data, the process of standardization should be conducted before density estimation with the following equations: and, where s , t were spatial and temporal raw data, s, t could be referenced values for standardizing raw spatial and temporal data and h s ,h t were their kernel bandwidths. The advantage of standardization of raw spatial and temporal data was to remove the different measurement units of spatial and temporal data. The results of standardization of spatial and temporal data was that they have similar ranges for easy integration. The calculation of kernel density estimation waŝ However, it is noted that bandwidth selection was a critical problem that will affect cluster results. The unit of time was another problem because different units lead to different density of clusters.

Spatiotemporal Data Clustering: A Survey of Methods <s> Windowed Nearest Neighbor Method <s> Vibratory power unit for vibrating conveyers and screens comprising an asynchronous polyphase motor, at least one pair of associated unbalanced masses disposed on the shaft of said motor, with the first mass of a pair of said unbalanced masses being rigidly fastened to said shaft and with said second mass of said pair being movably arranged relative to said first mass, means for controlling and regulating the conveying rate during conveyer operation by varying the rotational speed of said motor between predetermined minimum and maximum values, said second mass being movably outwardly by centrifugal force against the pressure of spring means, said spring means being prestressed in such a manner that said second mass is, at rotational motor speeds lower than said minimum speed, held in its initial position, and at motor speeds between said lower and upper values in positions which are radially offset with respect to the axis of said motor to an extent depending on the value of said rotational motor speed. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Windowed Nearest Neighbor Method <s> Abstract We consider the problem of detecting features in spatial point processes in the presence of substantial clutter. One example is the detection of minefields using reconnaissance aircraft images that identify many objects that are not mines. Our solution uses Kth nearest neighbor distances of points in the process to classify them as clutter or otherwise. The observed Kth nearest neighbor distances are modeled as a mixture distribution, the parameters of which are estimated by a simple EM algorithm. This method allows for detection of generally shaped features that need not be path connected. In the minefield example this method yields high detection and low false-positive rates. Another application, to outlining seismic faults, is considered with some success. The method works well in high dimensions. The method can also be used to produce very high-breakdown-point–robust estimators of a covariance matrix. <s> BIB002 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Windowed Nearest Neighbor Method <s> In a spatio-temporal data set, identifying spatio-temporal clusters is difficult because of the coupling of time and space and the interference of noise. Previous methods employ either the window scanning technique or the spatio-temporal distance technique to identify spatio-temporal clusters. Although easily implemented, they suffer from the subjectivity in the choice of parameters for classification. In this article, we use the windowed kth nearest (WKN) distance (the geographic distance between an event and its kth geographical nearest neighbour among those events from which to the event the temporal distances are no larger than the half of a specified time window width [TWW]) to differentiate clusters from noise in spatio-temporal data. The windowed nearest neighbour (WNN) method is composed of four steps. The first is to construct a sequence of TWW factors, with which the WKN distances of events can be computed at different temporal scales. Second, the appropriate values of TWW (i.e. the appropriate temporal scales, at which the number of false positives may reach the lowest value when classifying the events) are indicated by the local maximum values of densities of identified clustered events, which are calculated over varying TWW by using the expectation-maximization algorithm. Third, the thresholds of the WKN distance for classification are then derived with the determined TWW. In the fourth step, clustered events identified at the determined TWW are connected into clusters according to their density connectivity in geographic–temporal space. Results of simulated data and a seismic case study showed that the WNN method is efficient in identifying spatio-temporal clusters. The novelty of WNN is that it can not only identify spatio-temporal clusters with arbitrary shapes and different spatio-temporal densities but also significantly reduce the subjectivity in the classification process. <s> BIB003

Based on the idea of spatiotemporal k nearest neighbors test, windowed nearest neighbor method for mining spatiotemporal clusters was proposed several years ago BIB003 . Spatiotemporal point data could be represented by ST p , each point indicated by ST p (s i , t i ), and its neighbor could be defined as: For k nearest neighbors, the time interval of consecutive two points should be smaller than a threshold, |T i+1 − T i | ≤ ∆T. The distances D ST p from a given point to the rests are gradually increasing with time satisfied as: Similar to space-time scan statistics, each event could be regarded as a center of cylinder with a spatial radius and temporal height. A cylinder as a window includes spatiotemporal neighbors of a given event. A core event's neighbor should contain a minimum number of other points. The first step is to distinguish between a cluster of events and noise; second is to connect the cylinder into cluster events. Figure 5 shows the spatiotemporal density connectivity of events from a horizontal perspective to form the cluster. However, it is noted that bandwidth selection was a critical problem that will affect cluster results. The unit of time was another problem because different units lead to different density of clusters. Based on the idea of spatiotemporal nearest neighbors test, windowed nearest neighbor method for mining spatiotemporal clusters was proposed several years ago BIB003 . Spatiotemporal point data could be represented by , each point indicated by ( , ), and its neighbor could be defined as: For nearest neighbors, the time interval of consecutive two points should be smaller than a threshold, | − | ∆ . The distances from a given point to the rests are gradually increasing with time satisfied as: Similar to space-time scan statistics, each event could be regarded as a center of cylinder with a spatial radius and temporal height. A cylinder as a window includes spatiotemporal neighbors of a given event. A core event's neighbor should contain a minimum number of other points. The first step is to distinguish between a cluster of events and noise; second is to connect the cylinder into cluster events. Figure 5 shows the spatiotemporal density connectivity of events from a horizontal perspective to form the cluster. In their method, an ST Poisson point process was used to construct probability density function with the equation: where was the number of events in the volume of , was the constant. The density of cylinder can be calculated by: where was the number of events, and ∆ was the temporal interval constant. ∆ could be regarded as a threshold calculated by an expectation maximization (EM) algorithm BIB001 . A detailed process of the EM algorithm can be found in Byers and Raftery BIB002 . After the density connected In their method, an ST Poisson point process was used to construct probability density function with the equation: where k was the number of events in the volume of V, λ was the constant. The density of cylinder D can be calculated by: where k was the number of events, and ∆T was the temporal interval constant. ∆S t could be regarded as a threshold calculated by an expectation maximization (EM) algorithm BIB001 . A detailed process of the EM algorithm can be found in Byers and Raftery BIB002 . After the density connected events were divided into cluster events and noise features, they were linked by the cylinder for connecting events into clusters.

Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> —Earthquake clusterings in both space and time have various forms, in particular, two typical examples are the foreshock sequences and earthquake swarms. Based on the analysis of 8 foreshock sequences in mainland China during 1966–1996, this study concentrates on the pattern characteristics of foreshock sequences. The following pattern characteristics of foreshock sequences have been found (1) the epicenters of foreshock sequences were densely concentrated in space; (2) the focal mechanisms of foreshocks were similar to that of the main shock. Such consistency of focal mechanisms with main shocks did not exist in aftershock series as well as in several earthquake swarms; (3) we found no case in mainland China during the past thirty years that a main shock is preceded by an earthquake clustering with inconsistent focal mechanisms. Finally, we found 5% of the main shocks in mainland China are preceded by foreshock sequences. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> On September 3rd (22:07 UTC), 1997 a small earthquake with Mw=4.54 started the foreshocks sequence (≈1500 events with ML <3.1) of the September 26th seismic sequence. Two days after, three seismic stations of the University of Camerino were installed around the macroseismic epicenter of the foreshock. We present in this paper the location of foreshocks (with 2.1<ML<3.3) which occurred between September 3rd and 26th. Foreshocks location, with horizontal (ERH) and vertical (ERZ) error less than 1.5 km, define an area ≈4 km large. Foreshocks have been localized between the epicenters of the two major events of September 26th, which occurred at 00:33 UTC with Mw=5.6 and at 09:40 UTC with Mw=6.0 (Amato et al., 1998; Ekstrom et al., 1998). In a vertical cross-section, hypocenters show a low angle (≈30°) structure with SW dip-direction. Focal mechanisms for three of the major events show dip-slip fault solutions with strike direction of about N130, in agreement with the CMT solutions of September 3rd and September 26th earthquakes (Ekstrom et al., 1998). Data recorded at two stations Popola (POP) e Capodacqua (CPQ) located on the rupture area of the September 26th faults, allowed us to calculate a mean Vp/Vs ratio of 1.84±0.03 for the foreshock. This value is lower than the Vp/Vs ratio of 1.89±0.02 calculated for the aftershock sequence occurred in the same area. Besides, the Vp/Vs ratio during the foreshocks sequence is not stable in time but it seems to increase approaching September 26th. After September 26th mainshocks, this value tends to stabilize around a higher value of 1.89. Following the dilatancy model, we suggest that the relative low Vp/Vs ratio before the main shocks could indicate the presence of fluid in the focal volume. The presence of fluids could have increased the effective stress on the fault plane and could be responsible for the long foreshock activity before the two main earthquakes of September 26th. Therefore, we suggest that this foreshock activity could have also contributed to reduce the friction along the September 26th fault plane, breaking the active structure in two smaller segments. In this hypothesis, foreshock activity could have drastically contributed to mitigate the seismic potential of the Colfiorito's active structure. <s> BIB002 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> Identifying road accident hotspots is a key role in determining effective strategies for the reduction of high density areas of accidents. This paper presents (1) a methodology using Geographical Information Systems (GIS) and Kernel Density Estimation to study the spatial patterns of injury related road accidents in London, UK and (2) a clustering methodology using environmental data and results from the first section in order to create a classification of road accident hotspots. The use of this methodology will be illustrated using the London area in the UK. Road accident data collected by the Metropolitan Police from 1999 to 2003 was used. A kernel density estimation map was created and subsequently disaggregated by cell density to create a basic spatial unit of an accident hotspot. Appended environmental data was then added to the hotspot cells and using K-means clustering, an outcome of similar hotspots was deciphered. Five groups and 15 clusters were created based on collision and attribute data. These clusters are discussed and evaluated according to their robustness and potential uses in road safety campaigning. <s> BIB003 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> For an effective interpretation of spatio-temporal patterns of crime clusters/hotspots, we explore the possibility of three-dimensional mapping of crime events in a space-time cube with the aid of space-time variants of kernel density estimation and scan statistics. Using the crime occurrence dataset of snatch-and-run offences in Kyoto City from 2003 to 2004, we confirm that the proposed methodology enables simultaneous visualisation of the geographical extent and duration of crime clusters, by which stable and transient space-time crime clusters can be intuitively differentiated. Also, the combined use of the two statistical techniques revealed temporal inter-cluster associations showing that transient clusters alternatively appeared in a pair of hotspot regions, suggesting a new type of “displacement” phenomenon of crime. Highlighting the complementary aspects of the two space-time statistical approaches, we conclude that combining these approaches in a space-time cube display is particularly valuable for a spatio-temporal exploratory data analysis of clusters to extract new knowledge of crime epidemiology from a data set of space-time crime events. <s> BIB004 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> Twitter is a unique social media channel, in the sense that users discuss and talk about the most diverse topics, including their health conditions. In this paper we analyze how Dengue epidemic is reflected on Twitter and to what extent that information can be used for the sake of surveillance. Dengue is a mosquito-borne infectious disease that is a leading cause of illness and death in tropical and subtropical regions, including Brazil. We propose an active surveillance methodology that is based on four dimensions: volume, location, time and public perception. First we explore the public perception dimension by performing sentiment analysis. This analysis enables us to filter out content that is not relevant for the sake of Dengue surveillance. Then, we verify the high correlation between the number of cases reported by official statistics and the number of tweets posted during the same time period (i.e., R2 = 0.9578). A clustering approach was used in order to exploit the spatio-temporal dimension, and the quality of the clusters obtained becomes evident when they are compared to official data (i.e., RandIndex = 0.8914). As an application, we propose a Dengue surveillance system that shows the evolution of the dengue situation reported in tweets, which is implemented in www.observatorio.inweb.org.br/dengue/. <s> BIB005 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> Urban geographers, planners, and economists have long been studying urban spatial structure to understand the development of cities. Statistical and data mining techniques, as proposed in this paper, go a long way in improving our knowledge about human activities extracted from travel surveys. As of today, most urban simulators have not yet incorporated the various types of individuals by their daily activities. In this work, we detect clusters of individuals by daily activity patterns, integrated with their usage of space and time, and show that daily routines can be highly predictable, with clear differences depending on the group, e.g. students vs. part time workers. This analysis presents the basis to capture collective activities at large scales and expand our perception of urban structure from the spatial dimension to spatial-temporal dimension. It will be helpful for planers to understand how individuals utilize time and interact with urban space in metropolitan areas and crucial for the design of sustainable cities in the future. <s> BIB006 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> “Classic” kernel density estimations (KDE) can display static densities representing one point in time. It is not possible to visually identify which parts of the densities are moving. Therefore, within this paper we investigate how to display dynamic densities (and the density changes) to identify movement patterns. To deal with a temporal dimension (in our case study a dynamic crowd of individuals) we investigated the application of directed kernel density estimation (DKDE). In a case study we apply the DKDE to a point dataset presenting individuals approaching the Allianz Arena in Munich, Germany, with different speeds from different directions. Calculating the density using a directed kernel with this data, results in a density map indicating the movement direction with a visible “ripple” effect. Ripples move at different rates to the substances in which they occur. That tells us something about crowd dynamics and enables us to visually recognize the parts of the crowds that are moving plus the underlying movement directions. <s> BIB007 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> Dengue fever is an arboviral disease typical of the tropics that can be life-threatening and if not controlled properly may result in an epidemic. The absence of an effective vaccine makes strategies to prevent the virus transmission the most effective means of control. The planning of such strategies, however, is difficult due to the constant movement of individuals and mosquito host (Aedes aegypti). In this paper, the spatial and temporal relations that might exist between infected individuals during a dengue-epidemic year are explored. This research is motivated in that a deep understanding of potential transmission patterns between individuals might lead to a better design and planning of control strategies. A GIS-based Health Exploratory AnaLysis Tool (HELP) is used to compute space-time relationships by means of spatial K-function, kernel density, space-time K-function and linking pairs of cases within significant time and space intervals. Significant clustering was observed at a scale of 50 meters and 750 meters, respectively while temporal significance was determined at two days and five to eight days. While an increase of cases occurs in the months following severe droughts due to an El NiA±o phenomenon, the location of clusters remains relatively stable. These are observed near areas where potential habitats for the mosquito exist such as storm drains, hard surfaces where water accumulates (e.g., vases, containers), but also in poorer neighborhoods. The results from the spatial analysis provide valuable information for health care managers to take preventive actions at the municipality level. <s> BIB008 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> Background ::: Every day, around 400 million tweets are sent worldwide, which has become a rich source for detecting, monitoring and analysing news stories and special (disaster) events. Existing research within this field follows key words attributed to an event, monitoring temporal changes in word usage. However, this method requires prior knowledge of the event in order to know which words to follow, and does not guarantee that the words chosen will be the most appropriate to monitor. ::: ::: ::: Methods ::: This paper suggests an alternative methodology for event detection using space-time scan statistics (STSS). This technique looks for clusters within the dataset across both space and time, regardless of tweet content. It is expected that clusters of tweets will emerge during spatio-temporally relevant events, as people will tweet more than expected in order to describe the event and spread information. The special event used as a case study is the 2013 London helicopter crash. ::: ::: ::: Results and Conclusion ::: A spatio-temporally significant cluster is found relating to the London helicopter crash. Although the cluster only remains significant for a relatively short time, it is rich in information, such as important key words and photographs. The method also detects other special events such as football matches, as well as train and flight delays from Twitter data. These findings demonstrate that STSS is an effective approach to analysing Twitter data for event detection. <s> BIB009 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> In this article, we evaluate the impact of positional and temporal inaccuracies on the mapping and detection of potential outbreaks of dengue fever in Cali, an urban environment of Colombia. Positional uncertainties in input data are determined by comparison between coordinates following an automated geocoding process and those extracted from on-field GPS measurements. Temporal uncertainties are modeled around the incubation period for dengue fever. To test the robustness of disease intensities in space and time when accounting for the potential space-time error, each dengue case is perturbed using Monte Carlo simulations. A space-time kernel density estimation (STKDE) is conducted on both perturbed and observed sets of dengue cases. To reduce the computational effort, we use a parallel spatial computing solution. The results are visualized in a 3D framework, which facilitates the discovery of new, significant space-time patterns and shapes of dengue outbreaks while enhancing our understanding of complex and uncertain dynamics of vector-borne diseases. <s> BIB010 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> Predicting ambulance demand accurately at fine time and location scales is critical for ambulance fleet management and dynamic deployment. Large-scale datasets in this setting typically exhibit complex spatio-temporal dynamics and sparsity at high resolutions. We propose a predictive method using spatio-temporal kernel density estimation (stKDE) to address these challenges, and provide spatial density predictions for ambulance demand in Toronto, Canada as it varies over hourly intervals. Specifically, we weight the spatial kernel of each historical observation by its informativeness to the current predictive task. We construct spatio-temporal weight functions to incorporate various temporal and spatial patterns in ambulance demand, including location-specific seasonalities and short-term serial dependence. This allows us to draw out the most helpful historical data, and exploit spatio-temporal patterns in the data for accurate and fast predictions. We further provide efficient estimation and customizable prediction procedures. stKDE is easy to use and interpret by non-specialized personnel from the emergency medical service industry. It also has significantly higher statistical accuracy than the current industry practice, with a comparable amount of computational expense. <s> BIB011 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> Abstract Predictive hotspot mapping plays a critical role in hotspot policing. Existing methods such as the popular kernel density estimation (KDE) do not consider the temporal dimension of crime. Building upon recent works in related fields, this article proposes a spatio-temporal framework for predictive hotspot mapping and evaluation. Comparing to existing work in this scope, the proposed framework has four major features: (1) a spatio-temporal kernel density estimation (STKDE) method is applied to include the temporal component in predictive hotspot mapping, (2) a data-driven optimization technique, the likelihood cross-validation, is used to select the most appropriate bandwidths, (3) a statistical significance test is designed to filter out false positives in the density estimates, and (4) a new metric, the predictive accuracy index (PAI) curve, is proposed to evaluate predictive hotspots at multiple areal scales. The framework is illustrated in a case study of residential burglaries in Baton Rouge, Louisiana in 2011, and the results validate its utility. <s> BIB012 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> Population-level disease risk across a set of non-overlapping areal units varies in space and time, and a large research literature has developed methodology for identifying clusters of areal units exhibiting elevated risks. However, almost no research has extended the clustering paradigm to identify groups of areal units exhibiting similar temporal disease trends. We present a novel Bayesian hierarchical mixture model for achieving this goal, with inference based on a Metropolis coupled Markov chain Monte Carlo ((MC)3) algorithm. The effectiveness of the (MC)3 algorithm compared to a standard MCMC implementation is demonstrated in a simulation study, and the methodology is motivated by two important case studies in Scotland. The first concerns the impact on measles susceptibility of the discredited paper linking the Measles, Mumps and Rubella vaccination to an increased risk of Autism, and investigates whether all areas in Scotland were equally affected. The second concerns respiratory hospitalisations, and investigates over a 10 year period which parts of Glasgow have shown increased, decreased, and no change in risk. <s> BIB013 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Applications <s> Disease risk maps for areal unit data are often estimated from Poisson mixed models with local spatial smoothing, for example by incorporating random effects with a conditional autoregressive prior distribution. However, one of the limitations is that local discontinuities in the spatial pattern are not usually modelled, leading to over-smoothing of the risk maps and a masking of clusters of hot/coldspot areas. In this paper, we propose a novel two-stage approach to estimate and map disease risk in the presence of such local discontinuities and clusters. We propose approaches in both spatial and spatio-temporal domains, where for the latter the clusters can either be fixed or allowed to vary over time. In the first stage, we apply an agglomerative hierarchical clustering algorithm to training data to provide sets of potential clusters, and in the second stage, a two-level spatial or spatio-temporal model is applied to each potential cluster configuration. The superiority of the proposed approach with regard to a previous proposal is shown by simulation, and the methodology is applied to two important public health problems in Spain, namely stomach cancer mortality across Spain and brain cancer incidence in the Navarre and Basque Country regions of Spain. <s> BIB014

ST data clustering methods are widely used in many research areas, which we have divided into the following categories. Crime analysis: Criminal events usually repeatedly occur under the same situation and a similar time. Tracing the changes of a crime path is meaningful. Nakaya and Yano BIB004 explored the possibility of tracing crime events with three-dimensional attributes in a space-time cube relative to kernel density estimation and scan statistics to get the clustering of crime events and visualizing the crime events patterns. Hu, et al. BIB012 proposed a new modification of an existing method to increase the predictive accuracy of crime hotspots. They refined spatiotemporal kernel density estimation by generalized product kernels and adopted a data driven bandwidth selection to decide bandwidth. Residential burglaries data of Baton Rouge was used to predict crime hotspots. Events detection: Many events could be detected using clustering methods, such as helicopter crash accidents from social media data BIB009 . By using space-time scan statistics, a ST significant cluster of London helicopter crash locations were found. Many other events like football games and train and flight delays could also be detected. Clustering earthquake events could help to understand trends and mechanisms BIB001 BIB002 . Many small earthquakes can happen before or after a strong earthquake. By using ST clustering methods, clusters of earthquakes can be identified in space and time. ST kernel density estimation can be used for predicting ambulance demand. It is difficult to predict ambulance demand accurately from large-scale datasets of past events. Zhou and Matteson BIB011 proposed a model of spatiotemporal kernel density predictive method to explore ambulance demand precisely. KDE is also widely used in creating a density map of road accidents to identify its distribution pattern BIB003 . This could help to predict and reduce the number of incidents in the future. Mobility: Human mobility data such as phone call data could reflect urban growth in space and time. It would provide information for authorities to plan and manage cities in a smart way. It helps planners to understand where and when different groups of people interact in urban space. Jiang, et al. BIB006 discovered the clusters of human mobility pattern by kernel density estimation and integrating various spatial and temporal data to predict human daily routines. Krisp, et al. BIB007 proposed directed kernel density estimation to recognize movement and direction of crowds and was effective in visualizing the movement of crowds. Disease analysis: ST clustering methods could be applied in analyzing disease dispersion and trends. Visualizing space-time clusters of dengue fever pattern in Cali using extension of kernel density estimation method has been applied BIB008 BIB010 . The occurrence and spread of disease has a strong regular pattern in certain regions. Analyzing the former spread of disease to predict the future spread direction is meaningful for governments and hospitals to control diseases. Gomide, et al. BIB005 analyzed not only the location and time the disease was contracted, but also the reaction of the population when facing the disease. They used the ST-DBSCAN clustering method to explore the ST distribution characteristics of disease incidents to group nearby cities that have similar incident rates. A linear regression model was built to predict the number of diseases using the proportion of user experiences. Napier, et al. BIB013 proposed a novel Bayesian model to identify the cluster of similar temporal disease trends rather than disease estimation and prediction. Adin, et al. BIB014 proposed a two-stage approach to estimate disease risk maps. Compared with traditional methods, their method has the ability to overcome the problem of local discontinuities in the spatial pattern that cannot be modeled. It has a good performance of spatiotemporal smoothing for estimating risks of disease mapping.

Spatiotemporal Data Clustering: A Survey of Methods <s> Conclusion and Future Works <s> Spatial data mining is the process of discovering interesting and previously unknown, but potentially useful patterns from the spatial and spatiotemporal data. However, explosive growth in the spatial and spatiotemporal data, and the emergence of social media and location sensing technologies emphasize the need for developing new and computationally efficient methods tailored for analyzing big data. In this paper, we review major spatial data mining algorithms by closely looking at the computational and I/O requirements and allude to few applications dealing with big spatial data. <s> BIB001 </s> Spatiotemporal Data Clustering: A Survey of Methods <s> Conclusion and Future Works <s> Explosive growth in geospatial and temporal data as well as the emergence of new technologies emphasize the need for automated discovery of spatiotemporal knowledge. Spatiotemporal data mining studies the process of discovering interesting and previously unknown, but potentially useful patterns from large spatiotemporal databases. It has broad application domains including ecology and environmental management, public safety, transportation, earth science, epidemiology, and climatology. The complexity of spatiotemporal data and intrinsic relationships limits the usefulness of conventional data science techniques for extracting spatiotemporal patterns. In this survey, we review recent computational techniques and tools in spatiotemporal data mining, focusing on several major pattern families: spatiotemporal outlier, spatiotemporal coupling and tele-coupling, spatiotemporal prediction, spatiotemporal partitioning and summarization, spatiotemporal hotspots, and change detection. Compared with other surveys in the literature, this paper emphasizes the statistical foundations of spatiotemporal data mining and provides comprehensive coverage of computational approaches for various pattern families. ISPRS Int. J. Geo-Inf. 2015, 4 2307 We also list popular software tools for spatiotemporal data analysis. The survey concludes with a look at future research needs. <s> BIB002

ST data clustering analysis is a hot topic and has already been studied extensively BIB002 . ST data types can be classified into three categories, namely point, line, and polygon. In this paper, only point pattern is considered and existing clustering methods are divided into two parts, one is hypothesis testing based, and another is partitional clustering methods. ST data is more complicated than other types of data because of the additional dimension of time from two-dimensional spatial analysis. Some popular and representative methods are introduced in previous sections. However, simply regarding time as an extended dimension may ignore some important patterns that are hard to be detected. New methods should consider integrating time and other attributes together. Clustering is an important step to detect patterns from a large amount of data. It can be used in many application domains, including transportation, social media, and urban development. It focuses on finding hotspots from raw data. These hotspots are the foundation for pattern understanding. Adjusting different parameters of the clustering method for different data types is needed to get an optimum result. An appropriate clustering method can help discover potential and useful information from a large volume of data. Asides from investigating new algorithms, related research problems have been developed, such as the computational issues of ST data BIB001 . As mentioned before, even though extended algorithms could be used to detect clusters, these are more than mere geometrical considerations. There is a need to predefine thresholds such as radius, distance, and density based on the rules or knowledge from specific themes. As such, new research trends and methods need to be developed. ST data analysis has attracted much research attention and a lot of methods have been developed . However, there are still some issues and challenges to be solved. Several challenge issues are described as follows:

Sample transportation – an overview <s> Background <s> BACKGROUND ::: The stability of analyte concentrations in plasma after prolonged contact with blood cells in uncentrifuged lithium-heparin gel tubes was studied. ::: ::: ::: METHODS ::: To investigate the stability of concentrations of 26 chemistry and 15 immunochemistry analytes, the simultaneously drawn samples (n = 50) were measured after 6 h storage at +8 degrees C and +22 degrees C in whole blood and after immediate separation of plasma. The analyte concentrations were measured with a Roche Modular PPEE analyser using reagents from Roche Diagnostics. ::: ::: ::: RESULTS ::: After prolonged contact with cells a clinically significant change was only observed for potassium where the mean value increased from 4.0 mmol/L to 4.8 mmol/L (P < 0.001) when stored at +8 degrees C. ::: ::: ::: CONCLUSION ::: Immediate separation of plasma from cells is recommended. However, when prolonged contact of plasma with cell is unavoidable, samples can be kept uncentrifuged for up to 6 h at +8 degrees C or at +22 degrees C. The stability of potassium, however, is temperature-dependent and cannot be measured from refrigerated blood samples. <s> BIB001 </s> Sample transportation – an overview <s> Background <s> Abstract Laboratory diagnostics (i.e., the total testing process) develops conventionally through a virtual loop, originally referred to as "the brain to brain cycle" by George Lundberg. Throughout ... <s> BIB002 </s> Sample transportation – an overview <s> Background <s> OBJECTIVE ::: We studied the pre-analytical stability of 81 analytes based on the variables of delay before processing, storage as whole blood or serum/plasma, the storage temperature and the type of tube the sample was stored in. ::: ::: ::: DESIGN AND METHODS ::: The mean difference between assays for samples from 10 subjects was calculated with the samples being kept under different storage conditions and for different times between sampling time and analysis: up to 24h for biochemistry, coagulation and hematology, and up to 72 h for hormonology. This difference was compared to the acceptable limits derived from the analytical and the intra individual biological variation. ::: ::: ::: RESULTS ::: Most of the analytes investigated remained stable up to 24h under all storage conditions prior to centrifugation. However, some analytes were significantly affected either by delay, tube type or temperature, such as potassium, inorganic phosphorus, magnesium, LD, glucose, lactate, mean corpuscular volume, mean corpuscular hemoglobin, activated partial thromboplastin time, insulin, C-peptide, PTH, osteocalcin, C-telopeptide and ACTH. ::: ::: ::: CONCLUSION ::: This study may be useful to help define acceptable delay times and storage conditions when a short time between sample collection and processing is not possible. <s> BIB003 </s> Sample transportation – an overview <s> Background <s> Abstract Objectives Due to the consolidation of laboratory testing facilities, there is an increasing need for systems able to assure quality and safety in biological sample transportation, although little evidence on this aspect is available in literature. Design and methods An integrated system for sample transportation, implemented and monitored over a five-year period by our team, consists of secondary and tertiary containers, a device for temperature and time recording, and a system manager allowing the acceptance or rejection of biological samples through the immediate visualization and validation of registered data. Results Data collected between 2009 and October 2011, after a preliminary phase for optimizing the temperature inside the containers, demonstrated the frequency of transportations at an acceptable temperature ( 25 °C) had decreased by ~ 80%. Conclusions The integrated system and related operating instructions allow improvement in the quality of sample transportation over time. <s> BIB004 </s> Sample transportation – an overview <s> Background <s> Using simulation as an approach to display and improve internal logistics at hospitals has great potential. This study shows how a simulation model displaying the morning blood-taking round at a Danish public hospital can be developed and utilized with the aim of improving the logistics. The focus of the simulation was to evaluate changes made to the transportation of blood samples between wards and the laboratory. The average- (AWT) and maximum waiting time (MWT) from a blood sample was drawn at the ward until it was received at the laboratory, and the distribution of arrivals of blood samples in the laboratory were used as the evaluation criteria. Four different scenarios were tested and compared with the current approach: (1) Using AGVs (mobile robots), (2) using a pneumatic tube system, (3) using porters that are called upon, or (4) using porters that come to the wards every 45 minutes. Furthermore, each of the scenarios was tested in terms of what amount of resources would give the optimal result. The simulations showed a big improvement potential in implementing a new technology/mean for transporting the blood samples. The pneumatic tube system showed the biggest potential lowering the AWT and MWT with approx. 36% and 18%, respectively. Additionally, all of the scenarios had a more even distribution of arrivals except for porters coming to the wards every 45 min. As a consequence of the results obtained in the study, the hospital decided to implement a pneumatic tube system. <s> BIB005 </s> Sample transportation – an overview <s> Background <s> AbstractBackground. Suitable procedures for transport of blood samples from general practitioners to hospital laboratories are requested. Here we explore routine testing on samples stored and transported as whole blood in lithium-heparin or serum tubes. Methods. Blood samples were collected from 106 hospitalized patients, and analyzed on Architect c8000 or Advia Centaur XP for 35 analytes at base line, and after storage and transport of whole blood in lithium-heparin or serum tubes at 21 ± 1°C for 10 h. Bias and imprecision (representing variation from analysis and storage) were calculated from values at baseline and after storage, and differences tested by paired t-tests. Results were compared to goals set by the laboratory. Results. We observed no statistically significant bias and results within the goal for imprecision between baseline samples and 10-h samples for albumin, alkaline phosphatase, antitrypsin, bilirubin, creatinine, free triiodothyronine, γ-glutamyl transferase, haptoglobin, immunoglobul... <s> BIB006 </s> Sample transportation – an overview <s> Background <s> Blood samples are usually collected daily from different collection points, such hospitals and health centers, and transported to a core laboratory for testing. This paper presents a project to improve the collection routes of two of the largest clinical laboratories in Spain. These routes must be designed in a cost-efficient manner while satisfying two important constraints: (i) two-hour time windows between collection and delivery, and (ii) vehicle capacity. A heuristic method based on a genetic algorithm has been designed to solve the problem of blood sample collection. The user enters the following information for each collection point: postal address, average collecting time, and average demand (in thermal containers). After implementing the algorithm using C programming, this is run and, in few seconds, it obtains optimal (or near-optimal) collection routes that specify the collection sequence for each vehicle. Different scenarios using various types of vehicles have been considered. Unless new collection points are added or problem parameters are changed substantially, routes need to be designed only once. The two laboratories in this study previously planned routes manually for 43 and 74 collection points, respectively. These routes were covered by an external carrier company. With the implementation of this algorithm, the number of routes could be reduced from ten to seven in one laboratory and from twelve to nine in the other, which represents significant annual savings in transportation costs. The algorithm presented can be easily implemented in other laboratories that face this type of problem, and it is particularly interesting and useful as the number of collection points increases. The method designs blood collection routes with reduced costs that meet the time and capacity constraints of the problem. <s> BIB007 </s> Sample transportation – an overview <s> Background <s> BACKGROUND ::: The stability of biochemical analytes has already been investigated, but results strongly differ depending on parameters, methodologies, and sample storage times. We investigated the stability for many biochemical parameters after different storage times of both whole blood and plasma, in order to define acceptable pre- and postcentrifugation delays in hospital laboratories. ::: ::: ::: METHODS ::: Twenty-four analytes were measured (Modular® Roche analyzer) in plasma obtained from blood collected into lithium heparin gel tubes, after 2-6 hr of storage at room temperature either before (n = 28: stability in whole blood) or after (n = 21: stability in plasma) centrifugation. Variations in concentrations were expressed as mean bias from baseline, using the analytical change limit (ACL%) or the reference change value (RCV%) as acceptance limit. ::: ::: ::: RESULTS ::: In tubes stored before centrifugation, mean plasma concentrations significantly decreased after 3 hr for phosphorus (-6.1% [95% CI: -7.4 to -4.7%]; ACL 4.62%) and lactate dehydrogenase (LDH; -5.7% [95% CI: -7.4 to -4.1%]; ACL 5.17%), and slightly decreased after 6 hr for potassium (-2.9% [95% CI: -5.3 to -0.5%]; ACL 4.13%). In plasma stored after centrifugation, mean concentrations decreased after 6 hr for bicarbonates (-19.7% [95% CI: -22.9 to -16.5%]; ACL 15.4%), and moderately increased after 4 hr for LDH (+6.0% [95% CI: +4.3 to +7.6%]; ACL 5.17%). Based on RCV, all the analytes can be considered stable up to 6 hr, whether before or after centrifugation. ::: ::: ::: CONCLUSION ::: This study proposes acceptable delays for most biochemical tests on lithium heparin gel tubes arriving at the laboratory or needing to be reanalyzed. <s> BIB008 </s> Sample transportation – an overview <s> Background <s> diagnostic samples:from the patient to the laboratory , diagnostic samples:from the patient to the laboratory , کتابخانه دانشگاه علوم پزشکی و خدمات درمانی بوشهر <s> BIB009 </s> Sample transportation – an overview <s> Background <s> BACKGROUND: ::: ::: Blood specimens are transported from clinical departments to the biochemistry laboratory by hospital courier service, sometimes over long distances. The aim of this study was to assess the stability of common biochemical analytes in venous blood under our routine transport conditions and to evaluate analyte stability after prompt or delayed centrifugation. ::: METHODS: ::: ::: We investigated pre- and postanalytical contributions of 32 biochemical analytes in plasma and serum samples from 10 patients (healthy adults and patients from intensive care units). Differences in analyte concentrations between baseline (T0) and different time intervals (2, 4, 6, 8, 12 and 24 h) following storage after prompt and delayed centrifugation were reported. Evaluation was against the total change limit as described by Oddoze et al. (Oddoze C, Lombard E, Portugal H. Stability study of 81 analytes in human whole blood, in serum and in plasma. Clin Biochem 2012;45:464-9). ::: RESULTS: ::: ::: The majority of analytes were stable with delayed separation up to 12 h, except for potassium, C-peptide, osteocalcin, parathyroid hormone (PTH), bicarbonate and LDH. After prompt centrifugation and storage at 4°C, stability was greatly increased up to 48 h for most analytes. LDH and bicarbonate had the lowest stability after centrifugation; therefore, no reanalysis of these analytes in a centrifuged tube can be allowed. ::: CONCLUSIONS: ::: ::: Knowledge of analyte stability is crucial to interpret biological analysis with confidence. However, centrifugation prior to transport is time consuming, and the transfer of plasma or serum from a primary tube to a secondary tube increases the risk of preanalytical errors. For analytes that are stable in whole blood for 24 h or more, it seems that there is no benefit to centrifuge before transport. <s> BIB010

Transportation of blood samples is a major part of the preanalytical pathway and can be crucial in delaying laboratory results to the clinicians BIB002 . Due to increasing demands from the clinicians, it is important to have the primary venous specimen collection tubes transported to the laboratory as fast as possible to be able to measure analytes within the established stability time BIB009 BIB010 BIB006 BIB008 BIB003 BIB001 to maintain fast turnaround times and ensure sample integrity BIB007 BIB005 . Furthermore, the transportation process itself must be firmly controlled to assure that the analyses requested are not affected by temperature, agitation, or other physical or biological influences BIB004 . Finally, transportation logistics must be well arranged to satisfy sample flow from hospital wards as well as, e.g. general practitioners (GPs), while at the same time matching sample reception with the workflow at the laboratory in terms of numbers of samples, peak arrival during daytime, etc. Any delay from blood collection to centrifugation and analysis or any deviation from standard transportation conditions could potentially alter laboratory results and subsequently have a negative impact on patient safety . Of note, the impact of transportation time and conditions on test results is highly dependent on the analytes requested, time to centrifugation as well as on the analytical method applied. Multiple sample stability studies are available for separated as well as whole blood samples, though not necessarily for every analyte BIB009 BIB010 BIB006 BIB008 BIB003 BIB001 . All these aspects make sample transportation a complex, challenging and often overlooked task that needs thorough planning and resources taking into account a profound understanding of which preanalytical factors that could alternate the test results. This information has to be cascaded amongst all parties involved in the transportation process (e.g. clinicians, nurses, phlebotomists, carriers, etc.). The purpose of this review is to outline the options available for this task and to emphasize the preanalytical aspects that need consideration in this process.

Sample transportation – an overview <s> In-house sample transportation <s> BACKGROUND ::: The pneumatic tube system (PTS) has been implicated in inducing haemolysis. It is not known whether certain sample types are more susceptible to haemolysis than others. We assessed the level of haemolysis in commonly used sample types in the clinical biochemistry department when transported through the PTS. ::: ::: ::: METHOD ::: Blood was collected in pairs for different sample types and sent to the laboratory via the pneumatic tube or delivered by a porter. Haemolysis indices were measured spectrophotometrically and compared for each pair of sample type. ::: ::: ::: RESULTS ::: Our results suggest that plain serum samples are more susceptible to haemolysis than the other sample types when sent through our PTS (P <0.0001). Compared with serum with gel samples, plain serum samples are more prone to haemolysis (P <0.001). This suggests that gel may confer some protection against haemolysis. ::: ::: ::: CONCLUSION ::: Different hospitals will have varying system configurations and use different sample types. We recommend that each hospital investigate their own system to assess whether haemolysis is a recurring problem in any of the sample types transported. <s> BIB001 </s> Sample transportation – an overview <s> In-house sample transportation <s> BACKGROUND ::: Pneumatic tube transport of blood samples reduces turnaround times and labour. However, the preanalytical effects on new clinical chemistry parameters and instruments are not fully known. The aim of this study was to evaluate the effect of pneumatic tube transport on haematology and coagulation parameters, including platelet function with PFA-100, and global coagulation with a thromboelastograph. ::: ::: ::: METHODS ::: Paired venous blood samples from healthy volunteers were obtained before and after 1 week of treatment with acetylsalicylic acid. One sample was transported by pneumatic tube transport, while the other remained in the laboratory. ::: ::: ::: RESULTS ::: No preanalytical effect of pneumatic tube transport could be seen for most haematology and coagulation parameters, as well as analysis with PFA-100. For the thromboelastographic analysis, time to clot formation was shorter (-16%, p=0.037) in the transported samples. Treatment with acetylsalicylic acid had no effect on the majority of the test results. ::: ::: ::: CONCLUSIONS ::: Pneumatic tube transport does not introduce preanalytical errors when transporting samples for analysis of routine haematology, coagulation parameters and platelet function with the PFA-100. We recommend manual transport of samples for analysis with thromboelastographic techniques. <s> BIB002 </s> Sample transportation – an overview <s> In-house sample transportation <s> Abstract Objectives Due to the consolidation of laboratory testing facilities, there is an increasing need for systems able to assure quality and safety in biological sample transportation, although little evidence on this aspect is available in literature. Design and methods An integrated system for sample transportation, implemented and monitored over a five-year period by our team, consists of secondary and tertiary containers, a device for temperature and time recording, and a system manager allowing the acceptance or rejection of biological samples through the immediate visualization and validation of registered data. Results Data collected between 2009 and October 2011, after a preliminary phase for optimizing the temperature inside the containers, demonstrated the frequency of transportations at an acceptable temperature ( 25 °C) had decreased by ~ 80%. Conclusions The integrated system and related operating instructions allow improvement in the quality of sample transportation over time. <s> BIB003 </s> Sample transportation – an overview <s> In-house sample transportation <s> Pneumatic tube systems are widely used in hospitals. Advantages are high speed and rapid availability of the samples. However, the transportation by pneumatic tube promotes haemolysis. Haemolysis interferes with many spectrophotometric assays and is a common problem in clinical laboratories. The haemolysis index (HI) as a semi-quantitative representation of the level of haemolysis was compared in unpaired tube-transported and hand-delivered routine lithium heparinate plasma samples (n = 1368 and n = 837, respectively). Additionally, the HI distribution was measured in lithium heparinate plasma samples with a HI above the threshold value of 20 and in paired serum samples after transportation by pneumatic tube system. HI values above 20 can interfere with the selected assays: Creatine kinase (CK), creatine kinase-MB (CK-MB) and alanine aminotransferase (ALT) activities. These parameters were determined to demonstrate how haemolysis affects the results. 17.5% of the tube-transported plasma samples and 2.6% of the hand-delivered plasma samples had a HI above 20. The median HI in pneumatic tube-transported lithium heparinate plasma was 85 and 33 in the paired serum samples. The median HI difference between paired plasma and serum was 46. Blood samples in lithium heparinate tubes may be substantially more susceptible to haemolysis by pneumatic tube transportation than serum tube samples. Although our results cannot be universally applied to laboratories with different pneumatic tube systems, it is recommended that each laboratory evaluate carefully the degree of haemolysis after the transportation by the own pneumatic tube system and in terms of the sample type. <s> BIB004 </s> Sample transportation – an overview <s> In-house sample transportation <s> BACKGROUND ::: Pneumatic tube systems (PTSs) are convenient methods of patient sample transport in medical centers, but excessive acceleration force and time/distance traveled in the PTS have been correlated with increased blood-sample hemolysis. We investigated the utility of smartphones for monitoring of PTS-related variables. ::: ::: ::: METHODS ::: Smartphones were sent through the PTS from several hospital locations. Each smartphone used 2 apps as data-loggers to record force of acceleration vs time. To relate the smartphone data to sample integrity, blood samples were collected from 5 volunteers, and hemolysis of the samples was analyzed after they were transported by hand or via 1 of 2 PTS routes. Increased sample hemolysis as measured by plasma lactate dehydrogenase (LD) was also related to the amount of transport in the PTS. ::: ::: ::: RESULTS ::: The smartphones showed higher duration of forceful acceleration during transport through 1 of the 2 PTS routes, and the increased duration correlated with significant increases in hemolysis (H)-index and plasma LD. In addition, plasma LD showed a positive linear relationship with number of shock forces experienced during transport through the PTS. ::: ::: ::: CONCLUSIONS ::: Smartphones can monitor PTS variables that cause sample hemolysis. This provides an accessible method for investigating specific PTS routes in medical centers. <s> BIB005 </s> Sample transportation – an overview <s> In-house sample transportation <s> Introduction: Timeliness of laboratory results is crucial to patient care and outcome. Monitoring turnaround times (TAT), especially for emergency tests, is important to measure the effectiveness and efficiency of laboratory services. Laboratory-based clinical audits reveal opportunities for im proving quality. Our aim was to identify the most critical steps causing a high TAT for cerebrospinal fluid (CSF) chemistry analysis in our laboratory. Materials and methods: A 6-month retrospective audit was performed. The duration of each operational phase across the laboratory work flow was examined. A process-mapping audit trail of 60 randomly selected requests with a high TAT was conducted and reasons for high TAT were tested for significance. Results: A total of 1505 CSF chemistry requests were analysed. Transport of samples to the laboratory was primarily responsible for the high average TAT (median TAT = 170 minutes). Labelling accounted for most delays within the laboratory (median TAT = 71 minutes) with most delays occurring after regular work hours (P < 0.05). CSF chemistry requests without the appropriate number of CSF sample tubes were significantly associated with delays in movement of samples from the labelling area to the technologist’s work station (caused by a preference for microbiological testing prior to CSF chemistry). Conclusion: A laboratory-based clinical audit identified sample transportation, work shift periods and use of inappropriate CSF sample tubes as drivers of high TAT for CSF chemistry in our laboratory. The results of this audit will be used to change pre-analytical practices in our laboratory with the aim of improving TAT and customer satisfaction. <s> BIB006 </s> Sample transportation – an overview <s> In-house sample transportation <s> To the Editor: ::: ::: During the last decade the interest in using circulating cell-free DNA (ccfDNA)1 as a biomarker for the early diagnosis, prognosis, and monitoring of a variety of clinical conditions has increased rapidly. In daily clinical practice, it has been shown that preanalytical variables such as blood sampling, handling-procedurevariations, and blood storage conditions affect the ccfDNA concentration in human plasma (1). In this respect, optimal handling of blood samples between venipuncture and plasma preparation may be considered essential for downstream processes (1, 2) since time delay and/or agitation could trigger the release of genomic DNA from white blood cells (1). ::: ::: To reduce laboratory turnaround time, sample transport by pneumatic tube systems (PTS) has been established in many hospitals. However, rapid accelerations and decelerations during the PTS transport may cause shear stress and can potentially affect various laboratory measurements for coagulation, clinical chemistry, hematology, or blood gas analyses (3). ::: ::: At present, the literature lacks studies investigating the impact of PTS transport on quantitative ccfDNA measurements. Therefore, we compared ccfDNA plasma concentrations between paired whole … <s> BIB007 </s> Sample transportation – an overview <s> In-house sample transportation <s> Abstract Background Core laboratory (CL), as a new business model, facilitates consolidation and integration of laboratory services to enhance efficiency and reduce costs. This study evaluates the impact of total laboratory automation system (TLA), electric track vehicle (ETV) system and auto-verification (AV) of results on overall turnaround time (TAT) (phlebotomy to reporting TAT: PR-TAT) within a CL setting. Methods Mean, median and percentage of outlier (OP) for PR-TAT were compared for pre- and post-CL eras using five representative tests based on different request priorities. Comparison studies were also carried out on the intra-laboratory TAT (in-lab to reporting TAT: IR-TAT) and the delivery TAT (phlebotomy to in-lab TAT: PI-TAT) to reflect the efficiency of the TLA (both before and after introducing result AV) and ETV systems respectively. Results Median PR-TATs for the urgent samples were reduced on average by 16% across all representative analytes. Median PR-TATs for the routine samples were curtailed by 51%, 50%, 49%, 34% and 22% for urea, potassium, thyroid stimulating hormone (TSH), complete blood count (CBC) and prothrombin time (PT) respectively. The shorter PR-TAT was attributed to a significant reduction of IR-TAT through the TLA. However, the median PI-TAT was delayed when the ETV was used. Application of various AV rules shortened the median IR-TATs for potassium and urea. However, the OP of PR-TAT for the STAT requests exceeding 60 min were all higher than those from the pre-CL era. Conclusions TLA and auto-verification rules help to efficiently manage substantial volumes of urgent and routine samples. However, the ETV application as it stands shows a negative impact on the PR-TAT. <s> BIB008 </s> Sample transportation – an overview <s> In-house sample transportation <s> BACKGROUND ::: Pneumatic transportation systems (PTSs) are increasingly used for transportation of blood samples to the core laboratory. Many studies have investigated the impact of these systems on different types of analyses, but to elucidate whether PTSs in general are safe for transportation of blood samples, existing literature on the subject was systematically assessed. ::: ::: ::: METHODS ::: A systematic literature review was conducted following the preferred reporting items for systematic reviews and metaanalyses (PRISMA) Statement guidelines to gather studies investigating the impact of PTS on analyses in blood samples. Studies were extracted from PubMed and Embase. The search period ended November 2016. ::: ::: ::: RESULTS ::: A total of 39 studies were retrieved. Of these, only 12 studies were conducted on inpatients, mainly intensive care unit patients. Blood gases, hematology, and clinical chemistry were well investigated, whereas coagulation, rotational thromboelastometry, and platelet function in acutely ill patients were addressed by only 1 study each. Only a few parameters were affected in a clinically significant way (clotting time parameter in extrinsic system thromboelastometry, pO2 in blood gas, multiplate analysis, and the hemolysis index). ::: ::: ::: CONCLUSIONS ::: Owing to their high degree of heterogeneity, the retrieved studies were unable to supply evidence for the safety of using PTSs for blood sample transportation. In consequence, laboratories need to measure and document the actual acceleration forces in their existing PTS, instituting quality target thresholds for these measurements such as acceleration vector sums. Computer modeling might be applied to the evaluation of future PTS installations. With the increasing use of PTS, a harmonized, international recommendation on this topic is warranted. <s> BIB009 </s> Sample transportation – an overview <s> In-house sample transportation <s> Diagnostic pathology activities are largely based on fixation of tissues in 4% formaldehyde, which has recently been re-classified as a carcinogenic compound and banned in several countries. Hospitals that do not have in-house pathology services need to send surgical and biopsy specimens to referral centers. These are generally transferred in liquid containers, under suboptimal safety conditions, as accidental spillage of potentially dangerous substances may occur. A safe, innovative, two-step procedure for pathology sample transportation is presented. Formalin-fixed material from ten surgical cases was dissected (including surrogate biopsies) and preserved in liquid-free plastic bags under vacuum for up to 30 days and subsequently processed for conventional histology, several immunohistochemical markers, and molecular tests (e.g., RAS mutation). The data were compared with the corresponding routine analyses. Formalin-fixed specimens after up to 30 days under vacuum storage gave equivalent results compared to standard histopathological slides and molecular tests, regarding both hematoxylin-eosin, immuno-stained slides and also nucleic acid extracted for molecular tests. The proposal of under-vacuum sealing pathology specimens that were previously formalin fixed can be adopted to transfer liquid-free biopsy and surgical specimens to referral pathology services. In fact, it is easy to perform, less expensive (both plastic bags and domestic-type vacuum chamber machines are at affordable costs), and above all is fully safe and adequate in the pre-analytical processing of pathology specimens. <s> BIB010

For in-house samples, transportation can be manually or automated: Manual transportation by trolley or by hand (e.g. by the phlebotomist or the clinician drawing a STAT sample) are both very reliable as no technical instruments are involved and also, it is an accustomed transportation form. It is however slow, and if emergency samples are carried by hand the time spent with sample transportation is time lost to perform other tasks. Another hurdle is the tendency to gather samples at the wards in order to send them in batches to the laboratory. This perhaps saves transportation labour, but the first sample collected may have been waiting for hours at the ward before it is transported to the lab. Also, if samples are received batch-wise, it is not possible to maintain a "first-in-first-out" (FIFO) process, as the laboratory does not know in which order the samples were collected. As a consequence of these drawbacks, an increasing number of hospitals use automated transportation systems, which can be pneumatic tube transportation systems (PTS) BIB009 or, e.g. an electric track vehicle BIB008 . Both fit nicely into automation of the in-house sample reception at the laboratory; it also tends to make the wards send their samples in more "real time", enabling the laboratory to maintain the FIFO principle, and as a natural consequence, it facilitates speed, unidirectional flow and high throughput. However, it also poses obstacles to be addressed by the laboratory professionals: It is important to note that the test request information has to be available at the laboratory when the sample arrives, which makes electronic requisition more flow-efficient than paper requisitions. Separated transport of samples and test request information in terms of paper requisitions leads to further delay in the processing of the ordered analysis, which often leads to resampling and is the cause of laborious tracing efforts for both the wards and the laboratory. Thus, if an automated sample reception system is to work, test requests must be electronically handled. A more basic problem is the obvious risk of processing the serum samples too quick, i.e. not allowing time for clot formation before centrifugation. This results in fibrin issues, which can be either interference with a number of analyses or a fibrin clot blocking the sample pipetting needle leading to a false result and possibly "down time" of the analyser -in both instances a new blood sample is needed causing delay in the clinical process. This can be solved by programming a halt in the sample flow for these serum tubes, but one must however first be aware of the problem. Another main issue is the increased risk of haemolysis shown by a number of studies (e.g. BIB004 BIB001 ) and also demonstrated by use of a smart phone BIB005 . But as revealed by a systematic review there is no general evidence for the safety of using PTS for blood sample transportation BIB009 . This is mainly due to the high degree of heterogeneity of the retrieved studies, but also because the local, physical PTS arrangement impacts the sample transported in an unpredictable way: Automated PTS are very individually constructed, and conditions concerning the usability (how easy can the samples be loaded?), the g-force impact (due to speed, twist and turns), the actual physical impact on the sample (which along with the mentioned g impact also includes bumps of the tubes during "pit stops" during transport and at the arrival at the laboratory), and finally the temperature (if the PTS goes underground or outside a building) are all parameters that requires a closer scrutiny to assure sample integrity during automated transportation. So ideally, laboratories must measure and document the actual acceleration forces in their existing PTS. For this, use of G-loggers to measure acceleration vector sums has been suggested, and the laboratory should accordingly institute quality target thresholds for these values BIB003 . Another more simple possibility is the use of daily test results from a particular PTS, where, e.g. median potassium values can be used to monitor changes in the impact on blood samples. Single potassium values will thus not have any relevance, while running potassium median values will provide sufficient monitoring; again, the laboratory must establish a target threshold and define what action should be taken if exceeded. Haemolysis index values for each PTS can also be used, but notably, samples can be affected without haemolysis, e.g. if potassium leaks from leukocytes or platelets during transportation , a preanalytical error that will remain undetected by the haemolysis index. A significant preanalytical effect on samples transported by PTS has for instance been noted for thromboelastographic analyses. For these analyses, manual transport of samples is therefore recommended BIB002 , and in the case of platelet function testing best practice is to avoid any transportation at all and if possible, perform the blood collection at the test site. In order to continuously minimize turnaround times in all clinical settings, a forthcoming challenge will be the use of automated PTS for all different kinds of sample materials, e.g. paediatric samples, urine, cerebrospinal fluid BIB006 , free DNA measurement BIB007 , culture media and even pathology specimens BIB010 .

Sample transportation – an overview <s> External sample transportation (from general practitioners or external laboratories) <s> The aim of the study was to simulate physical conditions occurring during mail transport and to study how these conditions influence the stability of material for clinical chemistry analysis. Periods of transport at constant temperatures did not result in big changes. We found that all serum and plasma constituents tested should be considered stable for 4 days at -20 degrees C. Except for the coagulation factors studied all other serum and plasma constituents were stable also at 4 degrees C and 20 degrees C. Blood cells were studied only at 4 degrees C at which temperature they were stable for 4 days except for thrombocytes. The condition that caused the largest effect was changing of temperature. Temperature gradient from ambient to 50 degrees C influenced the concentrations of all cells in blood, of coagulation factors in plasma and of enzymes in serum. The constituents were remarkably stable during shaking. The results agreed with those of a field study on the effect of mail transport by Berg et al. [3]. <s> BIB001 </s> Sample transportation – an overview <s> External sample transportation (from general practitioners or external laboratories) <s> BACKGROUND ::: Safety transport boxes are increasingly used to ship laboratory specimens but there is little information on their capacity to maintain suitable transportation temperatures. ::: ::: ::: MATERIALS AND METHODS ::: Inner temperature was assessed using a commercially available transport box during an 8-h transportation period in the heat. ::: ::: ::: RESULTS ::: Temperature stability was unsatisfactory during approximately 64% of the transportation time (i.e., from 125 to 450 min). ::: ::: ::: CONCLUSIONS ::: Transport boxes might be unsuitable for shipping specimens over long periods. <s> BIB002 </s> Sample transportation – an overview <s> External sample transportation (from general practitioners or external laboratories) <s> Abstract Objectives Due to the consolidation of laboratory testing facilities, there is an increasing need for systems able to assure quality and safety in biological sample transportation, although little evidence on this aspect is available in literature. Design and methods An integrated system for sample transportation, implemented and monitored over a five-year period by our team, consists of secondary and tertiary containers, a device for temperature and time recording, and a system manager allowing the acceptance or rejection of biological samples through the immediate visualization and validation of registered data. Results Data collected between 2009 and October 2011, after a preliminary phase for optimizing the temperature inside the containers, demonstrated the frequency of transportations at an acceptable temperature ( 25 °C) had decreased by ~ 80%. Conclusions The integrated system and related operating instructions allow improvement in the quality of sample transportation over time. <s> BIB003 </s> Sample transportation – an overview <s> External sample transportation (from general practitioners or external laboratories) <s> Objectives ::: We addressed the stability of biological samples in prolonged drone flights by obtaining paired chemistry and hematology samples from 21 adult volunteers in a single phlebotomy event-84 samples total. ::: ::: ::: Methods ::: Half of the samples were held stationary, while the other samples were flown for 3 hours (258 km) in a custom active cooling box mounted on the drone. After the flight, 19 chemistry and hematology tests were performed. ::: ::: ::: Results ::: Seventeen analytes had small or no bias, but glucose and potassium in flown samples showed an 8% and 6.2% bias, respectively. The flown samples (mean, 24.8°C) were a mean of 2.5°C cooler than the stationary samples (mean, 27.3°C) during transportation to the flight field as well as during the flight. ::: ::: ::: Conclusions ::: The changes in glucose and potassium are consistent with the magnitude and duration of the temperature difference between the flown and stationary samples. Long drone flights of biological samples are feasible but require stringent environmental controls to ensure consistent results. <s> BIB004 </s> Sample transportation – an overview <s> External sample transportation (from general practitioners or external laboratories) <s> • The aim of the paper sended to the journal, is to stress the need to accurately control the transport conditions of biological samples, being the time and the temperature during transportation the major issue responsible to loss of samples quality. <s> BIB005

External samples are transported in a variety of ways, e.g. by car BIB001 , in boxes BIB002 , using drones BIB004 , or even by planes and trains. If the samples are properly protected from temperature deviation and agitation, none of these transportation forms should affect the samples significantly BIB003 . It is however crucial to monitor these conditions -and also a demand according to the ISO15189 accreditation that most laboratories carry. The transportation of venous blood samples from outside the hospital setting is often costly and logistically challenging. Flaws in tube labelling, packaging and transport cause delays and increased costs with several stakeholders involved. The transportation routines should be described and agreed upon, also including what to do when these routines are deviated from. A number of technical support systems for this, including software and intelligent transportation boxes with GPS and temperature loggers BIB005 , are already on the market, and such features will for certain be an important part of an improved preanalytical quality assurance in the future.

Sample transportation – an overview <s> Transport control systems <s> Blood specimens from primary care centres are normally transported to central laboratories by mail. This necessitates centrifugation and separation, especially since the potassium ion concentration in whole blood changes during storage at ambient temperature. Thus, because of the growing awareness of and concern for pre-analytical contributions to the uncertainty of measurements, we investigated 27 components and their stability under controlled temperature conditions from 17 to 238C. We found that storage of whole blood can be prolonged by up to 8–12 h for all components examined, including potassium ions, when stored at 20"0.28C. We conclude that this opens the possibility for establishing a pick-up service, by which whole blood specimens stored at 20–218C can be collected at the doctor’s office, making centrifugation, separation and mailing superfluous. In addition, the turn-around time from sample drawing to reporting the analytical result would be shortened. After investments in thermostatted boxes and logistics, the system could reduce costs for transporting blood samples from general practice centres to central laboratories. <s> BIB001 </s> Sample transportation – an overview <s> Transport control systems <s> Background: In many countries and especially in Scandinavia, blood samples drawn in primary healthcare are sent to a hospital laboratory for analysis. The samples are exposed to various conditions regarding storage time, storage temperature and transport form. As these factors can have a severe impact on the quality of results, we wanted to study which combination of transport conditions could fulfil our predefined goals for maximum allowable error. Methods: Samples from 406 patients from nine general practitioners (GPs) in two Danish counties were sent to two hospitals for analyses, during two periods (winter and summer). Transport conditions (mail, courier pick-up, or brought to hospital by public coach), storage time, storage temperature and centrifugation requirements were different in the two counties. Results were tested for deviation from a "0-sample", the blood sample taken, centrifuged and separated at the doctor's office within 45-60 min. This sample was considered as the best estimate of a comparison value. Results: The pre-set quality goals were fulfilled for all the investigated components for samples transported to hospital by courier either as whole blood or as "on gel" after centrifugation, as long as the samples were stored at 20-25°C and centrifuged/analysed within 5-6 h. A total of 4% of the samples sent by mail had mismatched identity, probably due to plasma being transferred to a new tube. Conclusions: Samples can be sent as unprocessed anticoagulated whole blood if the above mentioned conditions are met. There is no need for centrifugation in the primary sector. Neither mailing of samples with plasma "on gel" nor public transport by coach bus fulfil our analytical goals. <s> BIB002 </s> Sample transportation – an overview <s> Transport control systems <s> Abstract Objectives Due to the consolidation of laboratory testing facilities, there is an increasing need for systems able to assure quality and safety in biological sample transportation, although little evidence on this aspect is available in literature. Design and methods An integrated system for sample transportation, implemented and monitored over a five-year period by our team, consists of secondary and tertiary containers, a device for temperature and time recording, and a system manager allowing the acceptance or rejection of biological samples through the immediate visualization and validation of registered data. Results Data collected between 2009 and October 2011, after a preliminary phase for optimizing the temperature inside the containers, demonstrated the frequency of transportations at an acceptable temperature ( 25 °C) had decreased by ~ 80%. Conclusions The integrated system and related operating instructions allow improvement in the quality of sample transportation over time. <s> BIB003 </s> Sample transportation – an overview <s> Transport control systems <s> Tietz textbook of clinical chemistry and molecular diagnostics / , Tietz textbook of clinical chemistry and molecular diagnostics / , کتابخانه دیجیتالی دانشگاه علوم پزشکی و خدمات درمانی شهید بهشتی <s> BIB004 </s> Sample transportation – an overview <s> Transport control systems <s> diagnostic samples:from the patient to the laboratory , diagnostic samples:from the patient to the laboratory , کتابخانه دانشگاه علوم پزشکی و خدمات درمانی بوشهر <s> BIB005 </s> Sample transportation – an overview <s> Transport control systems <s> BACKGROUND ::: The stability limit of an analyte in a biological sample can be defined as the time required until a measured property acquires a bias higher than a defined specification. Many studies assessing stability and presenting recommendations of stability limits are available, but differences among them are frequent. The aim of this study was to classify and to grade a set of bibliographic studies on the stability of five common blood measurands and subsequently generate a consensus stability function. ::: ::: ::: METHODS ::: First, a bibliographic search was made for stability studies for five analytes in blood: alanine aminotransferase (ALT), glucose, phosphorus, potassium and prostate specific antigen (PSA). The quality of every study was evaluated using an in-house grading tool. Second, the different conditions of stability were uniformly defined and the percent deviation (PD%) over time for each analyte and condition were scattered while unifying studies with similar conditions. ::: ::: ::: RESULTS ::: From the 37 articles considered as valid, up to 130 experiments were evaluated and 629 PD% data were included (106 for ALT, 180 for glucose, 113 for phosphorus, 145 for potassium and 85 for PSA). Consensus stability equations were established for glucose, potassium, phosphorus and PSA, but not for ALT. ::: ::: ::: CONCLUSIONS ::: Time is the main variable affecting stability in medical laboratory samples. Bibliographic studies differ in recommedations of stability limits mainly because of different specifications for maximum allowable error. Definition of a consensus stability function in specific conditions can help laboratories define stability limits using their own quality specifications. <s> BIB006

Optimally, transportation should be performed under the same temperature conditions as storage before and after analysis until the analysis quality control has been performed and approved. Ideally, the time and temperature of transport boxes should be logged, which previously has been nicely described BIB003 . At best, temperature data from the transportation phase should be incorporated in the laboratory information system (LIS) in order to facilitate a swift, automated approval procedure when receiving outhouse samples. Such systems are not yet available, but LIS companies should be encouraged to develop such functionalities. There is a large body of evidence on sample stability (e.g. BIB002 BIB001 BIB006 ), but specification and documentation of sample stability under different circumstances and ambient temperatures are outside the scope of this review; to retrieve this information the reader could consult e.g. Tietz' textbook BIB004 or Guder's Diagnostic Samples BIB005 . But in order to judge stability of the analytes, information on specimen collection time, time to centrifugation and analysis time are inevitable and must be presented in the LIS. Samples that are not guaranteed to reach the laboratory the same day must be centrifuged, which necessitates centrifugation capability and educated personnel, e.g. at the GP. This however requires quality assurance of the centrifugation performed as well as of sample aliquoting (if performed). Therefore, sampling of specimens for analytes with short-time stability or demanding storage in plain tubes after centrifugation or freezing prior to transport must be specified in the laboratory's specimen collection guidelines, which according to the ISO15189 must be distributed to external samplers. Just as samples are rejected due to haemolysis, samples with analytes beyond the specified stability time and outside the approved temperature limits should be rejected for analysis. Importantly, routines should be established at the laboratory to avoid similar situations in the future.

Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> A procedure has been developed for measuring visibly different facial movements. The Facial Action Code was derived from an analysis of the anatomical basis of facial movement. The method can be used to describe any facial movement (observed in photographs, motion picture film or videotape) in terms of anatomically based action units. The development of the method is explained, contrasting it to other methods of measuring facial behavior. An example of how facial behavior is measured is provided, and ideas about research applications are discussed. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> Numerous previous studies found monopolar rather than bipolar dimensions of affect (defined as emotion represented in language), but may have included methodological biases against bipolarity. The present study of self-report data (N = 150) on 11 affect scales showed that response format and acquiescence response style significantly shifted correlations between hypothesized opposites away from showing bipolarity. When these biases were taken into account, pleasure was found to be the bipolar opposite of displeasure and arousal of sleepiness. In turn, pleasure-displeasure and degree of arousal formed a twodimensional bipolar space that accounted for almost all of the reliable variance in Thayer's four factors of activation plus a measure of depression. Dominance and submissiveness factors were also included in the study, but invalidity of the scales used precluded any conclusions regarding their bipolarity. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> Abstract Emotions are viewed as having evolved through their adaptive value in dealing with fundamental life-tasks. Each emotion has unique features: signal, physiology, and antecedent events. Each emotion also has characteristics in common with other emotions: rapid onset, short duration, unbidden occurrence, automatic appraisal, and coherence among responses. These shared and unique characteristics are the product of our evolution, and distinguish emotions from other affective phenomena. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> Although consumer research began focusing on emotional response to advertising during the 1980s (Goodstein, Edell, and Chapman Moore. 1990; Burke and Edell, 1989; Aaker, Stayman, and Vezina, 1988; Holbrook and Batra, 1988), perhaps one of the most practical measures of affective response has only recently emerged. Part of the difficulty in developing measures of emotional response stems from the complexity of emotion itself (Plummer and Leckenby, 1985). Researchers have explored several different measurement formats including: verbal self-reports (adjective checklists), physiological techniques, photodecks, and dial-turning instruments. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> The third and fourth books of Cicero's "Tusculan Disputations" deal with the nature and management of human emotion: first grief, then the emotions in general. Cicero presents the insights of Greek philosophers on the subject, reporting the views of Epicurians and Peripatetics and giving a detailed account of the Stoic position, which he himself favours for it's close reasoning and moral earnestness. He describes the Stoic's analysis of grief, their classification of emotions by genus and species, their lists of oddly named character flaws, and by the philosophical debate that develops over the utility of anger in politics and war. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> Little attention has been paid so far to physiological signals for emotion recognition compared to audiovisual emotion channels such as facial expression or speech. This paper investigates the potential of physiological signals as reliable channels for emotion recognition. All essential stages of an automatic recognition system are discussed, from the recording of a physiological dataset to a feature-based multiclass classification. In order to collect a physiological dataset from multiple subjects over many weeks, we used a musical induction method which spontaneously leads subjects to real emotional states, without any deliberate lab setting. Four-channel biosensors were used to measure electromyogram, electrocardiogram, skin conductivity and respiration changes. A wide range of physiological features from various analysis domains, including time/frequency, entropy, geometric analysis, subband spectra, multiscale entropy, etc., is proposed in order to find the best emotion-relevant features and to correlate them with emotional states. The best features extracted are specified in detail and their effectiveness is proven by classification results. Classification of four musical emotions (positive/high arousal, negative/high arousal, negative/low arousal, positive/low arousal) is performed by using an extended linear discriminant analysis (pLDA). Furthermore, by exploiting a dichotomic property of the 2D emotion model, we develop a novel scheme of emotion-specific multilevel dichotomous classification (EMDC) and compare its performance with direct multiclass classification using the pLDA. Improved recognition accuracy of 95\% and 70\% for subject-dependent and subject-independent classification, respectively, is achieved by using the EMDC scheme. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> MAHNOB-HCI is a multimodal database recorded in response to affective stimuli with the goal of emotion recognition and implicit tagging research. A multimodal setup was arranged for synchronized recording of face videos, audio signals, eye gaze data, and peripheral/central nervous system physiological signals. Twenty-seven participants from both genders and different cultural backgrounds participated in two experiments. In the first experiment, they watched 20 emotional videos and self-reported their felt emotions using arousal, valence, dominance, and predictability as well as emotional keywords. In the second experiment, short videos and images were shown once without any tag and then with correct or incorrect tags. Agreement or disagreement with the displayed tags was assessed by the participants. The recorded videos and bodily responses were segmented and stored in a database. The database is made available to the academic community via a web-based system. The collected data were analyzed and single modality and modality fusion results for both emotion recognition and implicit tagging experiments are reported. These results show the potential uses of the recorded modalities and the significance of the emotion elicitation protocol. <s> BIB007 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> We present a multimodal data set for the analysis of human affective states. The electroencephalogram (EEG) and peripheral physiological signals of 32 participants were recorded as each watched 40 one-minute long excerpts of music videos. Participants rated each video in terms of the levels of arousal, valence, like/dislike, dominance, and familiarity. For 22 of the 32 participants, frontal face video was also recorded. A novel method for stimuli selection is proposed using retrieval by affective tags from the last.fm website, video highlight detection, and an online assessment tool. An extensive analysis of the participants' ratings during the experiment is presented. Correlates between the EEG signal frequencies and the participants' ratings are investigated. Methods and results are presented for single-trial classification of arousal, valence, and like/dislike ratings using the modalities of EEG, peripheral physiological signals, and multimedia content analysis. Finally, decision fusion of the classification results from different modalities is performed. The data set is made publicly available and we encourage other researchers to use it for testing their own affective state estimation methods. <s> BIB008 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> Emotion recognition through computational modeling and analysis of physiological signals has been widely investigated in the last decade. Most of the proposed emotion recognition systems require relatively long-time series of multivariate records and do not provide accurate real-time characterizations using short-time series. To overcome these limitations, we propose a novel personalized probabilistic framework able to characterize the emotional state of a subject through the analysis of heartbeat dynamics exclusively. The study includes thirty subjects presented with a set of standardized images gathered from the international affective picture system, alternating levels of arousal and valence. Due to the intrinsic nonlinearity and nonstationarity of the RR interval series, a specific point-process model was devised for instantaneous identification considering autoregressive nonlinearities up to the third-order according to the Wiener-Volterra representation, thus tracking very fast stimulus-response changes. Features from the instantaneous spectrum and bispectrum, as well as the dominant Lyapunov exponent, were extracted and considered as input features to a support vector machine for classification. Results, estimating emotions each 10 seconds, achieve an overall accuracy in recognizing four emotional states based on the circumplex model of affect of 79.29%, with 79.15% on the valence axis, and 83.55% on the arousal axis. <s> BIB009 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> Automatic emotion recognition is one of the most challenging tasks. To detect emotion from nonstationary EEG signals, a sophisticated learning algorithm that can represent high-level abstraction is required. This study proposes the utilization of a deep learning network (DLN) to discover unknown feature correlation between input signals that is crucial for the learning task. The DLN is implemented with a stacked autoencoder (SAE) using hierarchical feature learning approach. Input features of the network are power spectral densities of 32-channel EEG signals from 32 subjects. To alleviate overfitting problem, principal component analysis (PCA) is applied to extract the most important components of initial input features. Furthermore, covariate shift adaptation of the principal components is implemented to minimize the nonstationary effect of EEG signals. Experimental results show that the DLN is capable of classifying three different levels of valence and arousal with accuracy of 49.52% and 46.03%, respectively. Principal component based covariate shift adaptation enhances the respective classification accuracy by 5.55% and 6.53%. Moreover, DLN provides better performance compared to SVM and naive Bayes classifiers. <s> BIB010 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> In this work, we present DECAF —a multimodal data set for dec oding user physiological responses to af fective multimedia content. Different from data sets such as DEAP [15] and MAHNOB-HCI [31] , DECAF contains (1) brain signals acquired using the Magnetoencephalogram (MEG) sensor, which requires little physical contact with the user’s scalp and consequently facilitates naturalistic affective response, and (2) explicit and implicit emotional responses of 30 participants to 40 one-minute music video segments used in [15] and 36 movie clips, thereby enabling comparisons between the EEG versus MEG modalities as well as movie versus music stimuli for affect recognition. In addition to MEG data, DECAF comprises synchronously recorded near-infra-red (NIR) facial videos, horizontal Electrooculogram (hEOG), Electrocardiogram (ECG), and trapezius-Electromyogram (tEMG) peripheral physiological responses. To demonstrate DECAF’s utility, we present (i) a detailed analysis of the correlations between participants’ self-assessments and their physiological responses and (ii) single-trial classification results for valence , arousal and dominance , with performance evaluation against existing data sets. DECAF also contains time-continuous emotion annotations for movie clips from seven users, which we use to demonstrate dynamic emotion prediction. <s> BIB011 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> Stress, anxiety and depression in the workplace are detrimental to human health and productivity with significant financial implications. Recent research in this area has focused on the use of sensor technologies, including smartphones and wearables embedded with physiological and movement sensors. In this work, we explore the possibility of using such devices for mood recognition, focusing on work environments. We propose a novel mood recognition framework that is able to identify five intensity levels for eight different types of moods every two hours. We further present a smartphone app ('HealthyOffice'), designed to facilitate self-reporting in a structured manner and provide our model with the ground truth. We evaluate our system in a small-scale user study where wearable sensing data is collected in an office environment. Our experiments exhibit promising results allowing us to reliably recognize various classes of perceived moods. <s> BIB012 </s> Wearable-Based Affect Recognition—A Review <s> Emotion Models <s> We present ASCERTAIN—a multimodal databa AS e for impli C it p ER sonali T y and A ffect recognit I o N using commercial physiological sensors. To our knowledge, ASCERTAIN is the first database to connect personality traits and emotional states via physiological responses . ASCERTAIN contains big-five personality scales and emotional self-ratings of 58 users along with their Electroencephalogram (EEG), Electrocardiogram (ECG), Galvanic Skin Response (GSR) and facial activity data, recorded using off-the-shelf sensors while viewing affective movie clips. We first examine relationships between users’ affective ratings and personality scales in the context of prior observations, and then study linear and non-linear physiological correlates of emotion and personality. Our analysis suggests that the emotion-personality relationship is better captured by non-linear rather than linear statistics. We finally attempt binary emotion and personality trait recognition using physiological features. Experimental results cumulatively confirm that personality differences are better revealed while comparing user responses to emotionally homogeneous videos, and above-chance recognition is achieved for both affective and personality dimensions. <s> BIB013

In this section emotional models frequently employed in AR literature are detailed. These are grouped into two distinct types: 1. Categorical models: Here different emotions are represented in discrete categories. 2. Dimensional models: Following this approach, emotions are mapped into a multidimensional space, where each of the axis represents a continuous variable. Categorical models date back to ancient Greek and Roman philosophers . Cicero, for instance, distinguished four basic categories of emotions, namely fear, pain, lust and pleasure BIB005 . Darwin [24] also conducted studies on emotions and came to the conclusion that emotions have an evolutionary history and, hence, are shared across cultures. Similar to Darwin , Ekman BIB003 argues that basic emotions are shared across cultures and appear to be universally recognised. Following Ekman and Friesen , six basic emotions can be distinguished: joy, sadness, anger, fear, disgust, and surprise BIB001 . These basic emotions are discrete and have distinct physiological patterns, for example, facial muscle movement. Being able to express basic emotions can be attributed with a number of (evolutionary evolved) physiological and communicative functions: Disgust, for example, is often expressed by a certain facial expression and a wrinkled nose. On a physiological level this facial expression limits inhalation of malodorous particles. On the communicative level, this distinct facial expression, performed for instance as reaction to rotten food, has the potential to warn others. In 1980, Plutchik developed another taxonomy to classify discrete emotions. The so-called 'wheel of emotions' comprises of eight primary emotions: grief, amazement, terror, admiration, ecstasy, vigilance, rage, and loathing. Following Plutchik , the primary emotions mix and give rise to more complex emotions. In addition, emotions are expressed at different intensity levels. In the domain of wearable AR, categorical models were for instance used by Zenonos et al. BIB012 . In their study the authors presented an approach to distinguish eight different emotions and moods (excited, happy, calm, tired, bored, sad, stressed, and angry). The above presented model of basic emotions is not unquestioned and one point of criticism is that some languages do not have words for certain basic emotions BIB002 . According to Reference BIB007 , in Polish, for instance, there is no exact translation for the English word disgust. Dimensional models where emotions are mapped into a multidimensional space, mitigate this shortcoming. The first dimensional approach dates back to Wundt , who describes momentary emotions as a single point in a three-dimensional space . Wundt's emotional space is spanned by the pleasure-displeasure, excitement-inhibition and tension-relaxation axes. At the end of the 1970s, Russell BIB002 postulated a two-dimensional model, namely the circumplex model (see Figure 1a) . This model has been very impactful and in the circumplex model, affective states are represented as discrete points in a two-dimensional space, spanned by the axes valence and arousal. The valence axis indicates the perception on how positive or negative the current affective state is. On the arousal axis, the state is rated in terms of the activation level, for example, how energised or enervated one feels. The four quadrants of the circumplex model (low arousal/low valence (LALV), low arousal/high valence (LAHV), high arousal/low valence (HALV) and high arousal/high valence (HAHV)) can be attributed with sad, relaxed, angry, and happy. By adding further orthogonal axes, for example, dominance, the circumplex model is easily extended. In AR, the circumplex model and its variants are frequently employed BIB006 BIB008 BIB009 BIB011 . Using the Self-Assessment Manikins (SAM) BIB004 , the circumplex model can easily be assessed. These Manikins offer an easy graphical way for subjects to report their current affective state (see Figure 1b ). In addition, the SAM are easily understood across cultures, due to their simple graphical representation. Another possible reason for the popularity of dimensional models in AR might arise from a machine learning (ML) point of view. The (at least two) independent axes of the circumplex model offer an interesting set of different classification tasks: The valence and arousal axes, for instance, can be binned into multiclass classification problems, for example, low/medium/high arousal or valence. In addition, posing classification problems based on the four quadrants named above is a frequently pursued task in AR, see for instance References BIB006 BIB013 . BIB009 ; (b) Exemplary Self-Assessment Manikins BIB004 , used to generate labels in the valence-arousal space. Adapted from Jirayucharoensak et al. BIB010 .

Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Stress is “the nonspecific response of the body to any demand made upon it,” that is, the rate at which we live at any one moment. All living beings are constantly under stress and anything, pleasant or unpleasant, that speeds up the intensity of life, causes a temporary increase in stress, the wear and tear exerted upon the body. A painful blow and a passionate kiss can be equally stressful. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Objective. —This article defines stress and related concepts and reviews their historical development. The notion of a stress system as the effector of the stress syndrome is suggested, and its physiologic and pathophysiologic manifestations are described. A new perspective on human disease states associated with dysregulation of the stress system is provided. Data Sources. —Published original articles from human and animal studies and selected reviews. Literature was surveyed utilizing MEDLINE and the Index Medicus . Study Selection. —Original articles from the basic science and human literature consisted entirely of controlled studies based on verified methodologies and, with the exception of the most recent studies, replicated by more than one laboratory. Many of the basic science and clinical studies had been conducted in our own laboratories and clinical research units. Reviews cited were written by acknowledged leaders in the fields of neurobiology, endocrinology, and behavior. Data Extraction. —Independent extraction and cross-referencing by the authors. Data Synthesis. —Stress and related concepts can be traced as far back as written science and medicine. The stress system coordinates the generalized stress response, which takes place when a stressor of any kind exceeds a threshold. The main components of the stress system are the corticotropin-releasing hormone and locus ceruleus-norepinephrine/autonomic systems and their peripheral effectors, the pituitary-adrenal axis, and the limbs of the autonomic system. Activation of the stress system leads to behavioral and peripheral changes that improve the ability of the organism to adjust homeostasis and increase its chances for survival. There has been an exponential increase in knowledge regarding the interactions among the components of the stress system and between the stress system and other brain elements involved in the regulation of emotion, cognitive function, and behavior, as well as with the axes responsible for reproduction, growth, and immunity. This new knowledge has allowed association of stress system dysfunction, characterized by sustained hyperactivity and/or hypoactivity, to various pathophysiologic states that cut across the traditional boundaries of medical disciplines. These include a range of psychiatric, endocrine, and inflammatory disorders and/or susceptibility to such disorders. Conclusions. —We hope that knowledge from apparently disparate fields of science and medicine integrated into a working theoretical framework will allow generation and testing of new hypotheses on the pathophysiology and diagnosis of, and therapy for, a variety of human illnesses reflecting systematic alterations in the principal effectors of the generalized stress response. We predict that pharmacologic agents capable of altering the central apparatus that governs the stress response will be useful in the treatment of many of these illnesses. ( JAMA . 1992;267:1244-1252) <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Objective: This article presents a new formulation of the relationship between stress and the processes leading to disease. It emphasizes the hidden cost of chronic stress to the body over long time periods, which act as a predisposing factor for the effects of acute, stressful life events. It also presents a model showing how individual differences in the susceptibility to stress are tied to individual behavioral responses to environmental challenges that are coupled to physiologic and pathophysiologic responses. Data Sources: Published original articles from human and animal studies and selected reviews. Literature was surveyed using MEDLINE. Data Extraction: Independent extraction and cross-referencing by us. Data Synthesis: Stress is frequently seen as a significant contributor to disease, and clinical evidence is mounting for specific effects of stress on immune and cardiovascular systems. Yet, until recently, aspects of stress that precipitate disease have been obscure. The concept of homeostasis has failed to help us understand the hidden toll of chronic stress on the body. Rather than maintaining constancy, the physiologic systems within the body fluctuate to meet demands from external forces, a state termed allostasis . In this article, we extend the concept of allostasis over the dimension of time and we define allostatic load as the cost of chronic exposure to fluctuating or heightened neural or neuroendocrine response resulting from repeated or chronic environmental challenge that an individual reacts to as being particularly stressful. Conclusions: This new formulation emphasizes the cascading relationships, beginning early in life, between environmental factors and genetic predispositions that lead to large individual differences in susceptibility to stress and, in some cases, to disease. There are now empirical studies based on this formulation, as well as new insights into mechanisms involving specific changes in neural, neuroendocrine, and immune systems. The practical implications of this formulation for clinical practice and further research are discussed. (Arch Intern Med. 1993;153:2093-2101) <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Abdominal obesity, anxiety, and depression have been found to cluster in several studies. To further characterize these associations, the following study was performed. In a population of 51-year-old men (N = 284), measurements of obesity (body mass index [BMI]) and body fat distribution (waist to hip ratio [WHR] and sagittal trunk recumbent diameter [D]) were analyzed in relation to dexamethasone (0.5 mg) inhibition of cortisol secretion, measured as salivary cortisol. Symptoms of anxiety and depression were defined by a validated questionnaire. Furthermore, testosterone, insulin-like growth factor-I (IGF-I), insulin, glucose, and serum lipid levels were measured. Twenty-five men (8.8%) had symptoms of anxiety and depression. BMI, WHR, and D correlated negatively with testosterone, except for BMI in the anxio-depressive (ADP) group. IGF-I showed no significant relationship. Furthermore, fasting insulin and the insulin to glucose ratio correlated positively and high-density lipoprotein (HDL) cholesterol correlated negatively with BMI, WHR, and D in the total study population and in the subgroups. Total and low-density lipoprotein (LDL) cholesterol showed no significant relationships. Correlation coefficients tended to be higher in ADP men. Dexamethasone inhibition showed a negative significant relationship with BMI (rho = -.47, P = .025), WHR (borderline, rho = -.37, P = .086), and D (rho = -.43, P = .046) only in the ADP group. Comparing the ADP group versus the group without anxio-depression (ADO) and high or low BMI (P = .008), WHR (P = .026), and D (P = .012) showed blunted dexamethasone inhibition only in ADP men with high anthropometric measurements. These findings suggest there is a subgroup with elevated BMI, WHR, and D in whom a blunted dexamethasone response is found associated with traits of anxiety and depression, conditions characterized by such an abnormality. The reason for the association might be insufficient control of cortisol secretion, followed by visceral fat accumulation. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> R. E. Thayer (1989) proposed 2 types of activation: energetic arousal (awake-tired) and tense arousal (tense-calm). This view has been challenged by claims that energetic arousal and tense arousal are mixtures of valence and a single activation dimension. The authors present a direct test of this hypothesis by computing the correlation between the residuals of energetic arousal and tense arousal after removing the shared variance with valence. Whereas the valence activation hypothesis predicts a strong positive correlation between the 2 residuals, the authors found that it was not significantly different from 0. This finding reaffirms the view of energetic arousal and tense arousal as 2 distinct types of activation. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> This essay describes the evolution of stress as a medical scientific idea. Claude Bernard, Walter B. Cannon and Hans Selye provided key founding concepts for the current view. Bernard introduced the idea of the internal environment bathing cells—the milieu interieur—maintained by continual compensatory changes of bodily functions. Cannon coined the word, “homeostasis,” referring to a set of acceptable ranges of values for internal variables. Cannon taught that threats to homeostasis evoke activation of the sympathoadrenal system as a functional unit. Selye defined stress as a state characterized by a uniform response pattern, regardless of the particular stressor, that could lead to long-term pathologic changes. “Allostasis” was introduced as a concept in recognition that there is no single ideal set of steady-state conditions in life; instead, setpoints and other response criteria change continuously. Stress is now viewed neither as a perturbation nor a stereotyped response pattern but as a condition cha... <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> We have designed a stress management biofeedback mobile service for everyday use, aiding users to reflect on both positive and negative patterns in their behavior. To do so, we embarked on a complex multidisciplinary design journey, learning that: detrimental stress results from complex processes related to e.g. the subjective experience of being able to cope (or not) and can therefore not be measured and diagnosed solely as a bodily state. We learnt that it is difficult, sometimes impossible, to make a robust analysis of stress symptoms based on biosensors worn outside the laboratory environment they were designed for. We learnt that rather than trying to diagnose stress, it is better to mirror short-term stress reactions back to them, inviting their own interpretations and reflections. Finally, we identified several experiential qualities that such an interface should entail: ambiguity and openness to interpretation, interactive history of prior states, fluency and aliveness. <s> BIB007 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Repeated exposures to psychological stress can lead to or worsen diseases of slow accumulation such as heart diseases and cancer. The main challenge in addressing the growing epidemic of stress is a lack of robust methods to measure a person's exposure to stress in the natural environment. Periodic self-reports collect only subjective aspects, often miss stress episodes, and impose significant burden on subjects. Physiological sensors provide objective and continuous measures of stress response, but exhibit wide between-person differences and are easily confounded by daily activities (e.g., speaking, physical movements, coffee intake, etc.). In this paper, we propose, train, and test two models for continuous prediction of stress from physiological measurements captured by unobtrusive, wearable sensors. The first model is a physiological classifier that predicts whether changes in physiology represent stress. Since the effect of stress may persist in the mind longer than its acute effect on physiology, we propose a perceived stress model to predict perception of stress. It uses the output of the physiological classifier to model the accumulation and gradual decay of stress in the mind. To account for wide between-person differences, both models self-calibrate to each subject. Both models were trained using data collected from 21 subjects in a lab study, where they were exposed to cognitive, physical, and social stressors representative of that experienced in the natural environment. Our physiological classifier achieves 90% accuracy and our perceived stress model achieves a median correlation of 0.72 with self-reported rating. We also evaluate the perceived stress model on data collected from 17 participants in a two-day field study, and find that the average rating of stress obtained from our model has a correlation of 0.71 with that obtained from periodic self-reports. <s> BIB008 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Stress can have long term adverse effects on individuals' physical and mental well-being. Changes in the speech production process is one of many physiological changes that happen during stress. Microphones, embedded in mobile phones and carried ubiquitously by people, provide the opportunity to continuously and non-invasively monitor stress in real-life situations. We propose StressSense for unobtrusively recognizing stress from human voice using smartphones. We investigate methods for adapting a one-size-fits-all stress model to individual speakers and scenarios. We demonstrate that the StressSense classifier can robustly identify stress across multiple individuals in diverse acoustic environments: using model adaptation StressSense achieves 81% and 76% accuracy for indoor and outdoor environments, respectively. We show that StressSense can be implemented on commodity Android phones and run in real-time. To the best of our knowledge, StressSense represents the first system to consider voice based stress detection and model adaptation in diverse real-life conversational situations using smartphones. <s> BIB009 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Emotion recognition through computational modeling and analysis of physiological signals has been widely investigated in the last decade. Most of the proposed emotion recognition systems require relatively long-time series of multivariate records and do not provide accurate real-time characterizations using short-time series. To overcome these limitations, we propose a novel personalized probabilistic framework able to characterize the emotional state of a subject through the analysis of heartbeat dynamics exclusively. The study includes thirty subjects presented with a set of standardized images gathered from the international affective picture system, alternating levels of arousal and valence. Due to the intrinsic nonlinearity and nonstationarity of the RR interval series, a specific point-process model was devised for instantaneous identification considering autoregressive nonlinearities up to the third-order according to the Wiener-Volterra representation, thus tracking very fast stimulus-response changes. Features from the instantaneous spectrum and bispectrum, as well as the dominant Lyapunov exponent, were extracted and considered as input features to a support vector machine for classification. Results, estimating emotions each 10 seconds, achieve an overall accuracy in recognizing four emotional states based on the circumplex model of affect of 79.29%, with 79.15% on the valence axis, and 83.55% on the arousal axis. <s> BIB010 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Stress remains a significant social problem for individuals in modern societies. This paper presents a machine learning approach for the automatic detection of stress of people in a social situation by combining two sensor systems that capture physiological and social responses. We compare the performance using different classifiers including support vector machine, AdaBoost, and k-nearest neighbour. Our experimental results show that by combining the measurements from both sensor systems, we could accurately discriminate between stressful and neutral situations during a controlled Trier social stress test (TSST). Moreover, this paper assesses the discriminative ability of each sensor modality individually and considers their suitability for real time stress detection. Finally, we present an study of the most discriminative features for stress detection. <s> BIB011 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB012 </s> Wearable-Based Affect Recognition—A Review <s> Stress Models <s> Affect recognition aims to detect a person's affective state based on observables, with the goal to e.g. improve human-computer interaction. Long-term stress is known to have severe implications on wellbeing, which call for continuous and automated stress monitoring systems. However, the affective computing community lacks commonly used standard datasets for wearable stress detection which a) provide multimodal high-quality data, and b) include multiple affective states. Therefore, we introduce WESAD, a new publicly available dataset for wearable stress and affect detection. This multimodal dataset features physiological and motion data, recorded from both a wrist- and a chest-worn device, of 15 subjects during a lab study. The following sensor modalities are included: blood volume pulse, electrocardiogram, electrodermal activity, electromyogram, respiration, body temperature, and three-axis acceleration. Moreover, the dataset bridges the gap between previous lab studies on stress and emotions, by containing three different affective states (neutral, stress, amusement). In addition, self-reports of the subjects, which were obtained using several established questionnaires, are contained in the dataset. Furthermore, a benchmark is created on the dataset, using well-known features and standard machine learning methods. Considering the three-class classification problem ( baseline vs. stress vs. amusement ), we achieved classification accuracies of up to 80%,. In the binary case ( stress vs. non-stress ), accuracies of up to 93%, were reached. Finally, we provide a detailed analysis and comparison of the two device locations ( chest vs. wrist ) as well as the different sensor modalities. <s> BIB013

In everyday life, stress or being stressed are terms used to describe the feeling of being under pressure. Stress is commonly elicited by an external and/or internal stimulus called stressor. However, from a scientific point of view, stress is primarily a physiological response. At the beginning of the 20th century, Cannon coined the terms homeostasis and "fight or flight" response. Homeostasis describes a balanced state of the organism where its physiological parameters stay within an acceptable range (e.g., a body temperature of 37 • C). Following Reference , both physiological and psychological stimuli can pose threats to homeostasis. Stressors can be seen as threats, disrupting homeostasis. In order to maintain homeostasis, even under extreme conditions, feedback loops (e.g., a fight or flight response) are triggered . In the 1970s, Selye BIB001 defined stress to be/result in a 'nonspecific response of the body to any demand upon it'. Following this definition, 'nonspecific' refers to a shared set of responses triggered regardless of the nature of the stressor, for example, physical or psychological. Recent stress models, for instance McEwen and Stellar BIB003 , incorporate multiple effectors and advocate that the stress response is to some degree specific. The stress response is mainly influenced by two aspects: first, the stressor itself and, second, the organism's perceived ability to cope with the posed threat BIB006 . Depending on the coping ability of the organism and estimated chances for success, eustress (positive outcome) and distress (negative outcome) are distinguished BIB009 . Eustress can have a positive (motivating) effect, while distress is perceived to be hindering (feeling worried or anxious). In order to illustrate this the following example can be used: Assume a person has to take an exam. Here, this exam represents an external stressor and the body reacts with a physiological stress response, for example, by increasing the blood glucose level. If the person feels well prepared for the exam and is looking forward to the challenge ahead, this can be interpreted as eustress. In contrast, if the person is not well prepared and feels like failing the exam, this can result in distress. Considering wearable stress recognition, distinguishing between eustress and distress is a largely unsolved problem due to the lack of adequate physiological indicators. However, long-term stress in general is associated with many severe health implications ranging from troubled sleeping and headaches to an increased risk for cardiovascular diseases BIB003 BIB002 BIB004 . Due to these severe side effects of long-term stress, the detection of stress is a frequent task in AR-Mozos et al. BIB011 , Plarre et al. BIB008 , Schmidt et al. BIB013 , for instance target binary stress recognition tasks (stress versus no stress) and Gjoreski et al. BIB012 aimed at distinguishing different levels of stress (no stress versus low stress versus high stress). Above different emotion and stress models were summarised. Although stress is not an emotion, a link between dimensional models and stress is readily established: Following Sanches et al. BIB007 , a direct link between stress and arousal can be drawn. Valenza et al. BIB010 maps stress into the high arousal/negative valence (quadrant II) of the circumplex model (see Figure 1a) . Following Thayer and later Schimmack and Reisenzein BIB005 , the arousal dimension of the 'classical circumplex' model can be split into tense arousal (stressed-relaxed) and energetic arousal (sleepy-active). According to Schimmack and Reisenzein BIB005 , this split is justified by the observation that only the energetic arousal component is influenced by the sleep-wake cycle. Considering the wearable affect and stress recognition literature, a recent study conducted by Mehrotra et al. uses this three-dimensional emotion model (valence, tense arousal and energetic arousal) to investigate correlation and causation between emotional states and cell phone interaction.

Wearable-Based Affect Recognition—A Review <s> Affective States and Their Physiological Indicators <s> Four experiments were conducted to determine whether voluntarily produced emotional facial configurations are associated with differentiated patterns of autonomic activity, and if so, how this might be mediated. Subjects received muscle-by-muscle instructions and coaching to produce facial configurations for anger, disgust, fear, happiness, sadness, and surprise while heart rate, skin conductance, finger temperature, and somatic activity were monitored. Results indicated that voluntary facial activity produced significant levels of subjective experience of the associated emotion, and that autonomic distinctions among emotions: (a) were found both between negative and positive emotions and among negative emotions, (b) were consistent between group and individual subjects' data, (c) were found in both male and female subjects, (d) were found in both specialized (actors, scientists) and nonspecialized populations, (e) were stronger when the voluntary facial configurations most closely resembled actual emotional expressions, and (f) were stronger when experience of the associated emotion was reported. The capacity of voluntary facial activity to generate emotion-specific autonomic activity: (a) did not require subjects to see facial expressions (either in a mirror or on an experimenter's face), and (b) could not be explained by differences in the difficulty of making the expressions or by differences in concomitant somatic activity. DESCRIPTORS: Facial action, Emotion, Autonomic activity during emotion, Emotion-specific autonomic activity. The experiments in this report are relevant to two major theoretical issues. The first, whether there are different patterns of autonomic nervous system (ANS) activity for different emotions, is one of psychophysiology's oldest. The second, under­ standing how voluntary facial activity can generate emotion-specific autonomic activity, has a much shorter history, but has potentially important im­ plications for emotion theory. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Affective States and Their Physiological Indicators <s> This manuscript discusses the physiology of the autonomic nervous system (ANS). The following topics are presented: regulation of activity; efferent pathways; sympathetic and parasympathetic divisions; neurotransmitters, their receptors and the termination of their activity; functions of the ANS; and the adrenal medullae. In addition, the application of this material to the practice of pharmacy is of special interest. Two case studies regarding insecticide poisoning and pheochromocytoma are included. The ANS and the accompanying case studies are discussed over 5 lectures and 2 recitation sections during a 2-semester course in Human Physiology. The students are in the first-professional year of the doctor of pharmacy program. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Affective States and Their Physiological Indicators <s> ‘‘What is an emotion?’’ William James’s seminal paper in Mind (1884) proposed the idea that physiological and behavioral responses precede subjective experience in emotions that are marked by ‘‘distinct bodily expression.’’ This notion has broadly inspired the investigation of emotion-specific autonomic nervous system activity, a research topic with great longevity. The trajectory of this literature is traced through its major theoretical challenges from the Cannon–Bard, activation, and Schachter– Singer theories, through its rich empirical history in the field of psychophysiology. Although these studies are marked by various findings, the overall trend of the research supports the notion of autonomic specificity for basic emotions. The construct of autonomic specificity continues to influence a number of core theoretical issues in affective science, such as the existence of basic or ‘natural kinds’ of emotion, the structure of affective space, the cognition–emotion relationship, and the function of emotion. Moreover, James’s classic paper, which stimulated the emergence of psychology from philosophy and physiology in the latter nineteenth century, remains a dynamic force in contemporary emotion research. <s> BIB003

Affective states and physiological changes are clearly linked, for example, if someone cracks a good joke we laugh or at least smile. With this physiological response we express amusement. Negative emotional states have even stronger physiological indicators. For instance, when being afraid or anxious one might start sweating, get a dry mouth, or feel sick. Stress was characterised primarily as a physiological response to a stimulus, see Section 2.3. The most severe physiological reaction to a stressor is the so called 'fight or flight' response . During this response the body prepares for a severe action, like fight or flight, releasing a mixture of hormones, like cortisol and adrenaline. This leads, for instance, to an increased breathing/heart rate, pupil dilation and muscle tension. The induced physiological responses are quite distinct and are a good example for the link between affective states and physiological changes. Above the link between affective states and physiological responses was established using examples. The direction/causality, for example, do affective states cause physiological changes or vice versa, is still an open research question: At the end of the 19th century postulated, that physiological changes precede emotions and that emotions arise from these changes. This is best illustrated considering the following example: Picture someone encountering a gigantic poisonous spider. After this encounter, the heart rate and the activity of the sweat glands of the subject would increase. Following the James-Lange-Theory, these physiological changes are not symptoms of fear/disgust but rather involuntary physiological responses. According to James these physiological responses, become an emotion/feeling, like fear/disgust, once a cognitive evaluation occurred. Hence, the subject could describe the process as "I feel afraid, because I have a racing heart". This theory is supported, for instance, by experiments conducted by Levenson et al. BIB001 , who found evidence that performing voluntary facial muscle movements exhibit similar changes in peripheral physiology as if the corresponding emotion is experienced. For instance, when the subjects were asked to make an angry face the heart rate was found to increase. This theory, of course, is not unchallenged. Following common sense, a stimulus is perceived, it elicits an feeling and the physiological responses are triggered. Hence, the subject could describe the process as "I have a racing heart, because I'm afraid of the poisonous spider". Following the Cannon-Bard-Theory, the perceived stimulus is processed in the brain and then the physiological response and affective states arise simultaneously BIB003 . Hence, the subject could describe the process as "The spider makes me feel afraid and I have a racing heart". The debate outlined above is, from a theoretical point of view, very interesting. However, it is out of scope of this review. Wearable-based AR, utilizes these affect-related changes in physiology. Affective states occur spontaneously and are accompanied by certain physiological pattern. These physiological responses are hard or even impossible to control for humans. The autonomic nervous system (ANS) directs these unconscious actions of the organism. Hence, the ANS plays a key role in directing the physiological response to an external (e.g., event) or internal (e.g., thought) affective stimulus. The ANS has two major branches: the sympathetic nervous system (SNS) and the parasympathetic nervous system (PNS). In Table 1 , the key contributions of the SNS and PNS are displayed. As the SNS is mainly associated with the 'fight or flight' response, an increased activity of the SNS indicates high arousal states. In other words, the main function of the SNS is to provide energy by increasing a number of physiological parameters (e.g., respiration rate, glucose level, etc.). The PNS, in contrast, regulates the 'rest and digest' functions BIB002 . Table 1 . Major functions of the sympathetic nervous system and parasympathetic nervous system.

Wearable-Based Affect Recognition—A Review <s> Sympathetic Nervous System (SNS) <s> Little attention has been paid so far to physiological signals for emotion recognition compared to audiovisual emotion channels such as facial expression or speech. This paper investigates the potential of physiological signals as reliable channels for emotion recognition. All essential stages of an automatic recognition system are discussed, from the recording of a physiological dataset to a feature-based multiclass classification. In order to collect a physiological dataset from multiple subjects over many weeks, we used a musical induction method which spontaneously leads subjects to real emotional states, without any deliberate lab setting. Four-channel biosensors were used to measure electromyogram, electrocardiogram, skin conductivity and respiration changes. A wide range of physiological features from various analysis domains, including time/frequency, entropy, geometric analysis, subband spectra, multiscale entropy, etc., is proposed in order to find the best emotion-relevant features and to correlate them with emotional states. The best features extracted are specified in detail and their effectiveness is proven by classification results. Classification of four musical emotions (positive/high arousal, negative/high arousal, negative/low arousal, positive/low arousal) is performed by using an extended linear discriminant analysis (pLDA). Furthermore, by exploiting a dichotomic property of the 2D emotion model, we develop a novel scheme of emotion-specific multilevel dichotomous classification (EMDC) and compare its performance with direct multiclass classification using the pLDA. Improved recognition accuracy of 95\% and 70\% for subject-dependent and subject-independent classification, respectively, is achieved by using the EMDC scheme. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Sympathetic Nervous System (SNS) <s> As is known for centuries, humans exhibit an electrical profile. This profile is altered through various psychological and physiological processes, which can be measured through biosignals; e.g., electromyography (EMG) and electrodermal activity (EDA). These biosignals can reveal our emotions and, as such, can serve as an advanced man-machine interface (MMI) for empathic consumer products. However, such a MMI requires the correct classification of biosignals to emotion classes. This chapter starts with an introduction on biosignals for emotion detection. Next, a state-of-the-art review is presented on automatic emotion classification. Moreover, guidelines are presented for affective MMI. Subsequently, a research is presented that explores the use of EDA and three facial EMG signals to determine neutral, positive, negative, and mixed emotions, using recordings of 21 people. A range of techniques is tested, which resulted in a generic framework for automated emotion classification with up to 61.31% correct classification of the four emotion classes, without the need of personal profiles. Among various other directives for future research, the results emphasize the need for parallel processing of multiple biosignals. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Sympathetic Nervous System (SNS) <s> Autonomic nervous system (ANS) activity is viewed as a major component of the emotion response in many recent theories of emotion. Positions on the degree of specificity of ANS activation in emotion, however, greatly diverge, ranging from undifferentiated arousal, over acknowledgment of strong response idiosyncrasies, to highly specific predictions of autonomic response patterns for certain emotions. A review of 134 publications that report experimental investigations of emotional effects on peripheral physiological responding in healthy individuals suggests considerable ANS response specificity in emotion when considering subtypes of distinct emotions. The importance of sound terminology of investigated affective states as well as of choice of physiological measures in assessing ANS reactivity is discussed. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Sympathetic Nervous System (SNS) <s> Chronic stress is endemic to modern society. However, as it is unfeasible for physicians to continuously monitor stress levels, its diagnosis is nontrivial. Wireless body sensor networks offer opportunities to ubiquitously detect and monitor mental stress levels, enabling improved diagnosis, and early treatment. This article describes the development of a wearable sensor platform to monitor a number of physiological correlates of mental stress. We discuss tradeoffs in both system design and sensor selection to balance information content and wearability. Using experimental signals collected from the wearable sensor, we describe a selected number of physiological features that show good correlation with mental stress. In particular, we propose a new spectral feature that estimates the balance of the autonomic nervous system by combining information from the power spectral density of respiration and heart rate variability. We validate the effectiveness of our approach on a binary discrimination problem when subjects are placed under two psychophysiological conditions: mental stress and relaxation. When used in a logistic regression model, our feature set is able to discriminate between these two mental states with a success rate of 81% across subjects. <s> BIB004

Parasympathetic Nervous System (PNS) • associated with 'fight or flight' • associated with 'rest and digest' • pupils dilate • pupils constrict • decreased salivation and digestion • increased salivation and digestion • increased heart and respiration rate • decreased heart and respiration rate • increased electrodermal activity • increased muscle activity • adrenalin and glucose release The interplay of sympathetic nervous system (SNS) and parasympathetic nervous system (PNS) is best illustrated considering the cardiovascular system. In reaction to a potential threat, the SNS increases the heart rate (HR). Once the threat is over, the PNS reduces the HR, bringing it back to normal BIB004 . A common measure to quantify the interaction of SNS and PNS is the heart rate variability (HRV). The HRV is defined as the variation in the beat-to-beat intervals. An increased/decreased HRV indicates increased activity of the PNS/SNS, respectively. As a result, the HRV is a rather simple but efficient measure to quantify the contributions of the PNS/SNS. Hence in related work, the HRV is employed to detect stress BIB004 . Changes in the electrodermal activity (EDA) are another simple but effective measure to assess the SNS activity, too. This is due to the fact, that changes in EDA are governed by the SNS BIB004 . Hence, following Dawson et al. the EDA is particularly sensitive to high arousal states, like fear, anger, and stress. EDA has two main components, namely the skin conductance level (SCL) and the skin conductance response (SCR). The SCL, also known as tonic component, represents a slowly varying baseline conductivity. In contrast, the SCR, also called phasic component, refers to peaks in the EDA signal. For most other vital parameters, the contributions of PNS and SNS are more interleaved. Hence, their responses are less specific. Nevertheless, also considering respiration and muscle activity, certain patterns can be attributed to different affective states. For instance, the respiration rate increases and becomes more irregular when a subject is more aroused BIB001 . Later, in Section 5.2, a detailed description of physiological features will be provided. As outlined above, the SNS contributions to high arousal states are quite distinct. In a recent meta analysis, Kreibig BIB003 investigated the specificity of the ANS response to certain affective states. A subset of these findings, including two positive and two negative affective states, is presented in Table 2 . Considering for instance anger: a majority of the analysed studies showed that it coincides with an increased heart rate (HR), skin conductance level (SCL), number of skin conductance response (SCR)s and a higher breathing rate. Since anger represents a high arousal state, governed by the SNS, these reactions were expected. Non-crying sadness was found to decrease HR, SCL and number of SCRs, while increasing the respiration rate. In the circumplex model (see Figure 1a) , sadness is mapped into the third quadrant (low valence, low arousal). Hence, the arousal level is expected to drop which is confirmed by Table 2 . Amusement and happiness are both positive affective states with a similar arousal level. Hence, it is not surprising that they have a similar physiological fingerprint. For more details, we refer the reader to Kreibig BIB003 . The findings of Kreibig BIB003 suggest that affective states have certain physiological fingerprints which are to some degree specific. These findings are promising, as they indicate that distinguishing affective states based on physiological indicators is feasible. However, in the context of wearable-based AR, the following aspects should be considered BIB002 : 1. Physiological measures are indirect measures of an affective state. 2. Emotions are subjective but physiological data are not. 3. Although some physiological patterns are shared across subjects, individual responses to a stimulus can differ strongly. 4. Multimodal affect detecting systems reach higher accuracies than unimodal systems . 5. The physiological signal quality often suffers from noise, induced by motion artefacts and misplacement.

Wearable-Based Affect Recognition—A Review <s> Frequently Employed Sensors <s> Autonomic nervous system (ANS) activity is viewed as a major component of the emotion response in many recent theories of emotion. Positions on the degree of specificity of ANS activation in emotion, however, greatly diverge, ranging from undifferentiated arousal, over acknowledgment of strong response idiosyncrasies, to highly specific predictions of autonomic response patterns for certain emotions. A review of 134 publications that report experimental investigations of emotional effects on peripheral physiological responding in healthy individuals suggests considerable ANS response specificity in emotion when considering subtypes of distinct emotions. The importance of sound terminology of investigated affective states as well as of choice of physiological measures in assessing ANS reactivity is discussed. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Frequently Employed Sensors <s> MAHNOB-HCI is a multimodal database recorded in response to affective stimuli with the goal of emotion recognition and implicit tagging research. A multimodal setup was arranged for synchronized recording of face videos, audio signals, eye gaze data, and peripheral/central nervous system physiological signals. Twenty-seven participants from both genders and different cultural backgrounds participated in two experiments. In the first experiment, they watched 20 emotional videos and self-reported their felt emotions using arousal, valence, dominance, and predictability as well as emotional keywords. In the second experiment, short videos and images were shown once without any tag and then with correct or incorrect tags. Agreement or disagreement with the displayed tags was assessed by the participants. The recorded videos and bodily responses were segmented and stored in a database. The database is made available to the academic community via a web-based system. The collected data were analyzed and single modality and modality fusion results for both emotion recognition and implicit tagging experiments are reported. These results show the potential uses of the recorded modalities and the significance of the emotion elicitation protocol. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Frequently Employed Sensors <s> Photoplethysmography (PPG) technology has been used to develop small, wearable, pulse rate sensors. These devices, consisting of infrared light-emitting diodes (LEDs) and photodetectors, offer a simple, reliable, low-cost means of monitoring the pulse rate noninvasively. Recent advances in optical technology have facilitated the use of high-intensity green LEDs for PPG, increasing the adoption of this measurement technique. In this review, we briefly present the history of PPG and recent developments in wearable pulse rate sensors with green LEDs. The application of wearable pulse rate monitors is discussed. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Frequently Employed Sensors <s> Because it is more convenient to measure photoplethysmography (PPG) than ECG, PPG is supposed as a surrogate of ECG for heart rate variability (HRV) analysis. In this study, we measured the spectrum of pulse rate variability (PRV) from PPG and its square coherence spectrum with HRV from ECG before and after exercise. The results showed that the spectrum of PRV corresponds almost well with HRV, especially for the healthy subjects at rest. However, stimuli conditions such as exercise will decrease the correlation, especially for the high frequency (HF) component. It indicates that whether PRV can be used as an alternative to HRV depends on the applications and conditions. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Frequently Employed Sensors <s> With the worldwide growth of mobile wireless technologies, healthcare services can be provided at anytime and anywhere. Usage of wearable wireless physiological monitoring system has been extensively increasing during the last decade. These mobile devices can continuously measure e.g. the heart activity and wirelessly transfer the data to the mobile phone of the patient. One of the significant restrictions for these devices is usage of energy, which leads to requiring low sampling rate. This article is presented in order to investigate the lowest adequate sampling frequency of ECG signal, for achieving accurate enough time domain heart rate variability (HRV) parameters. For this purpose the ECG signals originally measured with high 5 kHz sampling rate were down-sampled to simulate the measurement with lower sampling rate. Down-sampling loses information, decreases temporal accuracy, which was then restored by interpolating the signals to their original sampling rates. The HRV parameters obtained from the ECG signals with lower sampling rates were compared. The results represent that even when the sampling rate of ECG signal is equal to 50 Hz, the HRV parameters are almost accurate with a reasonable error. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Frequently Employed Sensors <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB006

This section provides an overview of the sensor modalities frequently employed in wearable-based AR. The clear aim of AR is to find robust methods assessing the affective state of a user in everyday life. Hence, a major goal is to use sensor setups which are minimally intrusive and pose only minor limitations to the mobility of the user. As detailed in Tables 1 and 2 , physiological changes in the cardiac system and electrodermal activity are key indicators for affective states. Therefore, most studies utilise these modalities. Nevertheless, sensors measuring other physiological parameter, like respiration or muscle activity, can also contain valuable information on the affective state of a person BIB001 . Table 3 lists the most relevant sensors, grouped according to their placement on the human body. Below, each of the listed modalities is discussed, detailing advantages and limitations. In order to assess the heart rate (HR), heart rate variability (HRV) and other parameters related to the cardiac cycle, the electrocardiogram (ECG) serves as gold standard. For a standard three-point ECG, three electrodes are placed on the subject's torso, measuring the depolarisation and repolarisation of the heart tissue during each heartbeat. ECG samples are collected with frequencies up to 1024 Hz. However, when acquired with such high frequency the signal can be downsampled to 256 Hz without loss of information BIB002 . Furthermore, experiments of Mahdiani et al. BIB005 indicate that a 50 Hz ECG sampling rate is sufficient to obtain HRV-related parameters with a reasonable error. Using photoplethysmogram (PPG) also provides information about the cardiac cycles. In contrast to ECG, PPG utilises an optical method: The skin voxel, beneath the sensor, is illuminated by a LED and a photodiode measures the amount of backscattered light. Alternatively if the detector is on the opposite side of the respective body part (e.g., fingertip or earlobe), the amount of transmitted light is measured. Hence, the cardiac cycle is captured by the PPG signal, where the pulsatile part of the PPG signal reflects the pulsatile component in arterial blood flow BIB003 . Data obtained from a PPG sensor tends to be noisier than ECG data. This is due to artefacts caused by motion, light from external sources, or different skin tones, which influence the reflection/absorption properties of the skin. PPG sensors can be attached to the ear, wrist BIB006 or the finger tip BIB004 of subjects. The PPG modality finds broad application in fitness trackers and smartwatches, which can be attributed to the small form factor of the sensory setup. Typical sampling rates of PPG devices are below 100 Hz.

Wearable-Based Affect Recognition—A Review <s> Electrodermal Activity <s> A provisional standard method of measuring tonic skin conductance (SCL) and GSR (SCR) is advocated, using a constant-voltage method for which circuits are provided useable with Beckman, Grass, and other common polygraphs. A standard electrode methodology is also presented. The problem of units of measurement is considered in detail with an analysis of the so-called Law of Initial Values. Methods are given for correcting both tonic SC and SCRs for individual differences in their respective ranges of variation and the purpose and relative advantages of these range-correction methods are discussed. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal Activity <s> This paper presents methods for collecting and analyzing physiological data during real-world driving tasks to determine a driver's relative stress level. Electrocardiogram, electromyogram, skin conductance, and respiration were recorded continuously while drivers followed a set route through open roads in the greater Boston area. Data from 24 drives of at least 50-min duration were collected for analysis. The data were analyzed in two ways. Analysis I used features from 5-min intervals of data during the rest, highway, and city driving conditions to distinguish three levels of driver stress with an accuracy of over 97% across multiple drivers and driving days. Analysis II compared continuous features, calculated at 1-s intervals throughout the entire drive, with a metric of observable stressors created by independent coders from videotapes. The results show that for most drivers studied, skin conductivity and heart rate metrics are most closely correlated with driver stress level. These findings indicate that physiological signals can provide a metric of driver stress in future cars capable of physiological monitoring. Such a metric could be used to help manage noncritical in-vehicle information systems and could also provide a continuous measure of how different road and traffic conditions affect drivers. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal Activity <s> Chronic stress is endemic to modern society. However, as it is unfeasible for physicians to continuously monitor stress levels, its diagnosis is nontrivial. Wireless body sensor networks offer opportunities to ubiquitously detect and monitor mental stress levels, enabling improved diagnosis, and early treatment. This article describes the development of a wearable sensor platform to monitor a number of physiological correlates of mental stress. We discuss tradeoffs in both system design and sensor selection to balance information content and wearability. Using experimental signals collected from the wearable sensor, we describe a selected number of physiological features that show good correlation with mental stress. In particular, we propose a new spectral feature that estimates the balance of the autonomic nervous system by combining information from the power spectral density of respiration and heart rate variability. We validate the effectiveness of our approach on a binary discrimination problem when subjects are placed under two psychophysiological conditions: mental stress and relaxation. When used in a logistic regression model, our feature set is able to discriminate between these two mental states with a success rate of 81% across subjects. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal Activity <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal Activity <s> In affective computing (AC) field studies it is impossible to obtain an objective ground truth. Hence, self-reports in form of ecological momentary assessments (EMAs) are frequently used in lieu of ground truth. Based on four paradigms, we formulate practical guidelines to increase the accuracy of labels generated via EMAs. In addition, we detail how these guidelines were implemented in a recent AC field study of ours. During our field study, 1081 EMAs were collected from 10 subjects over a duration of 148 days. Based on these EMAs, we perform a qualitative analysis of the effectiveness of our proposed guidelines. Furthermore, we present insights and lessons learned from the field study. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal Activity <s> A growing number of pervasive systems integrate emotion recognition capabilities with the aim of fostering and promoting human health and wellbeing. In this paper, we argue that enhancing these systems with the ability to detect laughter episodes automatically would improve their effectiveness. This is because laughter is one of the most expressive behavioral cues for positive emotions. However, the existing approaches for laughter recognition rely on the use of obtrusive devices or of video and audio cues and have thus limited applicability in real-world settings. To overcome this limitation, we evaluate the feasibility of a novel, multi-modal approach to recognize laughter episodes using physiological and body movement data gathered with unobtrusive, wrist-worn devices. To assess the performance of our method, we collect an extensive data set of laughter episodes, which we also make publicly available. Our results show that laughter episodes can be distinguished from non-laughter episodes with an accuracy of 81%. Further, we demonstrate that the signatures left by laughter episodes on physiological and body-movement data differ significantly from those caused by slightly intense motions or cognitive load tasks. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal Activity <s> Physiological responses to emotions play a vital role in the field of emotion recognition. Machine-learning models implemented in wristbands or wearables, already exploit unique patterns in physiological responses to provide information about humans emotional states. However, such responses are commonly interfered and overlapped by physical activities, posing a challenge for emotion recognition “in-the-wild”. In this paper, we address this challenge by investigating new features based on the linear regression line and machine-learning models for emotion recognition. We triggered emotions through audio samples and recorded physiological responses from 18 participants before and while performing physical activities. We trained models with the least strenuous physical activity (sitting) and tested with the remaining, more strenuous ones. For three different emotion categories, we achieved classification accuracies up to 67%. Considering individual activities and participants, we achieve up to 73% classification accuracy, indicating the viability of emotion recognition models and features non-sensitive to interferences caused by physical activities. <s> BIB007

The electrodermal activity (EDA) is commonly measured at locations with a high density of sweat glands, for example, palm/finger BIB003 or feet BIB002 . Alternative locations to measure an EDA signal are the wrist BIB004 or the torso BIB005 . In order to assess EDA, the resistance between two electrodes is measured. From a technical point of view, EDA data is recorded employing either constant-current (measuring skin resistance) or constant-voltage systems (recording skin conductance) . However, due to the more linear relationship between the skin conductance and the number of active sweat glands, Lykken and Venables BIB001 argues strongly for a direct measure of the skin conductance using constant-voltage systems . In recent AR research the Empatica E4 is a frequently employed device to collect EDA data BIB004 BIB005 BIB006 BIB007 . Having the form factor of a smartwatch, the E4 samples the EDA signal at 4 Hz, which is sufficient to distinguish the SCR from the SCL. Although the EDA is strongly influenced by the SNS, external parameters such as humidity, temperature, or the physical activity have a strong influence.

Wearable-Based Affect Recognition—A Review <s> Electromyogram <s> Spontaneous pericranial electromyographic (EMG) activity is generally small and is contaminated by strong low-frequency artifacts. High-pass filtering should suppress artifacts but affect EMG signal power only minimally. In 24 subjects who performed a warned simple reaction time task, the optimal high-pass cut-off frequency was examined for nine different pericranial muscles. From four experimental conditions (visual and auditory reaction signals combined with hand and foot responses), 1-min EMG recordings were selected (bandwidth: 0.4-512 Hz) and divided into 60 1-s data segments. These segments were high-pass filtered, the -3-dB cut-off frequency varying from 5 to 90 Hz, and subjected to power spectral analysis. Optimal high-pass filter frequencies were determined for the mean power spectra based on visual estimation or comparison with a theoretical spectrum of the artifact-free EMG signal. The optimal frequencies for the different muscles varied between 15 and 25 Hz and were not influenced by stimulus or response modality. For all muscles, a low-pass filter frequency between 400 and 500 Hz was appropriate. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Electromyogram <s> Little attention has been paid so far to physiological signals for emotion recognition compared to audiovisual emotion channels such as facial expression or speech. This paper investigates the potential of physiological signals as reliable channels for emotion recognition. All essential stages of an automatic recognition system are discussed, from the recording of a physiological dataset to a feature-based multiclass classification. In order to collect a physiological dataset from multiple subjects over many weeks, we used a musical induction method which spontaneously leads subjects to real emotional states, without any deliberate lab setting. Four-channel biosensors were used to measure electromyogram, electrocardiogram, skin conductivity and respiration changes. A wide range of physiological features from various analysis domains, including time/frequency, entropy, geometric analysis, subband spectra, multiscale entropy, etc., is proposed in order to find the best emotion-relevant features and to correlate them with emotional states. The best features extracted are specified in detail and their effectiveness is proven by classification results. Classification of four musical emotions (positive/high arousal, negative/high arousal, negative/low arousal, positive/low arousal) is performed by using an extended linear discriminant analysis (pLDA). Furthermore, by exploiting a dichotomic property of the 2D emotion model, we develop a novel scheme of emotion-specific multilevel dichotomous classification (EMDC) and compare its performance with direct multiclass classification using the pLDA. Improved recognition accuracy of 95\% and 70\% for subject-dependent and subject-independent classification, respectively, is achieved by using the EMDC scheme. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Electromyogram <s> Stress is a growing problem in society and can, amongst others, induce musculoskeletal complaints, related to sustained muscle tension. The ability to measure stress with a wireless system would be useful in the prevention of stress-related health problems. The aim of this experiment was to derive stress levels of subjects from electromyography (EMG) signals of the upper trapezius muscle. Two new stress tests were designed for this study, which aimed at creating circumstances that are similar to work stress. An experiment is described in which EMG signals of the upper trapezius muscle were measured during three different stressful situations. Stress tests included a calculation task (the Norinder test), a logical puzzle task and a memory task, of which the last two were newly designed. The results show significantly higher amplitudes of the EMG signals during stress compared to rest and fewer gaps (periods of relaxation) during stress. Also, mean and median frequencies were significantly lower during stress than during rest. The differences in EMG features between rest and stress conditions indicate that EMG is a useful parameter to detect stress. These results show opportunities for the inclusion of EMG sensors in a wireless system for ambulatory monitoring of stress levels. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Electromyogram <s> We present a multimodal data set for the analysis of human affective states. The electroencephalogram (EEG) and peripheral physiological signals of 32 participants were recorded as each watched 40 one-minute long excerpts of music videos. Participants rated each video in terms of the levels of arousal, valence, like/dislike, dominance, and familiarity. For 22 of the 32 participants, frontal face video was also recorded. A novel method for stimuli selection is proposed using retrieval by affective tags from the last.fm website, video highlight detection, and an online assessment tool. An extensive analysis of the participants' ratings during the experiment is presented. Correlates between the EEG signal frequencies and the participants' ratings are investigated. Methods and results are presented for single-trial classification of arousal, valence, and like/dislike ratings using the modalities of EEG, peripheral physiological signals, and multimedia content analysis. Finally, decision fusion of the classification results from different modalities is performed. The data set is made publicly available and we encourage other researchers to use it for testing their own affective state estimation methods. <s> BIB004

Muscle activity is measured using surface electromyogram (EMG). For this purpose, a pair (or array) of electrodes is attached to the skin above the muscle under consideration. The electrical potential is generated when the muscle cells are activated and the surface electrodes are used to recorded changes in the electric potential. The frequency range of the muscle activity ranges from 15 to 500 Hz BIB001 . Hence, in order to capture the full spectral range, the minimal sampling rate of the EMG modality should be around 1000 Hz. One source of noise in surface EMG are potential changes in adjacent muscles and cardiac activity. Depending on the measurement position, the QRS complex (indicating depolarization of the cardiac ventricles and the following contraction) can cause artefacts which require postprocessing beyond normal filtering. Considering related work in AR literature, EMG electrodes are often placed in the face (e.g., on the zygomaticus major BIB004 ) or on the shoulder (e.g., on the upper trapezius muscle BIB002 BIB004 BIB003 ).

Wearable-Based Affect Recognition—A Review <s> Respiration <s> This paper presents methods for collecting and analyzing physiological data during real-world driving tasks to determine a driver's relative stress level. Electrocardiogram, electromyogram, skin conductance, and respiration were recorded continuously while drivers followed a set route through open roads in the greater Boston area. Data from 24 drives of at least 50-min duration were collected for analysis. The data were analyzed in two ways. Analysis I used features from 5-min intervals of data during the rest, highway, and city driving conditions to distinguish three levels of driver stress with an accuracy of over 97% across multiple drivers and driving days. Analysis II compared continuous features, calculated at 1-s intervals throughout the entire drive, with a metric of observable stressors created by independent coders from videotapes. The results show that for most drivers studied, skin conductivity and heart rate metrics are most closely correlated with driver stress level. These findings indicate that physiological signals can provide a metric of driver stress in future cars capable of physiological monitoring. Such a metric could be used to help manage noncritical in-vehicle information systems and could also provide a continuous measure of how different road and traffic conditions affect drivers. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Respiration <s> Repeated exposures to psychological stress can lead to or worsen diseases of slow accumulation such as heart diseases and cancer. The main challenge in addressing the growing epidemic of stress is a lack of robust methods to measure a person's exposure to stress in the natural environment. Periodic self-reports collect only subjective aspects, often miss stress episodes, and impose significant burden on subjects. Physiological sensors provide objective and continuous measures of stress response, but exhibit wide between-person differences and are easily confounded by daily activities (e.g., speaking, physical movements, coffee intake, etc.). In this paper, we propose, train, and test two models for continuous prediction of stress from physiological measurements captured by unobtrusive, wearable sensors. The first model is a physiological classifier that predicts whether changes in physiology represent stress. Since the effect of stress may persist in the mind longer than its acute effect on physiology, we propose a perceived stress model to predict perception of stress. It uses the output of the physiological classifier to model the accumulation and gradual decay of stress in the mind. To account for wide between-person differences, both models self-calibrate to each subject. Both models were trained using data collected from 21 subjects in a lab study, where they were exposed to cognitive, physical, and social stressors representative of that experienced in the natural environment. Our physiological classifier achieves 90% accuracy and our perceived stress model achieves a median correlation of 0.72 with self-reported rating. We also evaluate the perceived stress model on data collected from 17 participants in a two-day field study, and find that the average rating of stress obtained from our model has a correlation of 0.71 with that obtained from periodic self-reports. <s> BIB002

Although respiration can be assessed indirectly from measuring the blood oxygen level, a direct measurement contains more information about the actual respiration pattern. Commonly, a chest belt (respiratory inductive plethysmograph (RIP) BIB002 ), which is either worn thoracically or abdominally, is utilised to measure the respiration pattern directly. During a respiration cycle (inhalation and exhalation), the thorax expands and constricts. Hence, the chest belt experiences a sinusoidal stretching and destretching process, from which different physiological parameters like respiration rate and volume can be derived. Healey and Picard BIB001 sampled their respiration sensor at 31 Hz. However, following the Nyquist theorem a lower bound on the sampling rate of a RIP setup can be around 10-15 Hz. Nowadays, chest belts are mainly used by athletes monitoring their training progress. However, these devices have not found broad applications outside this domain.

Wearable-Based Affect Recognition—A Review <s> Electroencephalogram and Electrooculography <s> We discuss the strong relationship between affect and cognition and the importance of emotions in multimodal human computer interaction (HCI) and user modeling. We introduce the overall paradigm for our multimodal system that aims at recognizing its users' emotions and at responding to them accordingly depending upon the current context or application. We then describe the design of the emotion elicitation experiment we conducted by collecting, via wearable computers, physiological signals from the autonomic nervous system (galvanic skin response, heart rate, temperature) and mapping them to certain emotions (sadness, anger, fear, surprise, frustration, and amusement). We show the results of three different supervised learning algorithms that categorize these collected signals in terms of emotions, and generalize their learning to recognize emotions from new collections of signals. We finally discuss possible broader impact and potential applications of emotion recognition for multimodal intelligent systems. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Electroencephalogram and Electrooculography <s> This paper presents methods for collecting and analyzing physiological data during real-world driving tasks to determine a driver's relative stress level. Electrocardiogram, electromyogram, skin conductance, and respiration were recorded continuously while drivers followed a set route through open roads in the greater Boston area. Data from 24 drives of at least 50-min duration were collected for analysis. The data were analyzed in two ways. Analysis I used features from 5-min intervals of data during the rest, highway, and city driving conditions to distinguish three levels of driver stress with an accuracy of over 97% across multiple drivers and driving days. Analysis II compared continuous features, calculated at 1-s intervals throughout the entire drive, with a metric of observable stressors created by independent coders from videotapes. The results show that for most drivers studied, skin conductivity and heart rate metrics are most closely correlated with driver stress level. These findings indicate that physiological signals can provide a metric of driver stress in future cars capable of physiological monitoring. Such a metric could be used to help manage noncritical in-vehicle information systems and could also provide a continuous measure of how different road and traffic conditions affect drivers. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Electroencephalogram and Electrooculography <s> A physiological signal-based emotion recognition system is reported. The system was developed to operate as a user-independent system, based on physiological signal databases obtained from multiple subjects. The input signals were electrocardiogram, skin temperature variation and electrodermal activity, all of which were acquired without much discomfort from the body surface, and can reflect the influence of emotion on the autonomic nervous system. The system consisted of preprocessing, feature extraction and pattern classification stages. Preprocessing and feature extraction methods were devised so that emotion-specific characteristics could be extracted from short-segment signals. Although the features were carefully extracted, their distribution formed a classification problem, with large overlap among clusters and large variance within clusters. A support vector machine was adopted as a pattern classifier to resolve this difficulty. Correct-classification ratios for 50 subjects were 78.4% and 61.8%, for the recognition of three and four categories, respectively. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Electroencephalogram and Electrooculography <s> Chronic stress is endemic to modern society. However, as it is unfeasible for physicians to continuously monitor stress levels, its diagnosis is nontrivial. Wireless body sensor networks offer opportunities to ubiquitously detect and monitor mental stress levels, enabling improved diagnosis, and early treatment. This article describes the development of a wearable sensor platform to monitor a number of physiological correlates of mental stress. We discuss tradeoffs in both system design and sensor selection to balance information content and wearability. Using experimental signals collected from the wearable sensor, we describe a selected number of physiological features that show good correlation with mental stress. In particular, we propose a new spectral feature that estimates the balance of the autonomic nervous system by combining information from the power spectral density of respiration and heart rate variability. We validate the effectiveness of our approach on a binary discrimination problem when subjects are placed under two psychophysiological conditions: mental stress and relaxation. When used in a logistic regression model, our feature set is able to discriminate between these two mental states with a success rate of 81% across subjects. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Electroencephalogram and Electrooculography <s> This paper presents a user-independent emotion recognition method with the goal of recovering affective tags for videos using electroencephalogram (EEG), pupillary response and gaze distance. We first selected 20 video clips with extrinsic emotional content from movies and online resources. Then, EEG responses and eye gaze data were recorded from 24 participants while watching emotional video clips. Ground truth was defined based on the median arousal and valence scores given to clips in a preliminary study using an online questionnaire. Based on the participants' responses, three classes for each dimension were defined. The arousal classes were calm, medium aroused, and activated and the valence classes were unpleasant, neutral, and pleasant. One of the three affective labels of either valence or arousal was determined by classification of bodily responses. A one-participant-out cross validation was employed to investigate the classification performance in a user-independent approach. The best classification accuracies of 68.5 percent for three labels of valence and 76.4 percent for three labels of arousal were obtained using a modality fusion strategy and a support vector machine. The results over a population of 24 participants demonstrate that user-independent emotion recognition can outperform individual self-reports for arousal assessments and do not underperform for valence assessments. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Electroencephalogram and Electrooculography <s> We present a multimodal data set for the analysis of human affective states. The electroencephalogram (EEG) and peripheral physiological signals of 32 participants were recorded as each watched 40 one-minute long excerpts of music videos. Participants rated each video in terms of the levels of arousal, valence, like/dislike, dominance, and familiarity. For 22 of the 32 participants, frontal face video was also recorded. A novel method for stimuli selection is proposed using retrieval by affective tags from the last.fm website, video highlight detection, and an online assessment tool. An extensive analysis of the participants' ratings during the experiment is presented. Correlates between the EEG signal frequencies and the participants' ratings are investigated. Methods and results are presented for single-trial classification of arousal, valence, and like/dislike ratings using the modalities of EEG, peripheral physiological signals, and multimedia content analysis. Finally, decision fusion of the classification results from different modalities is performed. The data set is made publicly available and we encourage other researchers to use it for testing their own affective state estimation methods. <s> BIB006

The physiological modalities detailed above are only minimally intrusive. Hence, they are frequently employed in AR lab and field studies BIB004 BIB002 BIB001 BIB003 . In addition to the modalities listed above electroencephalogram (EEG) and electrooculography (EOG) are also often applied in AR studies. EEG, measuring the ionic current of brain neurons using electrodes placed on the scalp, was for instance employed by Soleymani et al. BIB005 to detect video-elicited emotions. EOG, which records horizontal and vertical eye movements by placing electrodes above/below and left/right of the eye, has been used by Koelstra et al. BIB006 . In our opinion, these modalities have the following disadvantages: • Both require the placement of electrodes on face/scalp. Hence, EEG and EOG are quite intrusive and not practical for everyday life. • They pose strong limitations on the movement of the participants and, hence, are not really applicable in real world scenarios. • EOG and EEG are prone to noise generated by muscle activity. Therefore, in the remainder of this review EEG and EOG will be given very little attention.

Wearable-Based Affect Recognition—A Review <s> Inertial Sensors <s> Despite the potential for stress and emotion recognition outside the lab environment, very little work has been reported that is feasible for use in the real world and much less for activities involving physical activity. In this work, we move a step forward towards a stress recognition system that works on a close to real world data set and shows a significant improvement over classification only systems. Our method uses clustering to separate the data into physical exertion levels and later performs stress classification over the discovered clusters. We validate our approach on a physiological stress dataset from 20 participants who performed 3 different activities of varying intensity under 3 different types of stimuli intended to cause stress. The results show an f-measure improvement of 130% compared to using classification only. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Inertial Sensors <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB002

Inertial sensors, incorporating a 3-axes acceleration (ACC), gyroscope and magnetometer, are commonly used in human activity recognition (HAR). In AR field studies the ACC signal can provide context information about the physical activity of the user. Gjoreski et al. BIB002 , for instance, used ACC data to classify six different activity types (lying, sitting, standing, walking, running and cycling). These activities, were then used as an additional input into a stress detection system. This certainly highlights the value of contextual information. However, results of Ramos et al. BIB001 indicate that in order to detect stress it is sufficient to estimate the intensity level of an activity instead of performing an exact activity classification.

Wearable-Based Affect Recognition—A Review <s> Context <s> Little attention has been paid so far to physiological signals for emotion recognition compared to audiovisual emotion channels such as facial expression or speech. This paper investigates the potential of physiological signals as reliable channels for emotion recognition. All essential stages of an automatic recognition system are discussed, from the recording of a physiological dataset to a feature-based multiclass classification. In order to collect a physiological dataset from multiple subjects over many weeks, we used a musical induction method which spontaneously leads subjects to real emotional states, without any deliberate lab setting. Four-channel biosensors were used to measure electromyogram, electrocardiogram, skin conductivity and respiration changes. A wide range of physiological features from various analysis domains, including time/frequency, entropy, geometric analysis, subband spectra, multiscale entropy, etc., is proposed in order to find the best emotion-relevant features and to correlate them with emotional states. The best features extracted are specified in detail and their effectiveness is proven by classification results. Classification of four musical emotions (positive/high arousal, negative/high arousal, negative/low arousal, positive/low arousal) is performed by using an extended linear discriminant analysis (pLDA). Furthermore, by exploiting a dichotomic property of the 2D emotion model, we develop a novel scheme of emotion-specific multilevel dichotomous classification (EMDC) and compare its performance with direct multiclass classification using the pLDA. Improved recognition accuracy of 95\% and 70\% for subject-dependent and subject-independent classification, respectively, is achieved by using the EMDC scheme. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Context <s> Repeated exposures to psychological stress can lead to or worsen diseases of slow accumulation such as heart diseases and cancer. The main challenge in addressing the growing epidemic of stress is a lack of robust methods to measure a person's exposure to stress in the natural environment. Periodic self-reports collect only subjective aspects, often miss stress episodes, and impose significant burden on subjects. Physiological sensors provide objective and continuous measures of stress response, but exhibit wide between-person differences and are easily confounded by daily activities (e.g., speaking, physical movements, coffee intake, etc.). In this paper, we propose, train, and test two models for continuous prediction of stress from physiological measurements captured by unobtrusive, wearable sensors. The first model is a physiological classifier that predicts whether changes in physiology represent stress. Since the effect of stress may persist in the mind longer than its acute effect on physiology, we propose a perceived stress model to predict perception of stress. It uses the output of the physiological classifier to model the accumulation and gradual decay of stress in the mind. To account for wide between-person differences, both models self-calibrate to each subject. Both models were trained using data collected from 21 subjects in a lab study, where they were exposed to cognitive, physical, and social stressors representative of that experienced in the natural environment. Our physiological classifier achieves 90% accuracy and our perceived stress model achieves a median correlation of 0.72 with self-reported rating. We also evaluate the perceived stress model on data collected from 17 participants in a two-day field study, and find that the average rating of stress obtained from our model has a correlation of 0.71 with that obtained from periodic self-reports. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Context <s> Work should be a source of health, pride, and happiness, in the sense of enhancing motivation and strengthening personal development. Healthy and motivated employees perform better and remain loyal to the company for a longer time. But, when the person constantly experiences high workload over a longer period of time and is not able to recover, then work may lead to prolonged negative effects and might cause serious illnesses like chronic stress disease. In this work, we present a solution for assessing the stress experience of people, using features derived from smartphones and wearable chest belts. In particular, we use information from audio, physical activity, and communication data collected during workday and heart rate variability data collected at night during sleep to build multinomial logistic regression models. We evaluate our system in a real work environment and in daily-routine scenarios of 35 employees over a period of 4 months and apply the leave-one-day-out cross-validation method for each user individually to estimate the prediction accuracy. Using only smartphone features, we get an accuracy of 55 %, and using only heart rate variability features, we get an accuracy of 59 %. The combination of all features leads to a rate of 61 % for a three-stress level (low, moderate, and high perceived stress) classification problem. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Context <s> Neurological assessment can be used to monitor a person's neurological status. In this paper, we report collection and analysis of a multimodal dataset of Non-EEG physiological signals available in the public domain. We have found this signal set useful for inferring the neurological status of individuals. The data was collected using non-invasive wrist worn biosensors and consists of electrodermal activity (EDA), temperature, acceleration, heart rate (HR), and arterial oxygen level (SpO2). We applied an efficient non-linear dimension reduction technique to visualize the biosignals in a low dimension feature space. We could cluster the four neurological statuses using an unsupervised Gaussian Mixture Model. The experimental results show that our unsupervised method can accurately separate different neurological statuses with an accuracy of greater than 84%. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Context <s> Stress remains a significant social problem for individuals in modern societies. This paper presents a machine learning approach for the automatic detection of stress of people in a social situation by combining two sensor systems that capture physiological and social responses. We compare the performance using different classifiers including support vector machine, AdaBoost, and k-nearest neighbour. Our experimental results show that by combining the measurements from both sensor systems, we could accurately discriminate between stressful and neutral situations during a controlled Trier social stress test (TSST). Moreover, this paper assesses the discriminative ability of each sensor modality individually and considers their suitability for real time stress detection. Finally, we present an study of the most discriminative features for stress detection. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Context <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Context <s> The detection and monitoring of emotions are important in various applications, e.g. to enable naturalistic and personalised human-robot interaction. Emotion detection often require modelling of various data inputs from multiple modalities, including physiological signals (e.g.EEG and GSR), environmental data (e.g. audio and weather), videos (e.g. for capturing facial expressions and gestures) and more recently motion and location data. Many traditional machine learning algorithms have been utilised to capture the diversity of multimodal data at the sensors and features levels for human emotion classification. While the feature engineering processes often embedded in these algorithms are beneficial for emotion modelling, they inherit some critical limitations which may hinder the development of reliable and accurate models. In this work, we adopt a deep learning approach for emotion classification through an iterative process by adding and removing large number of sensor signals from different modalities. Our dataset was collected in a real-world study from smart-phones and wearable devices. It merges local interaction of three sensor modalities: on-body, environmental and location into global model that represents signal dynamics along with the temporal relationships of each modality. Our approach employs a series of learning algorithms including a hybrid approach using Convolutional Neural Network and Long Short-term Memory Recurrent Neural Network (CNN-LSTM) on the raw sensor data, eliminating the needs for manual feature extraction and engineering. The results show that the adoption of deep-learning approaches is effective in human emotion classification when large number of sensors input is utilised (average accuracy 95% and F-Measure=%95) and the hybrid models outperform traditional fully connected deep neural network (average accuracy 73% and F-Measure=73%). Furthermore, the hybrid models outperform previously developed Ensemble algorithms that utilise feature engineering to train the model average accuracy 83% and F-Measure=82%) <s> BIB007

Finally, following Muaremi et al. BIB003 , smartphones offer an ideal platform to collect context information. This contextual data is aggregated by utilising position (GPS), sound snippets, calendar events, ambient light and user interaction with the phone BIB005 BIB003 BIB007 . Table 4 summarises recent wearable-based AR studies aspiring to detect different affective states, using wearable-based data. In order to identify relevant studied, a keyword-based search was performed in archival databases for the keywords affective computing/recognition, stress/emotion detection physiology and wearable-based AR, while explicitly removing non-wearable and EEG-only work. A detailed comparison of the employed classification algorithms, number of target classes, setting (e.g., lab or field), number of subjects, validation procedure and obtained accuracies, will be presented in Table 9 . In the studies presented in Table 4 , the target affective states are rather diverse: Almost 39% of the presented studies aimed to detect stress. For this purpose, different types of stressors (e.g., mental, physical, or social BIB002 BIB004 ) or different stress levels BIB006 are distinguished. Both the severe health implications and the strong physiological stress response (see Section 2.3), explain the popularity of stress recognition. According to Table 4 , various studies aim to recognise different emotional categories, distinguishing up to eight different affective states. Dimensional models of emotions (e.g., valence-arousal space) were used in 37% of the analysed studies. In 15% of the considered studies EEG was recorded. Nevertheless, there exists a large body of work, utilizing EEG data to classify different affective states. However, as mentioned in Section 1 this modality is not in scope of this review. As a result, studies utilizing EEG data are given less attention here. Concluding from Table 4 , sensor modalities monitoring cardiac activity are employed in 87% of the studies. EDA data was recorded in 76% of the studies. The popularity of these signals, certainly is linked to the strong impact of arousal-related changes on cardiac and electrodermal activity (see Section 3.1). In 32% of the considered studies, respiration data was acquired. Kim and André BIB001 pointed out that increased arousal can lead to an irregular respiration pattern. Finally, ACC, EMG and TEMP data were recorded in 32% of the studies. In summary, it is observed that sensors measuring parameters directly influenced by the SNS are most popular. Sensory setups recording less distinct changes are employed less frequently. Table 4 . Affective states and sensor signals frequently employed in wearable-based AR. Table 9 provides further detail on algorithms, location and performance. Abbreviations: 3-axes acceleration (ACC), blood pressure (BP), electrocardiogram (ECG), electrodermal activity (EDA), electroencephalogram (EEG), electromyogram (EMG), electrooculography (EOG), heart rate (HR), magnetoencephalogram (MEG), pupil diameter (PD), photoplethysmogram (PPG), respiration (RESP), skin-temperature (TEMP), arterial oxygen level (SpO2), low arousal/low valence (LALV), low arousal/high valence (LAHV), high arousal/low valence (HALV), high arousal/high valence (HAHV).

Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> This paper describes a protocol for induction of moderate psychological stress in a laboratory setting and evaluates its effects on physiological responses. The ‘Trier Social Stress Test’ (TSST) mainl <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Abstract Researchers interested in emotion have long struggled with the problem of how to elicit emotional responses in the laboratory. In this article, we summarise five years of work to develop a set of films that reliably elicit each of eight emotional states (amusement, anger, contentment, disgust, fear, neutral, sadness, and surprise). After evaluating over 250 films, we showed selected film clips to an ethnically diverse sample of 494 English-speaking subjects. We then chose the two best films for each of the eight target emotions based on the intensity and discreteness of subjects' responses to each film. We found that our set of 16 films successfully elicited amusement, anger, contentment. disgust, sadness, surprise, a relatively neutral state, and, to a lesser extent, fear. We compare this set of films with another set recently described by Philippot (1993), and indicate that detailed instructions for creating our set of film stimuli will be provided on request. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> The International Affective Picture System (IAPS) is widely used in studies of emotion and has been characterized primarily along the dimensions of valence, arousal, and dominance. Even though research has shown that the IAPS is useful in the study of discrete emotions, the categorical structure of the IAPS has not been characterized thoroughly. The purpose of the present project was to collect descriptive emotional category data on subsets of the IAPS in an effort to identify images that elicit one discrete emotion more than others. These data reveal multiple emotional categories for the images and indicate that this image set has great potential in the investigation of discrete emotions. This article makes these data available to researchers with such interests. Data for all the pictures are archived at www .psychonomic.org/archive/. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> The mystery surrounding emotions, how they work and how they affect our lives has not yet been unravelled. Scientists still debate the real nature of emotions, whether they are evolutionary, physiological or cognitive are just a few of the different approaches used to explain affective states. Regardless of the various emotional paradigms, neurologists have made progress in demonstrating that emotion is as, or more, important than reason in the process of making decisions and deciding actions. The significance of these findings should not be overlooked in a world that is increasingly reliant on computers to accommodate to user needs. In this paper, a novel approach for recognizing and classifying positive and negative emotional changes in real time using physiological signals is presented. Based on sequential analysis and autoassociative networks, the emotion detection system outlined here is potentially capable of operating on any individual regardless of their physical state and emotional intensity without requiring an arduous adaptation or pre-analysis phase. Results from applying this methodology on real-time data collected from a single subject demonstrated a recognition level of 71.4% which is comparable to the best results achieved by others through off-line analysis. It is suggested that the detection mechanism outlined in this paper has all the characteristics needed to perform emotion recognition in pervasive computing. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Many studies showed the ability of movies and imagery techniques to elicit emotions. Nevertheless, it is less clear how to manipulate the content of interactive media to induce specific emotional responses. In particular, this is true for the emerging medium virtual reality (VR), whose main feature is the ability to induce a feeling of "presence" in the computer-generated world experienced by the user. The main goal of this study was to analyze the possible use of VR as an affective medium. Within this general goal, the study also analyzed the relationship between presence and emotions. The results confirmed the efficacy of VR as affective medium: the interaction with "anxious" and "relaxing" virtual environments produced anxiety and relaxation. The data also showed a circular interaction between presence and emotions: on one side, the feeling of presence was greater in the "emotional" environments; on the other side, the emotional state was influenced by the level of presence. The significance of these r... <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> In this review, we report on studies that have assessed the effects of exogenous and endogenous increases in stress hormones on human cognitive performance. We first describe the history of the studies on the effects of using exogenous stress hormones such as glucocorticoids as anti-inflammatory medications on human cognition and mental health. Here, we summarize the cases that led to the diagnosis of glucocorticoid-induced 'steroid psychosis' in human populations and which demonstrated that these stress hormones could thus cross the blood-brain barrier and access the brain where they could influence cognition and mental health. We then summarize studies that assessed the effects of the exogenous administration of glucocorticoids on cognitive performance supported by the hippocampus, the frontal lobes and amygdala. In the second section of the paper, we summarize the effects of the endogenous release of glucocorticoids induced by exposure to a stressful situation on human cognition and we further dissociate the effects of emotion from those of stress on human learning and memory. Finally, in the last section of the paper, we discuss the potential impact that the environmental context to which we expose participants when assessing their memory could have on their reactivity to stress and subsequent cognitive performance. In order to make our point, we discuss the field of memory and aging and we suggest that some of the 'age-related memory impairments' observed in the literature could be partly due to increased stress reactivity in older adults to the environmental context of testing. We also discuss the inverse negative correlations reported between hippocampal volume and memory for young and older adults and suggest that these inverse correlations could be partly due to the effects of contextual stress in young and older adults, as a function of age-related differences in hippocampal volume. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> In this paper we present a multimodal approach for the recognition of eight emotions. Our approach integrates information from facial expressions, body movement and gestures and speech. We trained and tested a model with a Bayesian classifier, using a multimodal corpus with eight emotions and ten subjects. Firstly, individual classifiers were trained for each modality. Next, data were fused at the feature level and the decision level. Fusing the multimodal data resulted in a large increase in the recognition rates in comparison with the unimodal systems: the multimodal approach gave an improvement of more than 10% when compared to the most successful unimodal system. Further, the fusion performed at the feature level provided better results than the one performed at the decision level. <s> BIB007 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Repeated exposures to psychological stress can lead to or worsen diseases of slow accumulation such as heart diseases and cancer. The main challenge in addressing the growing epidemic of stress is a lack of robust methods to measure a person's exposure to stress in the natural environment. Periodic self-reports collect only subjective aspects, often miss stress episodes, and impose significant burden on subjects. Physiological sensors provide objective and continuous measures of stress response, but exhibit wide between-person differences and are easily confounded by daily activities (e.g., speaking, physical movements, coffee intake, etc.). In this paper, we propose, train, and test two models for continuous prediction of stress from physiological measurements captured by unobtrusive, wearable sensors. The first model is a physiological classifier that predicts whether changes in physiology represent stress. Since the effect of stress may persist in the mind longer than its acute effect on physiology, we propose a perceived stress model to predict perception of stress. It uses the output of the physiological classifier to model the accumulation and gradual decay of stress in the mind. To account for wide between-person differences, both models self-calibrate to each subject. Both models were trained using data collected from 21 subjects in a lab study, where they were exposed to cognitive, physical, and social stressors representative of that experienced in the natural environment. Our physiological classifier achieves 90% accuracy and our perceived stress model achieves a median correlation of 0.72 with self-reported rating. We also evaluate the perceived stress model on data collected from 17 participants in a two-day field study, and find that the average rating of stress obtained from our model has a correlation of 0.71 with that obtained from periodic self-reports. <s> BIB008 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Information about a user's emotional state is a very important aspect of affective interaction with embodied conversational agents. Most research work aims at identifying emotions through speech or facial expressions. However, facial expressions and speech are not continuously available. Furthermore, in some cases, bio-signal data are also required in order to fully assess a user's emotional state. We aimed to recognize the six, basic, primary emotions proposed by Ekman, using a widely-available and low-cost brain-computer interface (BCI) and a biofeedback sensor that measures heart rate. We exposed participants to sets of 10 IAPS images that had been partially validated through a subjective rating protocol. Results showed that the collected signals allowed us identifying user's emotional state. In addition, a partial correlation between objective and subjective data can be observed. <s> BIB009 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> MAHNOB-HCI is a multimodal database recorded in response to affective stimuli with the goal of emotion recognition and implicit tagging research. A multimodal setup was arranged for synchronized recording of face videos, audio signals, eye gaze data, and peripheral/central nervous system physiological signals. Twenty-seven participants from both genders and different cultural backgrounds participated in two experiments. In the first experiment, they watched 20 emotional videos and self-reported their felt emotions using arousal, valence, dominance, and predictability as well as emotional keywords. In the second experiment, short videos and images were shown once without any tag and then with correct or incorrect tags. Agreement or disagreement with the displayed tags was assessed by the participants. The recorded videos and bodily responses were segmented and stored in a database. The database is made available to the academic community via a web-based system. The collected data were analyzed and single modality and modality fusion results for both emotion recognition and implicit tagging experiments are reported. These results show the potential uses of the recorded modalities and the significance of the emotion elicitation protocol. <s> BIB010 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> We present a multimodal data set for the analysis of human affective states. The electroencephalogram (EEG) and peripheral physiological signals of 32 participants were recorded as each watched 40 one-minute long excerpts of music videos. Participants rated each video in terms of the levels of arousal, valence, like/dislike, dominance, and familiarity. For 22 of the 32 participants, frontal face video was also recorded. A novel method for stimuli selection is proposed using retrieval by affective tags from the last.fm website, video highlight detection, and an online assessment tool. An extensive analysis of the participants' ratings during the experiment is presented. Correlates between the EEG signal frequencies and the participants' ratings are investigated. Methods and results are presented for single-trial classification of arousal, valence, and like/dislike ratings using the modalities of EEG, peripheral physiological signals, and multimedia content analysis. Finally, decision fusion of the classification results from different modalities is performed. The data set is made publicly available and we encourage other researchers to use it for testing their own affective state estimation methods. <s> BIB011 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> This paper presents a user-independent emotion recognition method with the goal of recovering affective tags for videos using electroencephalogram (EEG), pupillary response and gaze distance. We first selected 20 video clips with extrinsic emotional content from movies and online resources. Then, EEG responses and eye gaze data were recorded from 24 participants while watching emotional video clips. Ground truth was defined based on the median arousal and valence scores given to clips in a preliminary study using an online questionnaire. Based on the participants' responses, three classes for each dimension were defined. The arousal classes were calm, medium aroused, and activated and the valence classes were unpleasant, neutral, and pleasant. One of the three affective labels of either valence or arousal was determined by classification of bodily responses. A one-participant-out cross validation was employed to investigate the classification performance in a user-independent approach. The best classification accuracies of 68.5 percent for three labels of valence and 76.4 percent for three labels of arousal were obtained using a modality fusion strategy and a support vector machine. The results over a population of 24 participants demonstrate that user-independent emotion recognition can outperform individual self-reports for arousal assessments and do not underperform for valence assessments. <s> BIB012 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Chronic stress is endemic to modern society. However, as it is unfeasible for physicians to continuously monitor stress levels, its diagnosis is nontrivial. Wireless body sensor networks offer opportunities to ubiquitously detect and monitor mental stress levels, enabling improved diagnosis, and early treatment. This article describes the development of a wearable sensor platform to monitor a number of physiological correlates of mental stress. We discuss tradeoffs in both system design and sensor selection to balance information content and wearability. Using experimental signals collected from the wearable sensor, we describe a selected number of physiological features that show good correlation with mental stress. In particular, we propose a new spectral feature that estimates the balance of the autonomic nervous system by combining information from the power spectral density of respiration and heart rate variability. We validate the effectiveness of our approach on a binary discrimination problem when subjects are placed under two psychophysiological conditions: mental stress and relaxation. When used in a logistic regression model, our feature set is able to discriminate between these two mental states with a success rate of 81% across subjects. <s> BIB013 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> The paper presents a multi-modal emotion recognition system exploiting audio and video (i.e., facial expression) information. The system first processes both sources of information individually to produce corresponding matching scores and then combines the computed matching scores to obtain a classification decision. For the video part of the system, a novel approach to emotion recognition, relying on image-set matching, is developed. The proposed approach avoids the need for detecting and tracking specific facial landmarks throughout the given video sequence, which represents a common source of error in video-based emotion recognition systems, and, therefore, adds robustness to the video processing chain. The audio part of the system, on the other hand, relies on utterance-specific Gaussian Mixture Models (GMMs) adapted from a Universal Background Model (UBM) via the maximum a posteriori probability (MAP) estimation. It improves upon the standard UBM-MAP procedure by exploiting gender information when bu... <s> BIB014 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> More than 15 years after the early studies in Affective Computing (AC), [1] the problem of detecting and modeling emotions in the context of human-computer interaction (HCI) remains complex and largely unexplored. The detection and modeling of emotion is, primarily, the study and use of artificial intelligence (AI) techniques for the construction of computational models of emotion. The key challenges one faces when attempting to model emotion [2] are inherent in the vague definitions and fuzzy boundaries of emotion, and in the modeling methodology followed. In this context, open research questions are still present in all key components of the modeling process. These include, first, the appropriateness of the modeling tool employed to map emotional manifestations and responses to annotated affective states; second, the processing of signals that express these manifestations (i.e., model input); and third, the way affective annotation (i.e., model output) is handled. This paper touches upon all three key components of an affective model (i.e., input, model, output) and introduces the use of deep learning (DL) [3], [4], [5] methodologies for affective modeling from multiple physiological signals. <s> BIB015 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Timely mental stress detection can help to prevent stress-related health problems. The aim of this study was to identify those physiological signals and features suitable for detecting mental stress in office-like situations. Electrocardiogram (ECG), respiration, skin conductance and surface electromyogram (sEMG) of the upper trapezius muscle were measured with a wearable system during three distinctive stress tests. The protocol contained stress tests that were designed to represent office-like situations. Generalized Estimating Equations were used to classify the data into rest and stress conditions. We reached an average classification rate of 74.5%. This approach may be used for continuous stress measurement in daily office life to detect mental stress at an early stage. <s> BIB016 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Despite the potential for stress and emotion recognition outside the lab environment, very little work has been reported that is feasible for use in the real world and much less for activities involving physical activity. In this work, we move a step forward towards a stress recognition system that works on a close to real world data set and shows a significant improvement over classification only systems. Our method uses clustering to separate the data into physical exertion levels and later performs stress classification over the discovered clusters. We validate our approach on a physiological stress dataset from 20 participants who performed 3 different activities of varying intensity under 3 different types of stimuli intended to cause stress. The results show an f-measure improvement of 130% compared to using classification only. <s> BIB017 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> In this work, we present DECAF —a multimodal data set for dec oding user physiological responses to af fective multimedia content. Different from data sets such as DEAP [15] and MAHNOB-HCI [31] , DECAF contains (1) brain signals acquired using the Magnetoencephalogram (MEG) sensor, which requires little physical contact with the user’s scalp and consequently facilitates naturalistic affective response, and (2) explicit and implicit emotional responses of 30 participants to 40 one-minute music video segments used in [15] and 36 movie clips, thereby enabling comparisons between the EEG versus MEG modalities as well as movie versus music stimuli for affect recognition. In addition to MEG data, DECAF comprises synchronously recorded near-infra-red (NIR) facial videos, horizontal Electrooculogram (hEOG), Electrocardiogram (ECG), and trapezius-Electromyogram (tEMG) peripheral physiological responses. To demonstrate DECAF’s utility, we present (i) a detailed analysis of the correlations between participants’ self-assessments and their physiological responses and (ii) single-trial classification results for valence , arousal and dominance , with performance evaluation against existing data sets. DECAF also contains time-continuous emotion annotations for movie clips from seven users, which we use to demonstrate dynamic emotion prediction. <s> BIB018 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> In mobile computing, varying access to resources makes it difficult for developers to ensure that satisfactory system response times will be maintained at all times. Wearable physiological sensors offer a way to dynamically detect user frustration in response to increased system delays. However, most prior efforts have focused on binary classifiers designed to detect the presence or absence of a task-specific stimulus. In this paper, we make two contributions. Our first contribution is in identifying the use of variable length system response delays, a universal and task-independent feature of computing, as a stimulus for driving different levels of frustration. By doing so, we are able to make our second and primary contribution, which is the development of models that predict multiple levels of user frustration from psycho-physiological responses caused by system response delays. We investigate how incorporating different sensor features, application settings, and timing constraints impact the performance of our models. We demonstrate that our models of physiological responses can be used to classify five levels of frustration in near real-time with over 80% accuracy, which is comparable to the accuracy of binary classifiers. <s> BIB019 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> We describe the creation of a film library designed for researchers interested in positive (amusing), negative (repulsive), mixed (amusing and repulsive) and neutral emotional states. Three hundred 20- to 33-second film clips videotaped by amateurs were selected from video-hosting websites and screened in laboratory studies by 75 female participants on self-reported amusement and repulsion (Experiments 1 and 2). On the basis of pre-defined cut-off values, 51 positive, 39 negative, 59 mixed and 50 neutral film clips were selected. These film clips were then presented to 411 male and female participants in a large online study to identify film clips that reliably induced the target emotions (Experiment 3). Depending on the goal of the study, researchers may choose positive, negative, mixed or neutral emotional film clips on the basis of Experiments 1 and 2 or Experiment 3 ratings. <s> BIB020 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Continuous exposure to stress is harmful for mental and physical health, but to combat stress, one should first detect it. In this paper we propose a method for continuous detection of stressful events using data provided from a commercial wrist device. The method consists of three machine-learning components: a laboratory stress detector that detects short-term stress every 2 minutes; an activity recognizer that continuously recognizes user's activity and thus provides context information; and a context-based stress detector that exploits the output of the laboratory stress detector and the user's context in order to provide the final decision on 20 minutes interval. The method was evaluated in a laboratory and a real-life setting. The accuracy on 55 days of real-life data, for a 2-class problem, was 92%. The method is currently being integrated in a smartphone application for managing mental health and well-being. <s> BIB021 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Stress remains a significant social problem for individuals in modern societies. This paper presents a machine learning approach for the automatic detection of stress of people in a social situation by combining two sensor systems that capture physiological and social responses. We compare the performance using different classifiers including support vector machine, AdaBoost, and k-nearest neighbour. Our experimental results show that by combining the measurements from both sensor systems, we could accurately discriminate between stressful and neutral situations during a controlled Trier social stress test (TSST). Moreover, this paper assesses the discriminative ability of each sensor modality individually and considers their suitability for real time stress detection. Finally, we present an study of the most discriminative features for stress detection. <s> BIB022 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB023 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Affective Computing has emerged as an important field of study that aims to develop systems that can automatically recognize emotions. Up to the present, elicitation has been carried out with non-immersive stimuli. This study, on the other hand, aims to develop an emotion recognition system for affective states evoked through Immersive Virtual Environments. Four alternative virtual rooms were designed to elicit four possible arousal-valence combinations, as described in each quadrant of the Circumplex Model of Affects. An experiment involving the recording of the electroencephalography (EEG) and electrocardiography (ECG) of sixty participants was carried out. A set of features was extracted from these signals using various state-of-the-art metrics that quantify brain and cardiovascular linear and nonlinear dynamics, which were input into a Support Vector Machine classifier to predict the subject’s arousal and valence perception. The model’s accuracy was 75.00% along the arousal dimension and 71.21% along the valence dimension. Our findings validate the use of Immersive Virtual Environments to elicit and automatically recognize different emotional states from neural and cardiac dynamics; this development could have novel applications in fields as diverse as Architecture, Health, Education and Videogames. <s> BIB024 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> Affect recognition aims to detect a person's affective state based on observables, with the goal to e.g. improve human-computer interaction. Long-term stress is known to have severe implications on wellbeing, which call for continuous and automated stress monitoring systems. However, the affective computing community lacks commonly used standard datasets for wearable stress detection which a) provide multimodal high-quality data, and b) include multiple affective states. Therefore, we introduce WESAD, a new publicly available dataset for wearable stress and affect detection. This multimodal dataset features physiological and motion data, recorded from both a wrist- and a chest-worn device, of 15 subjects during a lab study. The following sensor modalities are included: blood volume pulse, electrocardiogram, electrodermal activity, electromyogram, respiration, body temperature, and three-axis acceleration. Moreover, the dataset bridges the gap between previous lab studies on stress and emotions, by containing three different affective states (neutral, stress, amusement). In addition, self-reports of the subjects, which were obtained using several established questionnaires, are contained in the dataset. Furthermore, a benchmark is created on the dataset, using well-known features and standard machine learning methods. Considering the three-class classification problem ( baseline vs. stress vs. amusement ), we achieved classification accuracies of up to 80%,. In the binary case ( stress vs. non-stress ), accuracies of up to 93%, were reached. Finally, we provide a detailed analysis and comparison of the two device locations ( chest vs. wrist ) as well as the different sensor modalities. <s> BIB025 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in Laboratory Settings <s> We present ASCERTAIN—a multimodal databa AS e for impli C it p ER sonali T y and A ffect recognit I o N using commercial physiological sensors. To our knowledge, ASCERTAIN is the first database to connect personality traits and emotional states via physiological responses . ASCERTAIN contains big-five personality scales and emotional self-ratings of 58 users along with their Electroencephalogram (EEG), Electrocardiogram (ECG), Galvanic Skin Response (GSR) and facial activity data, recorded using off-the-shelf sensors while viewing affective movie clips. We first examine relationships between users’ affective ratings and personality scales in the context of prior observations, and then study linear and non-linear physiological correlates of emotion and personality. Our analysis suggests that the emotion-personality relationship is better captured by non-linear rather than linear statistics. We finally attempt binary emotion and personality trait recognition using physiological features. Experimental results cumulatively confirm that personality differences are better revealed while comparing user responses to emotionally homogeneous videos, and above-chance recognition is achieved for both affective and personality dimensions. <s> BIB026

Humans differ in their personality. Hence, generating data that corresponds to a particular emotional state is a challenging task BIB009 . However, due to the controlled lab environment, researchers can conduct studies following well-designed protocols. Another advantage of lab studies is that their replication is possible, due to the well defined experimental protocol. Below a detailed overview of stimuli frequently employed to elicit affective states in AR lab studies is provided: Images: The International Affective Picture System (IAPS) is a dataset comprised of colour photographs. The IAPS was compiled such that each image elicits an emotional reaction. Each image was rated multiple times by study participants, providing labels in the valence and arousal space. Mikels et al. BIB003 identified a subset of IAPS images, which elicits certain discrete emotions. Hence, depending on the desired emotion, one can choose particularly strong images from this subset. In the AR domain, the IAPS has, for instance, been used by Leon et al. BIB004 and by Hamdi et al. BIB009 . In the experiments presented by Leon et al. BIB004 , 21 images from the IAPS were used to elicit three different affective states (neutral, positive, negative). Hamdi et al. BIB009 exposed their study participants to ten images from the IAPS and aimed at recognising six basic emotions (disgust, joy, surprise, sadness, fear, anger) based on physiological data. Videos: According to Gross and Levenson BIB002 , short audiovisual clips are very suitable to elicit discrete emotions. Hence, video clips are frequently employed as stimuli BIB010 BIB011 BIB018 . A common procedure to select a set of videos evoking certain target emotions is to chose them from a large pool of videos. The process of identifying the most appropriate subset often happens in two steps: First, the clips are watched and rated by a large number of individuals. Second, the clips which elicit a certain emotion most reliably are chosen as stimuli in the study BIB011 BIB012 . Recently, Samson et al. BIB020 published a study on 199 short amateur clips which were rated by 411 subjects with respect to three affective categories (neutral, positive, negative). In AR literature, there are many examples where audiovisual clips have been used to elicit different affective states. Koelstra et al. BIB011 chose in their experiments music clips with a length of 60 s. After each stimulus, the progress was displayed and a 5 s baseline was recorded. Soleymani et al. BIB012 showed their participants 60 to 120 s long excerpts from movies and after each clip a short neutral clip (15 s) was displayed. Acted emotions: In the above detailed protocols, emotions are event-elicited. Another way of generating affective states is to ask the subjects to purposefully elicit emotions, for example, act an emotion. For instance, Hanai and Ghassemi asked the study participants to tell at least one happy and one sad story. Other researchers asked trained actors to perform certain emotions BIB007 BIB014 . These types of approaches are frequently employed in sentiment analysis and emotion recognition from audio/video data. Game elicited emotions: Another way to elicit a target affective state is to ask the subjects to perform a certain task. Using a Breakout engine and introducing a latency between the user's input and the reaction in the game, Taylor et al. BIB019 elicited frustration in their study participants. Martinez et al. BIB015 used four different versions of a Maze-Ball game to generate pairwise preference scores. The scores were generated by asking the subjects which of two games felt more anxious, exciting, frustrating, fun, and relaxing. Affective states elicited by immersive media: Advances in Virtual Reality (VR), like head mounted displays or VR-headsets open new possibilities to elicit affective states. Up to now these methods have not found broad application in wearable-based AR. However, this is a particularly interesting elicitation method as it allows to simulate close to real world scenarios, hence, offering optimal control, while retaining ecological validity. Such a method was for instance applied by Riva et al. BIB005 , who used a head mounted display, showing three different VR environments (virtual parks), as an affective stimuli, eliciting a neutral, anxious and a relaxed affective state. More recently, Marín-Morales et al. BIB024 employed four architectural environments, displayed to the subjects via a VR-headset, to elicit different affective states, too. Stress inducing study protocols: There are numerous protocols aiming at eliciting stress in the study participants. Mason showed that in order to trigger a (physiological) stress response, the situation has to be either novel and/or unpredictable and/or beyond control for the subject BIB006 . Stressors frequently employed in the AR literature can be categorised as follows: C1 Social-evaluative Stressors: A task creating a socially relevant situation for the subject. For example, performing a task in front of a panel which evaluates the subject. C2 Cognitive Stressors: A task demanding significant mental engagement and attention. For example, performing an (challenging) arithmetic task under time pressure. C3 Physical Stressors: A task creating a physically uncomfortable situation. For example, being exposed to extreme hot or cold. A well-studied and frequently employed stress elicitation protocol is the Trier Social Stress Test BIB001 . The Trier Social Stress Test (TSST) has two conditions: a public speaking/job interview type of situation and a mental arithmetic task. Hence, the TSST incorporates both a social-evaluative (C1) and cognitive stressor (C2). Due to its reliability and easy set-up, the TSST was administered in numerous AR studies, e.g., Mozos et al. BIB022 , Plarre et al. BIB008 , Schmidt et al. BIB025 , Hovsepian et al. , Gjoreski et al. BIB021 . Another stressor employed to target cognitive load is the so called Stroop color test . In this condition, the subjects have to read out loud a sequence of colours written on a screen. However, the font colour does not match the written colour (e.g., green, blue, etc.). As a result, the task inflicts a high cognitive load and, hence, is a C2 stressor. The Stroop colour test has for instance been employed by Choi et al. BIB013 , who aimed for the development of a wearable-based stress monitoring system. Using computer tasks, stress can also be elicited reliable. Wijsman et al. BIB016 , for instance, asked the subjects to perform a calculation, to solve a logical puzzle and to do a memorisation task. These tasks can all be seen as C2 stressors. These tasks had to be completed under time pressure. In addition, the subjects were distracted with sounds and parts of the protocol (memorisation task) were also recorded on video. Furthermore, as the participants of Reference BIB016 were told that their scores would be made available to their colleagues, the study protocol also had a social-evaluative component (see C1). The cold pressor test, applied by Plarre et al. BIB008 , can be used to evoke physical stress, corresponding to a C3 stressor. Following this test, the subjects are asked to place their hand into a bucket of ice cold water and leave it there for a predefined time (e.g., 60 s). Now as a common set of stimuli has been detailed, the issue of obtaining ground truth in a lab setting is discussed briefly. Following for instance Plarre et al. BIB008 , employed conditions (e.g., stressors) can be used as ground truth. One way to ensure the validity of the employed stimulus is to utilize exactly the same set up as in a related study. In addition, questionnaires integrated into the protocol should be used to verify that the desired affective states were successfully evoked (see for instance BIB025 ). Typically, these questionnaires are used directly after each affective stimulus or condition. Ramos et al. BIB017 , for instance, collected subjective stress levels after each stressor. In addition, the Stait-Trait Anxiety Inventory also has been used to capture different stress levels BIB023 . In order to generate labels in valence-arousal space the SAM are employed frequently BIB011 BIB025 BIB012 . In addition, as the perception of a stimulus can be influenced by personality traits, collecting this information, can be useful too BIB026 .

Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> OBJECTIVE: To address the lack of a simple and standardized instrument to assess overall panic disorder severity, the authors developed a scale for the measurement of panic disorder severity. METHOD: Ten independent evaluators used the seven-item Panic Disorder Severity Scale to assess 186 patients with principal DSM-III-R diagnoses of panic disorder (with no or mild agoraphobia) who were participating in the Multicenter Collaborative Treatment Study of Panic Disorder. In addition, 89 of these patients were reevaluated with the same scale after short-term treatment. A subset of 24 patients underwent two independent assessments to establish interrater reliability. Internal consistency, convergent and discriminant validity, and sensitivity to change were also determined. RESULTS: The Panic Disorder Severity Scale was associated with excellent interrater reliability, moderate internal consistency, and favorable levels of validity and sensitivity to change. Individual items showed good convergent and discrimi... <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> Heart rate variability (HRV) has been well established to measure instantaneous levels of mental stress. Circadian patterns of HRV features have been reported but their use to estimate levels of mental stress were not studied thoroughly. In this study, we investigated time dependent variations of HRV features to detect subjects under chronic mental stress. Sixty eight subjects were divided into high (n=10) and low stress group (n=43) depending on their self-reporting stress scores. HRV features were calculated during three different time periods of the day. High stress group showed decreased patterns of HRV features compared to low stress group. When logistic regression analysis was performed with raw multiple HRV features, the classification was 63.2% accurate. A new % deviance score reflecting the degree of difference from normal reference patterns increased the accuracy to 66.1%. Our data suggested that HRV patterns obtained at multiple time points of the day could provide useful data to monitor subjects under chronic stress. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> We conducted a 19 participant study using a system comprised of wireless galvanic skin response (GSR), heart rate (HR), activity sensors and a mobile phone for aggregating sensor data and enabling affect logging by the user. Each participant wore the sensors daily for five days, generating approximately 900 hours of continuous data. We found that analysis of emotional events was highly dependent on correct windowing and report results on synthesized windows around annotated events. Where raters agreed on the timing and quality of the emotion we were able to recognize 85% of the high and low energy emotions and 70% of the positive and negative emotions. We also gained many insights regarding participant's perception of their emotional state and the complexity of emotion in real life. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> Nine call center employees wore a skin conductance sensor on the wrist for a week at work and reported stress levels of each call. Although everyone had the same job profile, we found large differences in how individuals reported stress levels, with similarity from day to day within the same participant, but large differences across the participants. We examined two ways to address the individual differences to automatically recognize classes of stressful/non-stressful calls, namely modifying the loss function of Support Vector Machines (SVMs) to adapt to the varying priors, and giving more importance to training samples from the most similar people in terms of their skin conductance lability. We tested the methods on 1500 calls and achieved an accuracy across participants of 78.03% when trained and tested on different days from the same person, and of 73.41% when trained and tested on different people using the proposed adaptations to SVMs. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> Work should be a source of health, pride, and happiness, in the sense of enhancing motivation and strengthening personal development. Healthy and motivated employees perform better and remain loyal to the company for a longer time. But, when the person constantly experiences high workload over a longer period of time and is not able to recover, then work may lead to prolonged negative effects and might cause serious illnesses like chronic stress disease. In this work, we present a solution for assessing the stress experience of people, using features derived from smartphones and wearable chest belts. In particular, we use information from audio, physical activity, and communication data collected during workday and heart rate variability data collected at night during sleep to build multinomial logistic regression models. We evaluate our system in a real work environment and in daily-routine scenarios of 35 employees over a period of 4 months and apply the leave-one-day-out cross-validation method for each user individually to estimate the prediction accuracy. Using only smartphone features, we get an accuracy of 55 %, and using only heart rate variability features, we get an accuracy of 59 %. The combination of all features leads to a rate of 61 % for a three-stress level (low, moderate, and high perceived stress) classification problem. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> In this study, we aim to find physiological or behavioral markers for stress. We collected 5 days of data for 18 participants: a wrist sensor (accelerometer and skin conductance), mobile phone usage (call, short message service, location and screen on/off) and surveys (stress, mood, sleep, tiredness, general health, alcohol or caffeinated beverage intake and electronics usage). We applied correlation analysis to find statistically significant features associated with stress and used machine learning to classify whether the participants were stressed or not. In comparison to a baseline 87.5% accuracy using the surveys, our results showed over 75% accuracy in a binary classification using screen on, mobility, call or activity level information (some showed higher accuracy than the baseline). The correlation analysis showed that the higher-reported stress level was related to activity level, SMS and screen on/off patterns. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> Much of the stress and strain of student life remains hidden. The StudentLife continuous sensing app assesses the day-to-day and week-by-week impact of workload on stress, sleep, activity, mood, sociability, mental well-being and academic performance of a single class of 48 students across a 10 week term at Dartmouth College using Android phones. Results from the StudentLife study show a number of significant correlations between the automatic objective sensor data from smartphones and mental health and educational outcomes of the student body. We also identify a Dartmouth term lifecycle in the data that shows students start the term with high positive affect and conversation levels, low stress, and healthy sleep and daily activity patterns. As the term progresses and the workload increases, stress appreciably rises while positive affect, sleep, conversation and activity drops off. The StudentLife dataset is publicly available on the web. <s> BIB007 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> In this paper, we present first steps towards a mobile and wearable system intended to help people who experience regular and spontaneous panic attacks due to panic disorder. The goal of the system is to predict oncoming panic attacks and to deliver in-the-moment interventions on a smartphone device. Interventions are intended to reduce symptom severity by enabling a user to respond to approaching panic episodes. An initial feasibility study is described where a small real-world data set was collected. Personalized prediction models were trained which take, as input, physiological data and output a binary classification of either pre-panic or non-panic. We demonstrate proof-of-concept of episode prediction on this small dataset. <s> BIB008 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> We collected and analyzed subjective and objective data using surveys and wearable sensors worn day and night from 68 participants, for 30 days each, to address questions related to the relationships among sleep duration, sleep irregularity, self-reported Happy-Sad mood and other factors in college students. We analyzed daily and monthly behavior and physiology and identified factors that affect mood, including how accurately sleep duration and sleep regularity for the past 1-5 days classified the participants into high/low mood using support vector machines. We found statistically significant associations among sad mood and poor health-related factors. Behavioral factors such as the percentage of neutral social interactions and the total academic activity hours showed the best performance in separating the Happy-Sad mood groups. Sleep regularity was a more important discriminator of mood than sleep duration for most participants, although both variables predicted happy/sad mood with from 70-82% accuracy. The number of nights giving the best prediction of happy/sad mood varied for different groups of individuals. <s> BIB009 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> Stress, anxiety and depression in the workplace are detrimental to human health and productivity with significant financial implications. Recent research in this area has focused on the use of sensor technologies, including smartphones and wearables embedded with physiological and movement sensors. In this work, we explore the possibility of using such devices for mood recognition, focusing on work environments. We propose a novel mood recognition framework that is able to identify five intensity levels for eight different types of moods every two hours. We further present a smartphone app ('HealthyOffice'), designed to facilitate self-reporting in a structured manner and provide our model with the ground truth. We evaluate our system in a small-scale user study where wearable sensing data is collected in an office environment. Our experiments exhibit promising results allowing us to reliably recognize various classes of perceived moods. <s> BIB010 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> This paper presents results of analysis performed on a physiologic time-series dataset that was collected from a wearable ECG monitoring system worn by individuals who suffer from panic disorder. Models are constructed and evaluated for distinguishing between pathologic and non-pathologic states, including panic (during panic attack), pre-panic (preceding panic attack) and non-panic (outside panic attack window). The models presented use data fusion to combine both traditional time and frequency domain heart rate variability analysis together with nonlinear/complexity analysis. The best performing model is shown to be a random forest classifier that achieves an accuracy of 97.2% and 90.7% for recognizing states of panic and pre-panic, respectively. The models presented have application in pervasive and ubiquitous mobile and wearable health management systems. <s> BIB011 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> While accurately predicting mood and wellbeing could have a number of important clinical benefits, traditional machine learning (ML) methods frequently yield low performance in this domain. We posit that this is because a one-size-fits-all machine learning model is inherently ill-suited to predicting outcomes like mood and stress, which vary greatly due to individual differences. Therefore, we employ Multitask Learning (MTL) techniques to train personalized ML models which are customized to the needs of each individual, but still leverage data from across the population. Three formulations of MTL are compared: i) MTL deep neural networks, which share several hidden layers but have final layers unique to each task; ii) Multi-task Multi-Kernel learning, which feeds information across tasks through kernel weights on feature types; and iii) a Hierarchical Bayesian model in which tasks share a common Dirichlet Process prior. We offer the code for this work in open source. These techniques are investigated in the context of predicting future mood, stress, and health using data collected from surveys, wearable sensors, smartphone logs, and the weather. Empirical results demonstrate that using MTL to account for individual differences provides large performance improvements over traditional machine learning methods and provides personalized, actionable insights. <s> BIB012 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB013 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> In affective computing (AC) field studies it is impossible to obtain an objective ground truth. Hence, self-reports in form of ecological momentary assessments (EMAs) are frequently used in lieu of ground truth. Based on four paradigms, we formulate practical guidelines to increase the accuracy of labels generated via EMAs. In addition, we detail how these guidelines were implemented in a recent AC field study of ours. During our field study, 1081 EMAs were collected from 10 subjects over a duration of 148 days. Based on these EMAs, we perform a qualitative analysis of the effectiveness of our proposed guidelines. Furthermore, we present insights and lessons learned from the field study. <s> BIB014 </s> Wearable-Based Affect Recognition—A Review <s> Affect-Related User Studies in The Field <s> We present ASCERTAIN—a multimodal databa AS e for impli C it p ER sonali T y and A ffect recognit I o N using commercial physiological sensors. To our knowledge, ASCERTAIN is the first database to connect personality traits and emotional states via physiological responses . ASCERTAIN contains big-five personality scales and emotional self-ratings of 58 users along with their Electroencephalogram (EEG), Electrocardiogram (ECG), Galvanic Skin Response (GSR) and facial activity data, recorded using off-the-shelf sensors while viewing affective movie clips. We first examine relationships between users’ affective ratings and personality scales in the context of prior observations, and then study linear and non-linear physiological correlates of emotion and personality. Our analysis suggests that the emotion-personality relationship is better captured by non-linear rather than linear statistics. We finally attempt binary emotion and personality trait recognition using physiological features. Experimental results cumulatively confirm that personality differences are better revealed while comparing user responses to emotionally homogeneous videos, and above-chance recognition is achieved for both affective and personality dimensions. <s> BIB015

To develop affect-aware systems designed for everyday usage, data collection in the wild is essential. However, as the affective states occur naturally, the generation of a reliable ground truth has to be ensured differently. In this setting one can distinguish between questionnaires used in ecological-momentary-assessments (EMAs) and questionnaires employed during the pre-and post study phase. In the latter case constructs which are said to be constant for a longer time period (e.g., personality traits) are being queried. To assess the momentary affective state of a user, EMAs, also known as the experience sampling method, are employed. EMAs are a short set of questionnaires which the study participants file occasionally, to report their current affective state. Using EMAs, an important trade-off has to be considered. On one hand the affective state of the subject should be probed frequently. On the other hand, the subject should not be overloaded with questionnaires. The scheduling of EMAs can be either done interval-based (e.g., at certain/random times during the day) or event-triggered. In a study of Zenonos et al. BIB010 , for instance, the subjects were prompted every two hours during their working hours. The EMAs employed, inquired eight different moods, asking for each the question How have you been feeling for the last two hours?. Another approach is to distribute a defined number of EMAs randomly over a time period. Muaremi et al. BIB005 , for instance, divided the day into four sections and during each section subjects had to complete a randomly scheduled self-report. If the focus of a study lies on certain affective states or events, event-triggered self-reports can be utilized. In a study conducted by Hernandez et al. BIB004 call centre employees rated personal stress level after each call. Another example of event-based scheduling can be found by Rubin et al. BIB008 : Here subjects were asked to file an EMA once they became aware of the symptoms of a panic attack. In order to gain a deeper understanding of EMAs filed by the subjects daily screenings can be conducted BIB014 . Following Healey et al. BIB003 , these screenings can be used to correct/extend participants' annotations. Besides the frequency of EMAs, the length and complexity of each single questionnaire are also important factors defining the burden for the subjects. In order to avoid overloading study participants, EMAs should focus on the main goal of the study and their completion should require only little effort. In Table 5 questionnaires used during the pre-and post study as well as questionnaires employed in EMAs are displayed. As mentioned earlier the pre-and post study questionnaires, are used to aggregate information about longer time periods or traits of the subjects. Subjects' personality traits can have an influence on their affective perception and physiological response BIB015 . Therefore, completing a personality-related questionnaire can provide valuable insights. These Big Five Inventory (BFI) personality traits were, for instance, used by Sano et al. BIB009 as features for predicting subjects' mood. In addition, Taylor et al. BIB012 used personality traits to perform a groupwise personalization. Moreover, Wang et al. BIB007 used questionnaires assessing the mental health of their participants. For this purpose, the depression level (e.g., Patient Health Questionnaire (PHQ-9)) and loneliness level (UCLA loneliness scale) were recorded. As shown by Sano and Picard BIB006 , Sano et al. BIB009 , information on subjects' sleep quality can be useful in affect-related studies. The Pittsburgh Sleep Quality Index (PSQI), inquiring information about the past four weeks, can serve as a suitable questionnaire for sleep behaviour and quality assessment. In order to assess the overall stress level of the study participants the Perceived Stress Scale (PSS), measuring the perception and awareness of stress, can be employed. The PSS has been used in field studies (e.g., References BIB007 BIB006 ) and in ambulatory setting . The severity of stress-related symptoms can be scored using the Stress Response Inventory (SRI), or a simplified version of it, as shown by Kim et al. BIB002 . As detailed in Table 4 , wearable-based AR studies, typically rely on well-known psychological constructs. Hence, in order to generate labels using EMAs these constructs are employed, too. However, standard questionnaires are often quite long and as a result not really applicable in EMAs. In order to mitigate this issue, standard questionnaires can be shortened, for example, using only a subset of items with the highest factor loads on the targeted construct. Such an approach was, for instance, presented by Muaremi et al. BIB005 using a shortened version of the Positive and Negative Affect Schedule (PANAS) as EMA, which consisted of five positive affect items (relaxed, happy, concentrated, interested and active) and five negative affect items (tired, stressed, sleepy, angry and depressed). One particularly frequently employed construct is the valence-arousal space. In order to generate valence and arousal labels, Healey et al. BIB003 , for instance, used a tool called Mood Map, while Schmidt et al. BIB014 used the SAM. Furthermore, Wang et al. BIB007 used the Photo Affect Meter (PAM), assessing a similar construct. The PAM is implemented as smartphone app and the user selects from a set of 16 images the one that corresponds best to his/her current affective state. Zenonos et al. BIB010 provides an example for a custom EMA tool used for overall mood assessment: participants were asked to rate eight different moods on a scale from 0-100. The stress level of subjects can be assessed using a Likert-scale BIB013 BIB004 . Moreover, the severity of a certain event can be scored using its' symptoms. Rubin et al. BIB008 , for instance, aimed to quantify the severity of panic attacks. Hence, they created a questionnaire including 15 panic attack symptoms. In case a panic attack occurred, subjects were asked to rate the severity of each of the 15 symptoms, using a severity rating of 1 (none) to 5 (extreme). Hernandez et al. BIB004 Severity of panic attack symptoms Symptoms from the DSM-IV and Panic Disorder Severity Scale standard instrument 15 Shear et al. BIB001 Rubin et al. BIB008 Historically, personal notebooks or journals were used for EMAs. However, these tools have been predominantly replaced by smartphone apps, as they offer an ideal platform to facilitate self-reports: Subjects do not need to carry a study-specific device, EMAs are automatically scheduled and uploaded, and contextual information available on the smartphone can be logged together with the ground truth information. A key to both frequency and completeness of EMA is participant's motivation and using an appropriate reward system was proven to be beneficial: Participants of the study conducted by Healey et al. BIB003 received a base reward and an incremental reward, depending on the number of annotations made per day. Another reward structure was introduced by Wang et al. BIB007 : They offered all subjects a base reward, and the participants who completed most EMAs had the chance to win additional prizes. In Table 6 an overview of recent wearable-based AR field studies is provided and the employed EMAs as well as their scheduling is summarized. This table illustrates that commonly a combination of pre-/post-study questionnaires are used. The pre-/post-study questionnaires can be employed as additional features or to group the participants BIB002 BIB012 . In contrast, the data gathered via EMAs is often used as a subjective ground truth BIB013 BIB011 . Table 6 . Questionnaires employed during recent field studies, focusing on the applied scheduling (Pre-, During or Post-study).

Wearable-Based Affect Recognition—A Review <s> Author Employed Questionnaires and Their Scheduling <s> We conducted a 19 participant study using a system comprised of wireless galvanic skin response (GSR), heart rate (HR), activity sensors and a mobile phone for aggregating sensor data and enabling affect logging by the user. Each participant wore the sensors daily for five days, generating approximately 900 hours of continuous data. We found that analysis of emotional events was highly dependent on correct windowing and report results on synthesized windows around annotated events. Where raters agreed on the timing and quality of the emotion we were able to recognize 85% of the high and low energy emotions and 70% of the positive and negative emotions. We also gained many insights regarding participant's perception of their emotional state and the complexity of emotion in real life. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Author Employed Questionnaires and Their Scheduling <s> In this paper, we present first steps towards a mobile and wearable system intended to help people who experience regular and spontaneous panic attacks due to panic disorder. The goal of the system is to predict oncoming panic attacks and to deliver in-the-moment interventions on a smartphone device. Interventions are intended to reduce symptom severity by enabling a user to respond to approaching panic episodes. An initial feasibility study is described where a small real-world data set was collected. Personalized prediction models were trained which take, as input, physiological data and output a binary classification of either pre-panic or non-panic. We demonstrate proof-of-concept of episode prediction on this small dataset. <s> BIB002

Emotion Healey et al. BIB001 During study: Participants completed EMAs whenever they felt a change in their affective/physiological state. EMAs included a form of the circumplex model and a field for free text. Conducted Interviews at the end of each workday to generate additional labels and revision. Rubin et al. BIB002 During study: Start/stop time and severity ratings of 15 panic attack symptoms were reported by the subject using a mobile app.

Wearable-Based Affect Recognition—A Review <s> Sano and Picard [92] <s> Identifying episodes of significant stress is a challenging problem with implications for personal health and interface adaptation. We present the results of a study comparing multiple modalities of minimally intrusive stress sensing in real-world environments, collected from seven participants as they carried out their everyday activities over a ten-day period. We compare the data streams produced by sensors and self-report measures, in addition to asking the participants, themselves, to reflect on the accuracy and completeness of the data that had been collected. Finally, we describe the range of participant experiences---both positive and negative---as they reported their everyday stress levels. As a result of this study, we demonstrate that voice-based stress sensing tracks with electrodermal activity and self-reported stress measures in real-world environments and we identify limitations of various sensing approaches. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Sano and Picard [92] <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB002

Pre-study: Participants filled in a PSS, PSQI, and BFI. During study: Morning/evening EMAs on sleep, mood, stress level, health, and so forth. Post-study: Participants filled in questionnaires on health, mood, and stress. Adams et al. BIB001 Pre-study: Participants completed a PANAS, PSS, and a measure of mindfulness. During study: Self-reports approximately every 30 min. (with small random variations). Participants reported on momentary stress and affect. Additional reports and a small free text field were available too. Post-study: Semi-structured interview at the end of the end data collection. Hovsepian et al. During study: EMAs randomly scheduled approximately 15 times. During each EMA subjects filled in a shortened version of the PSS containing 6 items. Gjoreski et al. BIB002 During study: Subjects replied to 4 to 6 randomly scheduled EMAs. During each EMA subjects reported on their current stress level.

Wearable-Based Affect Recognition—A Review <s> Sampling rate: <s> We conducted a 19 participant study using a system comprised of wireless galvanic skin response (GSR), heart rate (HR), activity sensors and a mobile phone for aggregating sensor data and enabling affect logging by the user. Each participant wore the sensors daily for five days, generating approximately 900 hours of continuous data. We found that analysis of emotional events was highly dependent on correct windowing and report results on synthesized windows around annotated events. Where raters agreed on the timing and quality of the emotion we were able to recognize 85% of the high and low energy emotions and 70% of the positive and negative emotions. We also gained many insights regarding participant's perception of their emotional state and the complexity of emotion in real life. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Sampling rate: <s> Work should be a source of health, pride, and happiness, in the sense of enhancing motivation and strengthening personal development. Healthy and motivated employees perform better and remain loyal to the company for a longer time. But, when the person constantly experiences high workload over a longer period of time and is not able to recover, then work may lead to prolonged negative effects and might cause serious illnesses like chronic stress disease. In this work, we present a solution for assessing the stress experience of people, using features derived from smartphones and wearable chest belts. In particular, we use information from audio, physical activity, and communication data collected during workday and heart rate variability data collected at night during sleep to build multinomial logistic regression models. We evaluate our system in a real work environment and in daily-routine scenarios of 35 employees over a period of 4 months and apply the leave-one-day-out cross-validation method for each user individually to estimate the prediction accuracy. Using only smartphone features, we get an accuracy of 55 %, and using only heart rate variability features, we get an accuracy of 59 %. The combination of all features leads to a rate of 61 % for a three-stress level (low, moderate, and high perceived stress) classification problem. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Sampling rate: <s> Much of the stress and strain of student life remains hidden. The StudentLife continuous sensing app assesses the day-to-day and week-by-week impact of workload on stress, sleep, activity, mood, sociability, mental well-being and academic performance of a single class of 48 students across a 10 week term at Dartmouth College using Android phones. Results from the StudentLife study show a number of significant correlations between the automatic objective sensor data from smartphones and mental health and educational outcomes of the student body. We also identify a Dartmouth term lifecycle in the data that shows students start the term with high positive affect and conversation levels, low stress, and healthy sleep and daily activity patterns. As the term progresses and the workload increases, stress appreciably rises while positive affect, sleep, conversation and activity drops off. The StudentLife dataset is publicly available on the web. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Sampling rate: <s> Stress, anxiety and depression in the workplace are detrimental to human health and productivity with significant financial implications. Recent research in this area has focused on the use of sensor technologies, including smartphones and wearables embedded with physiological and movement sensors. In this work, we explore the possibility of using such devices for mood recognition, focusing on work environments. We propose a novel mood recognition framework that is able to identify five intensity levels for eight different types of moods every two hours. We further present a smartphone app ('HealthyOffice'), designed to facilitate self-reporting in a structured manner and provide our model with the ground truth. We evaluate our system in a small-scale user study where wearable sensing data is collected in an office environment. Our experiments exhibit promising results allowing us to reliably recognize various classes of perceived moods. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Sampling rate: <s> Continuous exposure to stress is harmful for mental and physical health, but to combat stress, one should first detect it. In this paper we propose a method for continuous detection of stressful events using data provided from a commercial wrist device. The method consists of three machine-learning components: a laboratory stress detector that detects short-term stress every 2 minutes; an activity recognizer that continuously recognizes user's activity and thus provides context information; and a context-based stress detector that exploits the output of the laboratory stress detector and the user's context in order to provide the final decision on 20 minutes interval. The method was evaluated in a laboratory and a real-life setting. The accuracy on 55 days of real-life data, for a 2-class problem, was 92%. The method is currently being integrated in a smartphone application for managing mental health and well-being. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Sampling rate: <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Sampling rate: <s> Affect recognition aims to detect a person's affective state based on observables, with the goal to e.g. improve human-computer interaction. Long-term stress is known to have severe implications on wellbeing, which call for continuous and automated stress monitoring systems. However, the affective computing community lacks commonly used standard datasets for wearable stress detection which a) provide multimodal high-quality data, and b) include multiple affective states. Therefore, we introduce WESAD, a new publicly available dataset for wearable stress and affect detection. This multimodal dataset features physiological and motion data, recorded from both a wrist- and a chest-worn device, of 15 subjects during a lab study. The following sensor modalities are included: blood volume pulse, electrocardiogram, electrodermal activity, electromyogram, respiration, body temperature, and three-axis acceleration. Moreover, the dataset bridges the gap between previous lab studies on stress and emotions, by containing three different affective states (neutral, stress, amusement). In addition, self-reports of the subjects, which were obtained using several established questionnaires, are contained in the dataset. Furthermore, a benchmark is created on the dataset, using well-known features and standard machine learning methods. Considering the three-class classification problem ( baseline vs. stress vs. amusement ), we achieved classification accuracies of up to 80%,. In the binary case ( stress vs. non-stress ), accuracies of up to 93%, were reached. Finally, we provide a detailed analysis and comparison of the two device locations ( chest vs. wrist ) as well as the different sensor modalities. <s> BIB007

When defining the number of scheduled EMAs over the observation period, the trade-off should between sampling as frequently as possible while not overloading the subject needs to be leveraged. A good compromise is to schedule an EMA every two hours BIB004 or approximately five times over the day BIB005 . 2. General scheduling: A good practice is to schedule EMAs randomly. This ensures that the subjects are unprepared. If the EMAs shall be distribute approximately evenly over the observation, the following approach could be used: Divide the observation period into N sections (where N is the total number of EMAs over the observation period), and randomly schedule one EMA within each section. This approach was applied for example by Muaremi et al. BIB002 . Considering user studies in the lab, EMAs are typically scheduled directly after each affective stimulus or condition BIB007 . 3. Manual trigger: As EMAs are commonly scheduled randomly during field studies, these questionnaires are independent of the participants' affective states. Therefore, it is good practice to allow subjects to file an EMA (in addition to the generally scheduled ones) whenever they feel a change in their affective state. For example, Gjoreski et al. BIB006 enabled their study participants to log stressful events whenever they occurred. 4. Number of items: In order to avoid overloading subjects, the time required to answer an EMA should be minimized. Therefore, EMAs should focused on the goal of the study and include a minimal number items. A good compromise is to include at most ten items per scheduled EMA, as discussed by Muaremi et al. BIB002 . Considering lab studies, the length of an EMA is usually less critical: Here EMAs can be used during the cool-down phase after an affective stimulus, which allows the completion of longer EMAs. 5. Situation labels: It is important to generate labels on the spot and not in hindsight. This is due to memorization effects (e.g., halo effect), where the occurrence of a certain emotion can influence the perception of other affective states experienced during the observation period. Considering a field study, however, it is good practice to review the labels together with the study participant, for example, on a daily basis BIB001 . 6. Length of labels: For a (mentally) healthy subject, affective states are expected to be stable on short time scales. However, when labels are generated using EMAs, the question arises how long these labels are valid. Considering lab studies, the labels generate using a questionnaire usually refer to the preceding stimulus (e.g., TSST). Considering field studies, however, the validity of labels is not as trivial. Depending on the focus of the study, one has to decide on a label length. If the study addresses mood, longer label periods, for example, 2 h BIB004 , can be taken into account. If the study targets shorter affective states (e.g., emotions or stress), shorter label periods are used. For example, in order to detect and classify stress, Gjoreski et al. BIB006 considered ten minutes before and after each provided label. 7. Ensure engagement: Considering field studies, subjects motivation is key and keeping the subjects motivated will ensure high-quality labels, regarding both frequency and completeness. One way to boost motivation is an appropriate (incremental) reward system BIB003 BIB001 . Another, way to increase subjects motivation might be to make the EMA optical appealing, for example, including graphical measures like the SAM or PAM.

Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> The ability to recognize emotion is one of the hallmarks of emotional intelligence, an aspect of human intelligence that has been argued to be even more important than mathematical and verbal intelligences. This paper proposes that machine intelligence needs to include emotional intelligence and demonstrates results toward this goal: developing a machine's ability to recognize the human affective state given four physiological signals. We describe difficult issues unique to obtaining reliable affective data and collect a large set of data from a subject trying to elicit and experience each of eight emotional states, daily, over multiple weeks. This paper presents and compares multiple algorithms for feature-based recognition of emotional state from this data. We analyze four physiological signals that exhibit problematic day-to-day variations: The features of different emotions on the same day tend to cluster more tightly than do the features of the same emotion on different days. To handle the daily variations, we propose new features and algorithms and compare their performance. We find that the technique of seeding a Fisher Projection with the results of sequential floating forward search improves the performance of the Fisher Projection and provides the highest recognition rates reported to date for classification of affect from physiology: 81 percent recognition accuracy on eight classes of emotion, including neutral. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> This paper presents methods for collecting and analyzing physiological data during real-world driving tasks to determine a driver's relative stress level. Electrocardiogram, electromyogram, skin conductance, and respiration were recorded continuously while drivers followed a set route through open roads in the greater Boston area. Data from 24 drives of at least 50-min duration were collected for analysis. The data were analyzed in two ways. Analysis I used features from 5-min intervals of data during the rest, highway, and city driving conditions to distinguish three levels of driver stress with an accuracy of over 97% across multiple drivers and driving days. Analysis II compared continuous features, calculated at 1-s intervals throughout the entire drive, with a metric of observable stressors created by independent coders from videotapes. The results show that for most drivers studied, skin conductivity and heart rate metrics are most closely correlated with driver stress level. These findings indicate that physiological signals can provide a metric of driver stress in future cars capable of physiological monitoring. Such a metric could be used to help manage noncritical in-vehicle information systems and could also provide a continuous measure of how different road and traffic conditions affect drivers. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> Autonomic nervous system (ANS) activity is viewed as a major component of the emotion response in many recent theories of emotion. Positions on the degree of specificity of ANS activation in emotion, however, greatly diverge, ranging from undifferentiated arousal, over acknowledgment of strong response idiosyncrasies, to highly specific predictions of autonomic response patterns for certain emotions. A review of 134 publications that report experimental investigations of emotional effects on peripheral physiological responding in healthy individuals suggests considerable ANS response specificity in emotion when considering subtypes of distinct emotions. The importance of sound terminology of investigated affective states as well as of choice of physiological measures in assessing ANS reactivity is discussed. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> We present a multimodal data set for the analysis of human affective states. The electroencephalogram (EEG) and peripheral physiological signals of 32 participants were recorded as each watched 40 one-minute long excerpts of music videos. Participants rated each video in terms of the levels of arousal, valence, like/dislike, dominance, and familiarity. For 22 of the 32 participants, frontal face video was also recorded. A novel method for stimuli selection is proposed using retrieval by affective tags from the last.fm website, video highlight detection, and an online assessment tool. An extensive analysis of the participants' ratings during the experiment is presented. Correlates between the EEG signal frequencies and the participants' ratings are investigated. Methods and results are presented for single-trial classification of arousal, valence, and like/dislike ratings using the modalities of EEG, peripheral physiological signals, and multimedia content analysis. Finally, decision fusion of the classification results from different modalities is performed. The data set is made publicly available and we encourage other researchers to use it for testing their own affective state estimation methods. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> MAHNOB-HCI is a multimodal database recorded in response to affective stimuli with the goal of emotion recognition and implicit tagging research. A multimodal setup was arranged for synchronized recording of face videos, audio signals, eye gaze data, and peripheral/central nervous system physiological signals. Twenty-seven participants from both genders and different cultural backgrounds participated in two experiments. In the first experiment, they watched 20 emotional videos and self-reported their felt emotions using arousal, valence, dominance, and predictability as well as emotional keywords. In the second experiment, short videos and images were shown once without any tag and then with correct or incorrect tags. Agreement or disagreement with the displayed tags was assessed by the participants. The recorded videos and bodily responses were segmented and stored in a database. The database is made available to the academic community via a web-based system. The collected data were analyzed and single modality and modality fusion results for both emotion recognition and implicit tagging experiments are reported. These results show the potential uses of the recorded modalities and the significance of the emotion elicitation protocol. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> Much of the stress and strain of student life remains hidden. The StudentLife continuous sensing app assesses the day-to-day and week-by-week impact of workload on stress, sleep, activity, mood, sociability, mental well-being and academic performance of a single class of 48 students across a 10 week term at Dartmouth College using Android phones. Results from the StudentLife study show a number of significant correlations between the automatic objective sensor data from smartphones and mental health and educational outcomes of the student body. We also identify a Dartmouth term lifecycle in the data that shows students start the term with high positive affect and conversation levels, low stress, and healthy sleep and daily activity patterns. As the term progresses and the workload increases, stress appreciably rises while positive affect, sleep, conversation and activity drops off. The StudentLife dataset is publicly available on the web. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> In this work, we present DECAF —a multimodal data set for dec oding user physiological responses to af fective multimedia content. Different from data sets such as DEAP [15] and MAHNOB-HCI [31] , DECAF contains (1) brain signals acquired using the Magnetoencephalogram (MEG) sensor, which requires little physical contact with the user’s scalp and consequently facilitates naturalistic affective response, and (2) explicit and implicit emotional responses of 30 participants to 40 one-minute music video segments used in [15] and 36 movie clips, thereby enabling comparisons between the EEG versus MEG modalities as well as movie versus music stimuli for affect recognition. In addition to MEG data, DECAF comprises synchronously recorded near-infra-red (NIR) facial videos, horizontal Electrooculogram (hEOG), Electrocardiogram (ECG), and trapezius-Electromyogram (tEMG) peripheral physiological responses. To demonstrate DECAF’s utility, we present (i) a detailed analysis of the correlations between participants’ self-assessments and their physiological responses and (ii) single-trial classification results for valence , arousal and dominance , with performance evaluation against existing data sets. DECAF also contains time-continuous emotion annotations for movie clips from seven users, which we use to demonstrate dynamic emotion prediction. <s> BIB007 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> Neurological assessment can be used to monitor a person's neurological status. In this paper, we report collection and analysis of a multimodal dataset of Non-EEG physiological signals available in the public domain. We have found this signal set useful for inferring the neurological status of individuals. The data was collected using non-invasive wrist worn biosensors and consists of electrodermal activity (EDA), temperature, acceleration, heart rate (HR), and arterial oxygen level (SpO2). We applied an efficient non-linear dimension reduction technique to visualize the biosignals in a low dimension feature space. We could cluster the four neurological statuses using an unsupervised Gaussian Mixture Model. The experimental results show that our unsupervised method can accurately separate different neurological statuses with an accuracy of greater than 84%. <s> BIB008 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> We present ASCERTAIN—a multimodal databa AS e for impli C it p ER sonali T y and A ffect recognit I o N using commercial physiological sensors. To our knowledge, ASCERTAIN is the first database to connect personality traits and emotional states via physiological responses . ASCERTAIN contains big-five personality scales and emotional self-ratings of 58 users along with their Electroencephalogram (EEG), Electrocardiogram (ECG), Galvanic Skin Response (GSR) and facial activity data, recorded using off-the-shelf sensors while viewing affective movie clips. We first examine relationships between users’ affective ratings and personality scales in the context of prior observations, and then study linear and non-linear physiological correlates of emotion and personality. Our analysis suggests that the emotion-personality relationship is better captured by non-linear rather than linear statistics. We finally attempt binary emotion and personality trait recognition using physiological features. Experimental results cumulatively confirm that personality differences are better revealed while comparing user responses to emotionally homogeneous videos, and above-chance recognition is achieved for both affective and personality dimensions. <s> BIB009 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> Affect recognition aims to detect a person's affective state based on observables, with the goal to e.g. improve human-computer interaction. Long-term stress is known to have severe implications on wellbeing, which call for continuous and automated stress monitoring systems. However, the affective computing community lacks commonly used standard datasets for wearable stress detection which a) provide multimodal high-quality data, and b) include multiple affective states. Therefore, we introduce WESAD, a new publicly available dataset for wearable stress and affect detection. This multimodal dataset features physiological and motion data, recorded from both a wrist- and a chest-worn device, of 15 subjects during a lab study. The following sensor modalities are included: blood volume pulse, electrocardiogram, electrodermal activity, electromyogram, respiration, body temperature, and three-axis acceleration. Moreover, the dataset bridges the gap between previous lab studies on stress and emotions, by containing three different affective states (neutral, stress, amusement). In addition, self-reports of the subjects, which were obtained using several established questionnaires, are contained in the dataset. Furthermore, a benchmark is created on the dataset, using well-known features and standard machine learning methods. Considering the three-class classification problem ( baseline vs. stress vs. amusement ), we achieved classification accuracies of up to 80%,. In the binary case ( stress vs. non-stress ), accuracies of up to 93%, were reached. Finally, we provide a detailed analysis and comparison of the two device locations ( chest vs. wrist ) as well as the different sensor modalities. <s> BIB010 </s> Wearable-Based Affect Recognition—A Review <s> Publicly Available Datasets <s> A growing number of pervasive systems integrate emotion recognition capabilities with the aim of fostering and promoting human health and wellbeing. In this paper, we argue that enhancing these systems with the ability to detect laughter episodes automatically would improve their effectiveness. This is because laughter is one of the most expressive behavioral cues for positive emotions. However, the existing approaches for laughter recognition rely on the use of obtrusive devices or of video and audio cues and have thus limited applicability in real-world settings. To overcome this limitation, we evaluate the feasibility of a novel, multi-modal approach to recognize laughter episodes using physiological and body movement data gathered with unobtrusive, wrist-worn devices. To assess the performance of our method, we collect an extensive data set of laughter episodes, which we also make publicly available. Our results show that laughter episodes can be distinguished from non-laughter episodes with an accuracy of 81%. Further, we demonstrate that the signatures left by laughter episodes on physiological and body-movement data differ significantly from those caused by slightly intense motions or cognitive load tasks. <s> BIB011

Conducting a user study is both a time consuming and a challenging task. However, there is a number of publicly available datasets. Depending on the research idea these datasets make the overhead of recording an own dataset obsolete. Furthermore, they facilitate benchmarking and allow a direct comparison of different approaches. Up-to-date the wearable-based AR community has only a handful of publicly available datasets containing data solely gathered via wearables. Therefore, we extend the scope of this section to datasets with a broader relevance to wearable AR. Below we present datasets which meet one of the following criteria: (a) being publicly available; (b) including data recorded from study participants being subject either to emotional stimuli or a stressor; and (c) including at least a few sensor modalities which can be (theoretically) integrated into consumer-grade wearables, which are applicable in everyday life. The datasets included in our analysis are summarized in Table 7 . Considering the population column in Table 7 it becomes apparent, that the data available originates mostly from a young cohort of subjects. Only the dataset recorded by Taamneh et al. , features two different age groups, namely an elderly (age > 60) and a young group (aged btw. 18 and 28) . This is certainly a limitation that needs to be considered when working with these datasets. Below we describe the datasets in detail. The Eight-Emotion dataset BIB001 includes the data of one (female) study participant who was subject to the same set of stimuli over a time span of 20 days. The stimuli, a set of personally-significant imagery, were chosen by the subject to elicit the affective states neutral, anger, hate, grief, platonic love, romantic love, joy, and reverence. The physiological signals (ECG, EDA, EMG, and RESP) were sampled at 20 Hz. Major limitations of this dataset are: (a) only one subject is included, and (b) due to the low sampling rate aliasing artefacts are likely to occur. DEAP (Database for Emotion Analysis using Physiological signals), recorded by Koelstra et al. BIB004 , features physiological data of 32 study participants. In DEAP, one minute excerpts of music videos were used as stimuli. In total 40 clips were selected from a larger pool according to valence, arousal, and dominance ratings gathered during a pre-study. The physiological signals were all sampled with 512 Hz and later downsampled to 256 Hz. DEAP includes subjects' ratings of the videos (valence, arousal, dominance, and liking). However, due to the employed protocol and the sensor setup, the DEAP participants were very limited in terms of movement. Therefore, one can expect that models trained on the DEAP dataset will have a limited performance in real-life settings. The MAHNOB-HCI dataset, includes physiological data from 27 study participants (16 female) BIB005 . The dataset includes face and body video data from six cameras, data from an eye gaze tracker, and audio. The physiological data (ECG, EDA, EEG, RESP, and TEMP) was sampled at 1024 Hz. Apart from EEG data, the physiological data was downsampled to 256 Hz. The MAHNOB-HCI dataset includes data from two experiments: First, study participants watched a set of 20 video clips, each associated with an emotional keyword (disgust, amusement, joy, fear, sadness, and neutral). The goal of the second experiment was implicit tagging: Subjects were exposed to 28 images and 14 videos, and reported on the agreement with the displayed tags. For the AR community, especially the first experiment is of interest. DECAF (DECoding user physiological responses to AFfective multimedia content) BIB007 was recorded in a laboratory setting with 30 subjects (14 female). The data recording consisted of two sessions for each subject, presenting music videos and movie clips, respectively. In the first session (music videos) the same set of clips as in DEAP were employed. For the second session, 36 movie clips were used as stimuli. From this pool of videos always nine correspond to a quadrant in the valence-arousal space. These 36 movie clips were selected from a larger pool during a pre-study based on valence-arousal ratings from 42 participants. For a detailed description, we refer the reader to Abadi et al. BIB007 . DECAF contains image (near-infrared face videos), magnetoencephalogram (MEG), and peripheral sensory data (ECG, EOG, and EMG). A clear limitation of DECAF is that, due to the MEG recordings, subjects were very restricted in their movements. Therefore, in contrast to real-life data DECAF is almost free from motion artefacts. In ASCERTAIN (multimodal databASe for impliCit pERsonaliTy and Affect recognitIoN using commercial physiological sensors) BIB009 , the same 36 movie clips as in DECAF were employed as stimuli. ASCERTAIN provides data from 58 subjects (21 female), and includes physiological modalities (ECG, EDA, EEG) as well as data recorded from a facial feature tracker. In addition, self-reports including arousal, valence, engagement, liking, and familiarity obtained for each video are included. Moreover, the dataset contains the Big Five personality traits for each subject. Hence, based on the recorded data, not only models predicting emotions can be created but also personality traits can be assessed. USI_Laughs has been recently published by Di Lascio et al. BIB011 . The dataset contains physiological data recorded from 34 participants (6 female) recorded via an Empatica E4 smartwatch (ACC, EDA, PPG, TEMP). Similar to prior work, funny clips were used to induce laughter. Following Di Lascio et al. BIB011 the main aim of the dataset is to facilitate the detection of laughter episodes based on physiological data. Here, the laughter episodes are to be considered as surrogate to positive emotions. The Driver stress dataset BIB002 includes physiological data (ECG, EDA, EMG, RESP) from 24 participants. The dataset was recorded during one rest condition and two driving tasks (city streets and on a highway near Boston, Massachusetts). Depending on traffic the two driving tasks had a duration between 50 and 90 min. Using questionnaires and a score derived from observable events, the three study conditions (rest, highway, city) were mapped onto the stress levels low, medium, and high. Therefore, the dataset facilitates the development of real-life stress monitoring approaches. However, one limitation of the dataset is that the data was acquired at low sampling rates (e.g., EMG 15.5 Hz). Distracted Driving, recorded by Taamneh et al. , includes multimodal (physiological and eye tracking) data from 68 subjects driving in a simulator on a highway. All participants were subject to four different distractions: no, emotional, cognitive, and sensorimotor distraction. As the dataset includes among other modalities EDA, heart and respiration rate. This data can be used to study the influence of different distractions on these parameter. Non-EEG BIB008 is a dataset containing physiological data (EDA, HR, TEMP, SpO2, and ACC) from 20 subjects (4 female). The dataset was recorded during three different stress conditions (physical, cognitive, and emotional) and a relaxation task. Physical stress was evoked by asking the subjects to jog on a treadmill at three miles per hour. In order to elicit cognitive stress, the subjects had to count backwards from 2485 doing steps of seven. Lastly, emotional stress was triggered by anticipating and watching a clip from a zombie apocalypse movie. This dataset is particularly interesting as it contains only wearable-based data. Although the data collection was conducted in a lab setting, the subjects were (compared to the other datasets) less motion constrained due to the minimally intrusive nature of the sensors. However, a major limitation of the Non-EEG dataset is the low sampling rate of the employed devices (1 Hz and 8 Hz). In addition, as no ECG or PPG data was recorded, the HRV information can not be retrieved, a parameter shown to be relevant for stress recognition by various previous work (e.g., Kreibig BIB003 ). StudentLife BIB006 contains data from 48 college students (10 female). All participants were monitored over one academic semester (10 weeks). Unlike the afore described datasets StudentLife was recorded in the field. Considering the progress of the semester, it is expected that the students were more stressed towards the end of the data collection. This can be attributed to the examination period. StudentLife contains data recorded from the students' smartphones (e.g., ACC, microphone, light sensor, and GPS/Bluetooth data). Moreover, various information related to the students' context (e.g., class attendance) and smartphone usage (e.g., conversation frequency and duration) were recorded. In addition, StudentLife includes a large number of self-reports targeting physical activity, sleep, perceived stress, mood, mental well-being, and so forth. Due to the popularity of smartphones, the dataset is certainly of interest by facilitating affect and stress recognition purely based on smartphone usage patterns.' However, a drawback of the StudentLife is that it does not include any physiological data. WESAD (dataset for WEarable Stress and Affect Detection) is, to the best of our knowledge, the only publicly available dataset which contains data of subjects experiencing both an emotional and a stress stimulus BIB010 . WESAD includes data from 15 subjects (3 female) recorded in a laboratory setting. Each subject experienced three conditions: baseline (neutral reading task), amusement (watching a set of funny video clips), and stress (being exposed to the TSST). WESAD features physiological and motion data, recorded from both a wrist-and a chest-worn device. The following sensor modalities are included: ECG, PPG, EDA, EMG, RESP, TEMP, and ACC. Moreover, the high sampling rate (700 Hz) of the chest-worn device should be emphasised. Overall, WESAD is a fitting dataset for benchmarking affect and stress recognition algorithms based on physiological data.

Wearable-Based Affect Recognition—A Review <s> 3-axes acceleration Preprocessing: <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> 3-axes acceleration Preprocessing: <s> Stress remains a significant social problem for individuals in modern societies. This paper presents a machine learning approach for the automatic detection of stress of people in a social situation by combining two sensor systems that capture physiological and social responses. We compare the performance using different classifiers including support vector machine, AdaBoost, and k-nearest neighbour. Our experimental results show that by combining the measurements from both sensor systems, we could accurately discriminate between stressful and neutral situations during a controlled Trier social stress test (TSST). Moreover, this paper assesses the discriminative ability of each sensor modality individually and considers their suitability for real time stress detection. Finally, we present an study of the most discriminative features for stress detection. <s> BIB002

A detailed analysis of preprocessing applied to ACC data can be found in Figo et al. . In AR, the ACC data is often considered as a surrogate for the performed activity BIB001 BIB002 .

Wearable-Based Affect Recognition—A Review <s> Electrocardiogram Preprocessing: <s> We have developed a real-time algorithm for detection of the QRS complexes of ECG signals. It reliably recognizes QRS complexes based upon digital analyses of slope, amplitude, and width. A special digital bandpass filter reduces false detections caused by the various types of interference present in ECG signals. This filtering permits use of low thresholds, thereby increasing detection sensitivity. The algorithm automatically adjusts thresholds and parameters periodically to adapt to such ECG changes as QRS morphology and heart rate. For the standard 24 h MIT/BIH arrhythmia database, this algorithm correctly detects 99.3 percent of the QRS complexes. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Electrocardiogram Preprocessing: <s> A photoplethysmography (PPG) signal provides very useful information about a subject's hemodynamic status in a hospital or ubiquitous environment. However, PPG is very vulnerable to motion artifacts, which can significantly distort the information belonging to the PPG signal itself. Thus, the reduction of the effects of motion artifacts is an important issue when monitoring the cardiovascular system by PPG. There have been many adaptive techniques to reduce motion artifacts from PPG signals. In the present study, we compared a method based on the fixed-interval Kalman smoother with the usual adaptive filtering algorithms, e.g. the normalized least mean squares, recursive least squares and the conventional Kalman filter. We found that the fixed-interval Kalman smoother reduced motion artifacts from the PPG signal most effectively. Therefore, the use of the fixed-interval Kalman smoother can reduce motion artifacts in PPG, thus providing the most reliable information that can be deduced from the reconstructed PPG signals. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Electrocardiogram Preprocessing: <s> Photoplethysmography (PPG) is used to estimate the skin blood flow using infrared light. Researchers from different domains of science have become increasingly interested in PPG because of its advantages as non-invasive, inexpensive, and convenient diagnostic tool. Traditionally, it measures the oxygen saturation, blood pressure, cardiac output, and for assessing autonomic functions. Moreover, PPG is a promising technique for early screening of various atherosclerotic pathologies and could be helpful for regular GP-assessment but a full understanding of the diagnostic value of the different features is still lacking. Recent studies emphasise the potential information embedded in the PPG waveform signal and it deserves further attention for its possible applications beyond pulse oximetry and heart-rate calculation. Therefore, this overview discusses different types of artifact added to PPG signal, characteristic features of PPG waveform, and existing indexes to evaluate for diagnoses. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Electrocardiogram Preprocessing: <s> The performance of pulse oximeters is highly influenced by motion artifacts (MAs) in photoplethysmographic (PPG) signals. In this paper, we propose a simple and efficient approach based on adaptive step-size least mean squares (AS-LMS) adaptive filter for reducing MA in corrupted PPG signals. The presented method is an extension to our prior work on efficient use of adaptive filters for reduction of MA in PPG signals. The novelty of the method lies in the fact that a synthetic noise reference signal for an adaptive filtering process, representing MA noise, is generated internally from the MA-corrupted PPG signal itself instead of using any additional hardware such as accelerometer or source-detector pair for acquiring noise reference signal. Thus, the generated noise reference signal is then applied to the AS-LMS adaptive filter for artifact removal. While experimental results proved the efficacy of the proposed scheme, the merit of the method is clearly demonstrated using convergence and correlation analysis, thus making it best suitable for present-day pulse oximeters utilizing PPG sensor head with a single pair of source and detector, which does not have any extra hardware meant for capturing noise reference signal. In addition to arterial oxygen saturation estimation, the artifact reduction method facilitated the waveform contour analysis on artifact-reduced PPG, and the conventional parameters were evaluated for assessing the arterial stiffness. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Electrocardiogram Preprocessing: <s> In this work, we describe a beat-by-beat method for assessing the clinical utility of pulsatile waveforms, primarily recorded from cardiovascular blood volume or pressure changes, concentrating on the photoplethysmogram (PPG). Physiological blood flow is nonstationary, with pulses changing in height, width and morphology due to changes in heart rate, cardiac output, sensor type and hardware or software pre-processing requirements. Moreover, considerable inter-individual and sensor-location variability exists. Simple template matching methods are therefore inappropriate, and a patient-specific adaptive initialization is therefore required. We introduce dynamic time warping to stretch each beat to match a running template and combine it with several other features related to signal quality, including correlation and the percentage of the beat that appeared to be clipped. The features were then presented to a multi-layer perceptron neural network to learn the relationships between the parameters in the presence of good- and bad-quality pulses. An expert-labeled database of 1055 segments of PPG, each 6 s long, recorded from 104 separate critical care admissions during both normal and verified arrhythmic events, was used to train and test our algorithms. An accuracy of 97.5% on the training set and 95.2% on test set was found. The algorithm could be deployed as a stand-alone signal quality assessment algorithm for vetting the clinical utility of PPG traces or any similar quasi-periodic signal. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Electrocardiogram Preprocessing: <s> An automated algorithm to assess electrocardiogram (ECG) quality for both normal and abnormal rhythms is presented for false arrhythmia alarm suppression of intensive care unit (ICU) monitors. A particular focus is given to the quality assessment of a wide variety of arrhythmias. Data from three databases were used: the Physionet Challenge 2011 dataset, the MIT-BIH arrhythmia database, and the MIMIC II database. The quality of more than 33 000 single-lead 10 s ECG segments were manually assessed and another 12 000 bad-quality single-lead ECG segments were generated using the Physionet noise stress test database. Signal quality indices (SQIs) were derived from the ECGs segments and used as the inputs to a support vector machine classifier with a Gaussian kernel. This classifier was trained to estimate the quality of an ECG segment. Classification accuracies of up to 99% on the training and test set were obtained for normal sinus rhythm and up to 95% for arrhythmias, although performance varied greatly depending on the type of rhythm. Additionally, the association between 4050 ICU alarms from the MIMIC II database and the signal quality, as evaluated by the classifier, was studied. Results suggest that the SQIs should be rhythm specific and that the classifier should be trained for each rhythm call independently. This would require a substantially increased set of labeled data in order to train an accurate algorithm. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Electrocardiogram Preprocessing: <s> Accurate estimation of heart rates from photoplethysmogram (PPG) signals during intense physical activity is a very challenging problem. This is because strenuous and high intensity exercise can result in severe motion artifacts in PPG signals, making accurate heart rate (HR) estimation difficult. In this study we investigated a novel technique to accurately reconstruct motion-corrupted PPG signals and HR based on time-varying spectral analysis. The algorithm is called Spectral filter algorithm for Motion Artifacts and heart rate reconstruction (SpaMA). The idea is to calculate the power spectral density of both PPG and accelerometer signals for each time shift of a windowed data segment. By comparing time-varying spectra of PPG and accelerometer data, those frequency peaks resulting from motion artifacts can be distinguished from the PPG spectrum. The SpaMA approach was applied to three different datasets and four types of activities: (1) training datasets from the 2015 IEEE Signal Process. Cup Database recorded from 12 subjects while performing treadmill exercise from 1 km/h to 15 km/h; (2) test datasets from the 2015 IEEE Signal Process. Cup Database recorded from 11 subjects while performing forearm and upper arm exercise. (3) Chon Lab dataset including 10 min recordings from 10 subjects during treadmill exercise. The ECG signals from all three datasets provided the reference HRs which were used to determine the accuracy of our SpaMA algorithm. The performance of the SpaMA approach was calculated by computing the mean absolute error between the estimated HR from the PPG and the reference HR from the ECG. The average estimation errors using our method on the first, second and third datasets are 0.89, 1.93 and 1.38 beats/min respectively, while the overall error on all 33 subjects is 1.86 beats/min and the performance on only treadmill experiment datasets (22 subjects) is 1.11 beats/min. Moreover, it was found that dynamics of heart rate variability can be accurately captured using the algorithm where the mean Pearson’s correlation coefficient between the power spectral densities of the reference and the reconstructed heart rate time series was found to be 0.98. These results show that the SpaMA method has a potential for PPG-based HR monitoring in wearable devices for fitness tracking and health monitoring during intense physical activities. <s> BIB007 </s> Wearable-Based Affect Recognition—A Review <s> Electrocardiogram Preprocessing: <s> Photoplethysmography (PPG) is a low-cost, non-invasive, and optical technique used to detect blood volume changes in the microvascular tissue bed, measured from the skin surface. It has traditionally been used in commercial medical devices for oxygen saturation, blood pressure monitoring, and cardiac activity for assessing peripheral vascular disease and autonomic function. There has been a growing interest to incorporate PPG sensors in daily life, capable of use in ambulatory settings. However, inferring cardiac information (e.g. heart rate) from PPG traces in such situations is extremely challenging, because of interferences caused by motion. Following the IEEE Signal Processing Cup in 2015, numerous methods have been proposed for estimating particularly the average heart rate using wrist-worn PPG during physical activity. Details on PPG technology, sensor development, and applications have been well documented in the literature. Hence, in this paper, we have presented a comprehensive review of state-of-the-art research on heart rate estimation from wrist-worn PPG signals. Our review also encompasses brief theoretical details about PPG sensing and other potential applications–biometric identification, disease diagnosis using wrist PPG. This paper will set a platform for future research on pervasive monitoring using wrist PPG. <s> BIB008 </s> Wearable-Based Affect Recognition—A Review <s> Electrocardiogram Preprocessing: <s> Photoplethysmography (PPG)-based continuous heart rate monitoring is essential in a number of domains, e.g., for healthcare or fitness applications. Recently, methods based on time-frequency spectra emerged to address the challenges of motion artefact compensation. However, existing approaches are highly parametrised and optimised for specific scenarios of small, public datasets. We address this fragmentation by contributing research into the robustness and generalisation capabilities of PPG-based heart rate estimation approaches. First, we introduce a novel large-scale dataset (called PPG-DaLiA), including a wide range of activities performed under close to real-life conditions. Second, we extend a state-of-the-art algorithm, significantly improving its performance on several datasets. Third, we introduce deep learning to this domain, and investigate various convolutional neural network architectures. Our end-to-end learning approach takes the time-frequency spectra of synchronised PPG- and accelerometer-signals as input, and provides the estimated heart rate as output. Finally, we compare the novel deep learning approach to classical methods, performing evaluation on four public datasets. We show that on large datasets the deep learning model significantly outperforms other methods: The mean absolute error could be reduced by 31 % on the new dataset PPG-DaLiA, and by 21 % on the dataset WESAD. <s> BIB009

In the raw ECG signal the R-peaks need to be identified. For this purpose, the Pan and Tompkin's algorithm can be applied BIB001 . Once the R-peaks have been detected, the next step is to determine the RR intervals and assess their validity. For example, Hovsepian et al. present an algorithm to assess the validity of candidate RR intervals. Behar et al. BIB006 , presented an approach to assess the ECG signal quality in regards to arrhythmia in the context of intensive care units. Similar approaches could also be utilized to assess the ECG quality during affect-related user studies. 3. Photoplethysmogram Preprocessing: A detailed description on PPG signal preprocessing methods applied to PPG data can be found in Elgendi BIB003 or Biswas et al. BIB008 . In order to remove motion artefacts, adaptive (filtering) approaches can be applied BIB002 BIB004 . In more recent work, peak matching approaches in the spectral domain were employed to remove movement artefacts BIB009 BIB007 . For the determination of RR intervals from identified R-peaks, similar algorithms as mentioned with ECG preprocessing can be applied. In addition, as shown by Li and Clifford BIB005 , the quality of a PPG signal can be assessed using a combination of dynamical time warping and multilayer perceptron.

Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> Human activity is one of the most important ingredients of context information. In wearable computing scenarios, activities such as walking, standing and sitting can be inferred from data provided by body-worn acceleration sensors. In such settings, most approaches use a single set of features, regardless of which activity to be recognized. In this paper we show that recognition rates can be improved by careful selection of individual features for each activity. We present a systematic analysis of features computed from a real-world data set and show how the choice of feature and the window length over which the feature is computed affects the recognition rates for different activities. Finally, we give a recommendation of suitable features and window lengths for a set of common activities. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> This paper proposed a multimodal fusion between brain and peripheral signals for emotion detection. The input signals were electroencephalogram, galvanic skin resistance, temperature, blood pressure and respiration, which can reflect the influence of emotion on the central nervous system and autonomic nervous system respectively. The acquisition protocol is based on a subset of pictures which correspond to three specific areas of valance-arousal emotional space (positively excited, negatively excited, and calm). The features extracted from input signals, and to improve the results, correlation dimension as a strong nonlinear feature is used for brain signals. The performance of the Quadratic Discriminant Classifier has been evaluated on different feature sets: peripheral signals, EEG's, and both. In comparison among the results of different feature sets, EEG signals seem to perform better than other physiological signals, and the results confirm the interest of using brain signals as peripherals in emotion assessment. According to the improvement in EEG results compare in each raw of the table, it seems that nonlinear features would lead to better understanding of how emotional activities work. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> The inferred cost of work-related stress call for prevention strategies that aim at detecting early warning signs at the workplace. This paper goes one step towards the goal of developing a personal health system for detecting stress. We analyze the discriminative power of electrodermal activity (EDA) in distinguishing stress from cognitive load in an office environment. A collective of 33 subjects underwent a laboratory intervention that included mild cognitive load and two stress factors, which are relevant at the workplace: mental stress induced by solving arithmetic problems under time pressure and psychosocial stress induced by social-evaluative threat. During the experiments, a wearable device was used to monitor the EDA as a measure of the individual stress reaction. Analysis of the data showed that the distributions of the EDA peak height and the instantaneous peak rate carry information about the stress level of a person. Six classifiers were investigated regarding their ability to discriminate cognitive load from stress. A maximum accuracy of 82.8% was achieved for discriminating stress from cognitive load. This would allow keeping track of stressful phases during a working day by using a wearable EDA device. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> Skin conductance (SC) data are usually characterized by a sequence of overlapping phasic skin conductance responses (SCRs) overlying a tonic component. The variability of SCR shapes hereby complicates the proper decomposition of SC data. A method is proposed for full decomposition of SC data into tonic and phasic components. A two-compartment diffusion model was found to adequately describe a standard SCR shape based on the process of sweat diffusion. Nonnegative deconvolution is used to decompose SC data into discrete compact responses and at the same time assess deviations from the standard SCR shape, which could be ascribed to the additional process of pore opening. Based on the result of single non-overlapped SCRs, response parameters can be estimated precisely as shown in a paradigm with varying inter-stimulus intervals. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> Stress is a growing problem in society and can, amongst others, induce musculoskeletal complaints, related to sustained muscle tension. The ability to measure stress with a wireless system would be useful in the prevention of stress-related health problems. The aim of this experiment was to derive stress levels of subjects from electromyography (EMG) signals of the upper trapezius muscle. Two new stress tests were designed for this study, which aimed at creating circumstances that are similar to work stress. An experiment is described in which EMG signals of the upper trapezius muscle were measured during three different stressful situations. Stress tests included a calculation task (the Norinder test), a logical puzzle task and a memory task, of which the last two were newly designed. The results show significantly higher amplitudes of the EMG signals during stress compared to rest and fewer gaps (periods of relaxation) during stress. Also, mean and median frequencies were significantly lower during stress than during rest. The differences in EMG features between rest and stress conditions indicate that EMG is a useful parameter to detect stress. These results show opportunities for the inclusion of EMG sensors in a wireless system for ambulatory monitoring of stress levels. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> We conducted a 19 participant study using a system comprised of wireless galvanic skin response (GSR), heart rate (HR), activity sensors and a mobile phone for aggregating sensor data and enabling affect logging by the user. Each participant wore the sensors daily for five days, generating approximately 900 hours of continuous data. We found that analysis of emotional events was highly dependent on correct windowing and report results on synthesized windows around annotated events. Where raters agreed on the timing and quality of the emotion we were able to recognize 85% of the high and low energy emotions and 70% of the positive and negative emotions. We also gained many insights regarding participant's perception of their emotional state and the complexity of emotion in real life. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> Autonomic nervous system (ANS) activity is viewed as a major component of the emotion response in many recent theories of emotion. Positions on the degree of specificity of ANS activation in emotion, however, greatly diverge, ranging from undifferentiated arousal, over acknowledgment of strong response idiosyncrasies, to highly specific predictions of autonomic response patterns for certain emotions. A review of 134 publications that report experimental investigations of emotional effects on peripheral physiological responding in healthy individuals suggests considerable ANS response specificity in emotion when considering subtypes of distinct emotions. The importance of sound terminology of investigated affective states as well as of choice of physiological measures in assessing ANS reactivity is discussed. <s> BIB007 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> MAHNOB-HCI is a multimodal database recorded in response to affective stimuli with the goal of emotion recognition and implicit tagging research. A multimodal setup was arranged for synchronized recording of face videos, audio signals, eye gaze data, and peripheral/central nervous system physiological signals. Twenty-seven participants from both genders and different cultural backgrounds participated in two experiments. In the first experiment, they watched 20 emotional videos and self-reported their felt emotions using arousal, valence, dominance, and predictability as well as emotional keywords. In the second experiment, short videos and images were shown once without any tag and then with correct or incorrect tags. Agreement or disagreement with the displayed tags was assessed by the participants. The recorded videos and bodily responses were segmented and stored in a database. The database is made available to the academic community via a web-based system. The collected data were analyzed and single modality and modality fusion results for both emotion recognition and implicit tagging experiments are reported. These results show the potential uses of the recorded modalities and the significance of the emotion elicitation protocol. <s> BIB008 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> Chronic stress is endemic to modern society. However, as it is unfeasible for physicians to continuously monitor stress levels, its diagnosis is nontrivial. Wireless body sensor networks offer opportunities to ubiquitously detect and monitor mental stress levels, enabling improved diagnosis, and early treatment. This article describes the development of a wearable sensor platform to monitor a number of physiological correlates of mental stress. We discuss tradeoffs in both system design and sensor selection to balance information content and wearability. Using experimental signals collected from the wearable sensor, we describe a selected number of physiological features that show good correlation with mental stress. In particular, we propose a new spectral feature that estimates the balance of the autonomic nervous system by combining information from the power spectral density of respiration and heart rate variability. We validate the effectiveness of our approach on a binary discrimination problem when subjects are placed under two psychophysiological conditions: mental stress and relaxation. When used in a logistic regression model, our feature set is able to discriminate between these two mental states with a success rate of 81% across subjects. <s> BIB009 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> Abstract Trunk muscle electromyography (EMG) is often contaminated by the electrocardiogram (ECG), which hampers data analysis and potentially yields misinterpretations. We propose the use of independent component analysis (ICA) for removing ECG contamination and compared it with other procedures previously developed to decontaminate EMG. To mimic realistic contamination while having uncontaminated reference signals, we employed EMG recordings from peripheral muscles with different activation patterns and superimposed distinct ECG signals that were recorded during rest at conventional locations for trunk muscle EMG. ICA decomposition was performed with and without a separately collected ECG signal as part of the data set and contaminated ICA modes representing ECG were identified automatically. Root mean squared relative errors and correlations between the linear envelopes of uncontaminated and contaminated EMG were calculated to assess filtering effects on EMG amplitude. Changes in spectral content were quantified via mean power frequencies. ICA-based filtering largely preserved the EMG's spectral content. Performance on amplitude measures was especially successful when a separate ECG recording was included. That is, the ICA-based filtering can produce excellent results when EMG and ECG are indeed statistically independent and when mode selection is flexibly adjusted to the data set under study. <s> BIB010 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> Recently, wearable devices have allowed for long term, ambulatory measurement of electrodermal activity (EDA). Despite the fact that ambulatory recording can be noisy, and recording artifacts can easily be mistaken for a physiological response during analysis, to date there is no automatic method for detecting artifacts. This paper describes the development of a machine learning algorithm for automatically detecting EDA artifacts, and provides an empirical evaluation of classification performance. We have encoded our results into a freely available web-based tool for artifact and peak detection. <s> BIB011 </s> Wearable-Based Affect Recognition—A Review <s> Electrodermal activity Preprocessing: <s> Goal: This paper reports on a novel algorithm for the analysis of electrodermal activity (EDA) using methods of convex optimization. EDA can be considered as one of the most common observation channels of sympathetic nervous system activity, and manifests itself as a change in electrical properties of the skin, such as skin conductance (SC). Methods: The proposed model describes SC as the sum of three terms: the phasic component, the tonic component, and an additive white Gaussian noise term incorporating model prediction errors as well as measurement errors and artifacts. This model is physiologically inspired and fully explains EDA through a rigorous methodology based on Bayesian statistics, mathematical convex optimization, and sparsity. Results: The algorithm was evaluated in three different experimental sessions to test its robustness to noise, its ability to separate and identify stimulus inputs, and its capability of properly describing the activity of the autonomic nervous system in response to strong affective stimulation. Significance: Results are very encouraging, showing good performance of the proposed method and suggesting promising future applicability, e.g., in the field of affective computing. <s> BIB012

In order to remove artefacts from EDA data different approaches were presented. The approaches can be grouped into filtering and machine learning-based approaches. Only changes in the low-frequency domain of the EDA signal are physiologically plausible. Hence, low-pass filtering with a cut-off of, for example, 5 Hz BIB003 can be applied to remove high-frequency noise. After the noise removal, for example, Soleymani et al. BIB008 , detrended the EDA signal by subtracting a moving average, computed on smoothed version of the signal. Machine learning-based approaches, using support vector machines or convex optimization, to identify and remove artefacts in EDA data can be found in Taylor et al. BIB011 , Greco et al. BIB012 . As detailed in Section 3, the EDA signal consists of two components: A slowly varying baseline conductivity referred to as skin conductance level (SCL) and a series of peaks referred to as skin conductance response (SCR). In literature different approaches to separate these two components can be found: Benedek and Kaernbach BIB004 , for instance, present an approach to separate SCL and SCR relying on nonnegative devolution. Alternatively, Choi et al. BIB009 utilized, a regularized least-squares detrending method, to separate the two components. 5. Electromyogram Preprocessing: Raw EMG data is often filtered to remove noise. For example, Wijsman et al. BIB005 report on a two step procedure. First, a bandpass filter, allowing frequencies from 20 to 450 Hz, was applied. Then, in order to remove residual power line interference from data, notch filters were applied. The notch filters attenuated the 50, 100, 150, 200, 250, and 350 Hz components of the signal. Cardiac artefacts are another common source of noise in EMG data. Hence, Willigenburg et al. BIB010 propose and compare different filtering procedures to remove ECG interference from the EMG signal. 6. Respiration Preprocessing: Depending on the signal quality, noise removal filtering techniques (e.g., bandpass filter with cut-off frequencies at 0.1 and 0.35 Hz) have to be applied. In addition, the raw respiration signal can be detrended by subtracting a moving average BIB002 . In the classical processing chain these preprocessing steps are followed by the segmentation. During this procedure the data is segmented using a sliding window of fixed size. The appropriate window size is crucial and depends on several aspects, such as the classification task or the considered sensor modality. Below appropriate choices for the window length of motion (ACC) and physiological data will be provided. In human activity recognition (HAR), ACC data is most frequently employed to detect activities and there exists a body of work, identifying appropriate window sizes for HAR BIB006 BIB001 153] . A common finding is that in HAR the relevant patterns occur on short time scales. Therefore, window sizes of ∼5 s are common. The time scales on which physiological responses to emotional stimuli occur are hard to define. Hence, considering physiological signals, finding an appropriate window size is difficult BIB006 . Moreover, due to inter-subject and inter-modality (e.g., ECG vs. EDA) differences, defining an appropriate window size becomes even more challenging. However, a meta analysis conducted by Kreibig BIB007 found that physiological features are commonly aggregated over fixed window lengths of 30 to 60 s.

Wearable-Based Affect Recognition—A Review <s> ACC-based Feature <s> Automatic estimation of physical activity using wearable sensors can be used for promotion of a healthier lifestyle. In this study, accelerometers and gyroscopes attached to ankle, wrist and hip were used to estimate intensity of physical activity. The estimates are compared to metabolic equivalent (MET) obtained from a portable cardiopulmonary exercise testing system. Data from common everyday tasks and exercise were collected with 11 subjects. The tasks include, e.g., ironing, vacuuming, walking, running and cycling on exercise bicycle (ergometer). The strongest linear correlation with metabolic equivalent was obtained with the tri-axial accelerometer attached to the ankle (r=0.86). <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> ACC-based Feature <s> Stress remains a significant social problem for individuals in modern societies. This paper presents a machine learning approach for the automatic detection of stress of people in a social situation by combining two sensor systems that capture physiological and social responses. We compare the performance using different classifiers including support vector machine, AdaBoost, and k-nearest neighbour. Our experimental results show that by combining the measurements from both sensor systems, we could accurately discriminate between stressful and neutral situations during a controlled Trier social stress test (TSST). Moreover, this paper assesses the discriminative ability of each sensor modality individually and considers their suitability for real time stress detection. Finally, we present an study of the most discriminative features for stress detection. <s> BIB002

From the HAR domain, a large set of ACC-based features is known. These features are often also employed in AR. Statistical features (mean, median, standard deviation, etc.) are often computed for each channel (x, y, z) separately and combined. Parkka et al. BIB001 showed that the absolute integral of acceleration can be used to estimate the metabolic equivalent of physical activities, which can be an interesting feature for affect recognition as well. Mozos et al. BIB002 used the first and second derivative of the accelerometer's energy as feature, for example, to indicate the direction of change in activity level. Considering frequency-domain features, the power ratio of certain defined frequency bands, the peak frequency or the entropy of the power spectral density have been applied successfully. Table 8 . Features commonly extracted and applied in the wearable-based AR.

Wearable-Based Affect Recognition—A Review <s> ACC <s> Automatic estimation of physical activity using wearable sensors can be used for promotion of a healthier lifestyle. In this study, accelerometers and gyroscopes attached to ankle, wrist and hip were used to estimate intensity of physical activity. The estimates are compared to metabolic equivalent (MET) obtained from a portable cardiopulmonary exercise testing system. Data from common everyday tasks and exercise were collected with 11 subjects. The tasks include, e.g., ironing, vacuuming, walking, running and cycling on exercise bicycle (ergometer). The strongest linear correlation with metabolic equivalent was obtained with the tri-axial accelerometer attached to the ankle (r=0.86). <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> ACC <s> In mobile computing, varying access to resources makes it difficult for developers to ensure that satisfactory system response times will be maintained at all times. Wearable physiological sensors offer a way to dynamically detect user frustration in response to increased system delays. However, most prior efforts have focused on binary classifiers designed to detect the presence or absence of a task-specific stimulus. In this paper, we make two contributions. Our first contribution is in identifying the use of variable length system response delays, a universal and task-independent feature of computing, as a stimulus for driving different levels of frustration. By doing so, we are able to make our second and primary contribution, which is the development of models that predict multiple levels of user frustration from psycho-physiological responses caused by system response delays. We investigate how incorporating different sensor features, application settings, and timing constraints impact the performance of our models. We demonstrate that our models of physiological responses can be used to classify five levels of frustration in near real-time with over 80% accuracy, which is comparable to the accuracy of binary classifiers. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> ACC <s> Stress remains a significant social problem for individuals in modern societies. This paper presents a machine learning approach for the automatic detection of stress of people in a social situation by combining two sensor systems that capture physiological and social responses. We compare the performance using different classifiers including support vector machine, AdaBoost, and k-nearest neighbour. Our experimental results show that by combining the measurements from both sensor systems, we could accurately discriminate between stressful and neutral situations during a controlled Trier social stress test (TSST). Moreover, this paper assesses the discriminative ability of each sensor modality individually and considers their suitability for real time stress detection. Finally, we present an study of the most discriminative features for stress detection. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> ACC <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB004

Time-domain: Statistical features (e.g., mean, median, standard deviation, absolute integral, correlation between axes), first and second derivative of acceleration energy Frequency-domain: Power ratio (0-2.75 Hz and 0-5 Hz band), peak frequency, entropy of the normalised power spectral density References: BIB003 153, BIB001 ECG/ PPG Time-domain: Statistical features (e.g., mean, median, 20th and 80th percentile), heart rate (HR), heart rate variability (HRV), statistical features on HRV (e.g., Root Mean Square of Successive Differences (RMSSD), Standard Deviation of the RR Intervals (SDNN)), number and percentage of successive RR intervals differing by more than 20 ms (NN20, pNN20) or 50 ms (NN50, pNN50), pNN50/pNN20 ratio, TEMP Time-domain: Statistical features (e.g., mean, slope), intersection of the y-axis with a linear regression applied to the signal References: BIB004 BIB002

Wearable-Based Affect Recognition—A Review <s> ECG-and PPG-based Features <s> This paper presents methods for collecting and analyzing physiological data during real-world driving tasks to determine a driver's relative stress level. Electrocardiogram, electromyogram, skin conductance, and respiration were recorded continuously while drivers followed a set route through open roads in the greater Boston area. Data from 24 drives of at least 50-min duration were collected for analysis. The data were analyzed in two ways. Analysis I used features from 5-min intervals of data during the rest, highway, and city driving conditions to distinguish three levels of driver stress with an accuracy of over 97% across multiple drivers and driving days. Analysis II compared continuous features, calculated at 1-s intervals throughout the entire drive, with a metric of observable stressors created by independent coders from videotapes. The results show that for most drivers studied, skin conductivity and heart rate metrics are most closely correlated with driver stress level. These findings indicate that physiological signals can provide a metric of driver stress in future cars capable of physiological monitoring. Such a metric could be used to help manage noncritical in-vehicle information systems and could also provide a continuous measure of how different road and traffic conditions affect drivers. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> ECG-and PPG-based Features <s> This paper reports on a new methodology for the automatic assessment of emotional responses. More specifically, emotions are elicited in agreement with a bidimensional spatial localization of affective states, that is, arousal and valence dimensions. A dedicated experimental protocol was designed and realized where specific affective states are suitably induced while three peripheral physiological signals, i.e., ElectroCardioGram (ECG), ElectroDermal Response (EDR), and ReSPiration activity (RSP), are simultaneously acquired. A group of 35 volunteers was presented with sets of images gathered from the International Affective Picture System (IAPS) having five levels of arousal and five levels of valence, including a neutral reference level in both. Standard methods as well as nonlinear dynamic techniques were used to extract sets of features from the collected signals. The goal of this paper is to implement an automatic multiclass arousal/valence classifier comparing performance when extracted features from nonlinear methods are used as an alternative to standard features. Results show that, when nonlinearly extracted features are used, the percentages of successful recognition dramatically increase. A good recognition accuracy (>;90 percent) after 40-fold cross-validation steps for both arousal and valence classes was achieved by using the Quadratic Discriminant Classifier (QDC). <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> ECG-and PPG-based Features <s> Chronic stress is endemic to modern society. However, as it is unfeasible for physicians to continuously monitor stress levels, its diagnosis is nontrivial. Wireless body sensor networks offer opportunities to ubiquitously detect and monitor mental stress levels, enabling improved diagnosis, and early treatment. This article describes the development of a wearable sensor platform to monitor a number of physiological correlates of mental stress. We discuss tradeoffs in both system design and sensor selection to balance information content and wearability. Using experimental signals collected from the wearable sensor, we describe a selected number of physiological features that show good correlation with mental stress. In particular, we propose a new spectral feature that estimates the balance of the autonomic nervous system by combining information from the power spectral density of respiration and heart rate variability. We validate the effectiveness of our approach on a binary discrimination problem when subjects are placed under two psychophysiological conditions: mental stress and relaxation. When used in a logistic regression model, our feature set is able to discriminate between these two mental states with a success rate of 81% across subjects. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> ECG-and PPG-based Features <s> This paper presents results of analysis performed on a physiologic time-series dataset that was collected from a wearable ECG monitoring system worn by individuals who suffer from panic disorder. Models are constructed and evaluated for distinguishing between pathologic and non-pathologic states, including panic (during panic attack), pre-panic (preceding panic attack) and non-panic (outside panic attack window). The models presented use data fusion to combine both traditional time and frequency domain heart rate variability analysis together with nonlinear/complexity analysis. The best performing model is shown to be a random forest classifier that achieves an accuracy of 97.2% and 90.7% for recognizing states of panic and pre-panic, respectively. The models presented have application in pervasive and ubiquitous mobile and wearable health management systems. <s> BIB004

From ECG and PPG data, various features related to cardiac activity are derived. Below, we provide a description of features commonly used in AR. For an in-depth analysis of features based on the cardiac cycle we refer to Malik . Commonly the heart rate is used as feature. Based on the location of the R-peaks (or the systolic peak in the PPG signal) the inter beat interval (IBI) can be computed. The IBI serves as a new time series signal, from which various HRV features can be derived, both in time-and frequency-domain. For instance, from the IBI the number and percentage of successive RR intervals differing by more than a certain amount of time (e.g., 20 or 50 milliseconds) can be computed. These feature are referred to as NNX and pNNX, where X is the time difference threshold in milliseconds. Based on the Fourier-transformation of the IBI time series, various frequency-domain features can be computed, which reflect the sympathetic and parasympathetic activities of the autonomic nervous system. Four different frequency bands are established in this respect BIB004 . The ultra low frequency (ULF) and very low frequency (VLF) bands range from 0 to 0.003 Hz and from 0.003 to 0.03 Hz, respectively. Changes in low frequency (LF) band, ranged between 0.03 and 0.15 Hz, are mostly associated with the activity of the sympathetic nervous system (SNS). In contrast, the high frequency (HF) band, ranged from 0.15 to 0.4 Hz, is believed to reflect mostly the activity of the parasympathetic nervous system (PNS) BIB004 . Therefore, the LF/HF ratio quantizes is a descriptive feature indicating the influence of both, SNS and PNS, on the cardiac activity. In literature, for example, Healey and Picard BIB001 , it was shown that the LF/HF ratio is a good indicator for stress. In addition to time and frequency domain-based features, non-linear features derived from ECG data were employed successfully wearable-based AR. Rubin et al. BIB004 , for instance, presents a detailed description of non-linear ECG features (e.g., maximal Lyapunov exponent, standard deviations (SD 1 and SD 2 ) along major axes of a Poincaré plot, the SD 1 /SD 2 ratio, sample entropy, etc.). Moreover, Valenza et al. BIB002 , aiming to detect five levels of valence and arousal, compared the performance of linear and non-linear features. Their results indicate that non-linear features are able to improve classification scores significantly. Another class of features based on the cardiac cycle are referred to as geometrical features. An example is the triangular interpolation index BIB002 BIB004 -A histogram of the RR intervals is computed, a triangular interpolation performed, and the baseline of the distribution is computed. Finally, the respiration is known to have an impact on the ECG signal. In literature, there exist different approaches for quantifying the effect of the RESP on the ECG data: Hovsepian et al. , for instance, employed the respiratory sinus arrhythmia (RSA), which is calculated by subtracting the shortest RR interval from the longest RR interval within one respiration cycle. In addition, Choi et al. BIB003 proposed a method of decomposing the HRV into a respiration-and a stress-driven component.

Wearable-Based Affect Recognition—A Review <s> EDA-Based Features <s> Abstract Overlapping phasic skin conductance responses (SCRs) obtained using short interstimulus interval (ISI) paradigms such as those employed in cognitive research, confound measurement of each discrete phasic SCR as well as the tonic skin conductance level (SCL). We report a method of resolving this problem using a modelling technique that takes advantage of the stereotyped nature of the within-subject SCR waveform. A four-parameter sigmoid-exponential SCR model that describes the entire response, was developed and extended to five-, six- and eight-parameter skin conductance (SC) models. These SC models were successfully curve-fitted to more than 60 SC segments, each containing one SCR or two overlapping SCRs on a sloping baseline obtained from 20 normal subjects. The SC segments were consequently decomposed into their components: the tail of the previous response, one or two SCRs and the SCL. The SCRs free of the complication of overlap were then quantified. The raw SCRs of the same data set were also measured using a standard method. The standard measurement showed a significant reduction of 15% in amplitude and 140 ms in peak latency compared to our method. The basic four SCR model parameters—onset time, rise time, decay time constant and gain—showed increasing inter-subject variability in that order. These SCR model parameters may be studied as variables in normal and patient groups and as indices of treatment response. This quantitative method also provides a means to assess the relationships between central and autonomic psychophysiologic measures. © 1997 Elsevier Science B.V. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> EDA-Based Features <s> This paper presents methods for collecting and analyzing physiological data during real-world driving tasks to determine a driver's relative stress level. Electrocardiogram, electromyogram, skin conductance, and respiration were recorded continuously while drivers followed a set route through open roads in the greater Boston area. Data from 24 drives of at least 50-min duration were collected for analysis. The data were analyzed in two ways. Analysis I used features from 5-min intervals of data during the rest, highway, and city driving conditions to distinguish three levels of driver stress with an accuracy of over 97% across multiple drivers and driving days. Analysis II compared continuous features, calculated at 1-s intervals throughout the entire drive, with a metric of observable stressors created by independent coders from videotapes. The results show that for most drivers studied, skin conductivity and heart rate metrics are most closely correlated with driver stress level. These findings indicate that physiological signals can provide a metric of driver stress in future cars capable of physiological monitoring. Such a metric could be used to help manage noncritical in-vehicle information systems and could also provide a continuous measure of how different road and traffic conditions affect drivers. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> EDA-Based Features <s> Little attention has been paid so far to physiological signals for emotion recognition compared to audiovisual emotion channels such as facial expression or speech. This paper investigates the potential of physiological signals as reliable channels for emotion recognition. All essential stages of an automatic recognition system are discussed, from the recording of a physiological dataset to a feature-based multiclass classification. In order to collect a physiological dataset from multiple subjects over many weeks, we used a musical induction method which spontaneously leads subjects to real emotional states, without any deliberate lab setting. Four-channel biosensors were used to measure electromyogram, electrocardiogram, skin conductivity and respiration changes. A wide range of physiological features from various analysis domains, including time/frequency, entropy, geometric analysis, subband spectra, multiscale entropy, etc., is proposed in order to find the best emotion-relevant features and to correlate them with emotional states. The best features extracted are specified in detail and their effectiveness is proven by classification results. Classification of four musical emotions (positive/high arousal, negative/high arousal, negative/low arousal, positive/low arousal) is performed by using an extended linear discriminant analysis (pLDA). Furthermore, by exploiting a dichotomic property of the 2D emotion model, we develop a novel scheme of emotion-specific multilevel dichotomous classification (EMDC) and compare its performance with direct multiclass classification using the pLDA. Improved recognition accuracy of 95\% and 70\% for subject-dependent and subject-independent classification, respectively, is achieved by using the EMDC scheme. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> EDA-Based Features <s> The inferred cost of work-related stress call for prevention strategies that aim at detecting early warning signs at the workplace. This paper goes one step towards the goal of developing a personal health system for detecting stress. We analyze the discriminative power of electrodermal activity (EDA) in distinguishing stress from cognitive load in an office environment. A collective of 33 subjects underwent a laboratory intervention that included mild cognitive load and two stress factors, which are relevant at the workplace: mental stress induced by solving arithmetic problems under time pressure and psychosocial stress induced by social-evaluative threat. During the experiments, a wearable device was used to monitor the EDA as a measure of the individual stress reaction. Analysis of the data showed that the distributions of the EDA peak height and the instantaneous peak rate carry information about the stress level of a person. Six classifiers were investigated regarding their ability to discriminate cognitive load from stress. A maximum accuracy of 82.8% was achieved for discriminating stress from cognitive load. This would allow keeping track of stressful phases during a working day by using a wearable EDA device. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> EDA-Based Features <s> We present a multimodal data set for the analysis of human affective states. The electroencephalogram (EEG) and peripheral physiological signals of 32 participants were recorded as each watched 40 one-minute long excerpts of music videos. Participants rated each video in terms of the levels of arousal, valence, like/dislike, dominance, and familiarity. For 22 of the 32 participants, frontal face video was also recorded. A novel method for stimuli selection is proposed using retrieval by affective tags from the last.fm website, video highlight detection, and an online assessment tool. An extensive analysis of the participants' ratings during the experiment is presented. Correlates between the EEG signal frequencies and the participants' ratings are investigated. Methods and results are presented for single-trial classification of arousal, valence, and like/dislike ratings using the modalities of EEG, peripheral physiological signals, and multimedia content analysis. Finally, decision fusion of the classification results from different modalities is performed. The data set is made publicly available and we encourage other researchers to use it for testing their own affective state estimation methods. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> EDA-Based Features <s> Chronic stress is endemic to modern society. However, as it is unfeasible for physicians to continuously monitor stress levels, its diagnosis is nontrivial. Wireless body sensor networks offer opportunities to ubiquitously detect and monitor mental stress levels, enabling improved diagnosis, and early treatment. This article describes the development of a wearable sensor platform to monitor a number of physiological correlates of mental stress. We discuss tradeoffs in both system design and sensor selection to balance information content and wearability. Using experimental signals collected from the wearable sensor, we describe a selected number of physiological features that show good correlation with mental stress. In particular, we propose a new spectral feature that estimates the balance of the autonomic nervous system by combining information from the power spectral density of respiration and heart rate variability. We validate the effectiveness of our approach on a binary discrimination problem when subjects are placed under two psychophysiological conditions: mental stress and relaxation. When used in a logistic regression model, our feature set is able to discriminate between these two mental states with a success rate of 81% across subjects. <s> BIB006

Considering the EDA signal, basic statistical features (e.g., mean, standard deviation, min, max) are commonly used BIB004 . In addition, Koelstra et al. BIB005 provides a list of statistical (e.g., average rising time and decay rate) and frequency domain-based (spectral power values in the 0-2.4 Hz frequency bands) EDA features. Furthermore, the EDA is known to consist of two components-the skin conductance level (SCL) and skin conductance response (SCR) component. Approaches to separate these components were, for instance, presented by Choi et al. BIB006 or Lim et al. BIB001 . The degree of linearity of the SCL component was shown to be a useful feature BIB006 . Considering the SCR component, the identified SCR segments are counted and further statistical features derived: sum of the SCR startle magnitudes and response durations, in the area under the identified SCRs BIB002 . The SCR-related features were found to be particularly interesting as they are closely linked to high arousal states BIB003 .

Wearable-Based Affect Recognition—A Review <s> EMG-Based Features <s> We present a multimodal data set for the analysis of human affective states. The electroencephalogram (EEG) and peripheral physiological signals of 32 participants were recorded as each watched 40 one-minute long excerpts of music videos. Participants rated each video in terms of the levels of arousal, valence, like/dislike, dominance, and familiarity. For 22 of the 32 participants, frontal face video was also recorded. A novel method for stimuli selection is proposed using retrieval by affective tags from the last.fm website, video highlight detection, and an online assessment tool. An extensive analysis of the participants' ratings during the experiment is presented. Correlates between the EEG signal frequencies and the participants' ratings are investigated. Methods and results are presented for single-trial classification of arousal, valence, and like/dislike ratings using the modalities of EEG, peripheral physiological signals, and multimedia content analysis. Finally, decision fusion of the classification results from different modalities is performed. The data set is made publicly available and we encourage other researchers to use it for testing their own affective state estimation methods. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> EMG-Based Features <s> Timely mental stress detection can help to prevent stress-related health problems. The aim of this study was to identify those physiological signals and features suitable for detecting mental stress in office-like situations. Electrocardiogram (ECG), respiration, skin conductance and surface electromyogram (sEMG) of the upper trapezius muscle were measured with a wearable system during three distinctive stress tests. The protocol contained stress tests that were designed to represent office-like situations. Generalized Estimating Equations were used to classify the data into rest and stress conditions. We reached an average classification rate of 74.5%. This approach may be used for continuous stress measurement in daily office life to detect mental stress at an early stage. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> EMG-Based Features <s> In this work, we present DECAF —a multimodal data set for dec oding user physiological responses to af fective multimedia content. Different from data sets such as DEAP [15] and MAHNOB-HCI [31] , DECAF contains (1) brain signals acquired using the Magnetoencephalogram (MEG) sensor, which requires little physical contact with the user’s scalp and consequently facilitates naturalistic affective response, and (2) explicit and implicit emotional responses of 30 participants to 40 one-minute music video segments used in [15] and 36 movie clips, thereby enabling comparisons between the EEG versus MEG modalities as well as movie versus music stimuli for affect recognition. In addition to MEG data, DECAF comprises synchronously recorded near-infra-red (NIR) facial videos, horizontal Electrooculogram (hEOG), Electrocardiogram (ECG), and trapezius-Electromyogram (tEMG) peripheral physiological responses. To demonstrate DECAF’s utility, we present (i) a detailed analysis of the correlations between participants’ self-assessments and their physiological responses and (ii) single-trial classification results for valence , arousal and dominance , with performance evaluation against existing data sets. DECAF also contains time-continuous emotion annotations for movie clips from seven users, which we use to demonstrate dynamic emotion prediction. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> EMG-Based Features <s> This paper addresses the problem of emotion recognition from physiological signals. Features are extracted and ranked based on their effect on classification accuracy. Different classifiers are compared. The inter-subject variability and the personalization effect are thoroughly investigated, through trial-based and subject-based cross-validation. Finally, a personalized model is introduced, that would allow for enhanced emotional state prediction, based on the physiological data of subjects that exhibit a certain degree of similarity, without the requirement of further feedback. <s> BIB004

From the EMG signal, various time-and frequency-domain features can be extracted. Christy et al. , working on the DEAP dataset, computed statistical features such as mean, median, standard deviation, and interquartile ranges on the EMG data. Other researchers used frequency-based features such as peak or mean frequencies BIB002 BIB004 . Another frequently used feature is the signal energy of either the complete signal BIB001 or specific frequency ranges (e.g., 55-95 Hz, 105-145 Hz) BIB003 . Wijsman et al. BIB002 performed a reference voluntary contraction measurement to compute a personalised EMG gap feature. This feature is defined as the relative time the EMG amplitude is below a specific percentage of the amplitude of the reference measurements.

Wearable-Based Affect Recognition—A Review <s> Respiration-Based Features <s> The existence of specific somatic states associated with different emotions remains controversial. In this study, we investigated the profile of cardiorespiratory activity during the experience of fear, anger, sadness and happiness. ECG and respiratory activity was recorded in 43 healthy volunteers during the recall and experiential reliving of one or two potent emotional autobiographical episodes and a neutral episode. Univariate statistics indicated that the four emotions differed from each other and from the neutral control condition on several linear and spectral indices of cardiorespiratory activity. Dependent variables were further reduced to five physiologically meaningful factors using an exploratory principal component analysis (PCA). Multivariate analyses of variance and effect size estimates calculated on those factors confirmed the differences between the four emotion conditions. A stepwise discriminant analyses predicting emotions using the PCA factors led to a classification rate of 65.3% for the four emotions (chance = 25%; p = 0.001) and of 72.0–83.3% for pair-wise discrimination (chance = 50%; p's < 0.05). These findings may be considered preliminary in view of the small sample on which the multivariate approach has been applied. However, this study emphasizes the need to better characterize the multidimensional factors involved in cardio-respiratory regulation during emotion. These results are consistent with the notion that distinct patterns of peripheral physiological activity are associated with different emotions. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Respiration-Based Features <s> Little attention has been paid so far to physiological signals for emotion recognition compared to audiovisual emotion channels such as facial expression or speech. This paper investigates the potential of physiological signals as reliable channels for emotion recognition. All essential stages of an automatic recognition system are discussed, from the recording of a physiological dataset to a feature-based multiclass classification. In order to collect a physiological dataset from multiple subjects over many weeks, we used a musical induction method which spontaneously leads subjects to real emotional states, without any deliberate lab setting. Four-channel biosensors were used to measure electromyogram, electrocardiogram, skin conductivity and respiration changes. A wide range of physiological features from various analysis domains, including time/frequency, entropy, geometric analysis, subband spectra, multiscale entropy, etc., is proposed in order to find the best emotion-relevant features and to correlate them with emotional states. The best features extracted are specified in detail and their effectiveness is proven by classification results. Classification of four musical emotions (positive/high arousal, negative/high arousal, negative/low arousal, positive/low arousal) is performed by using an extended linear discriminant analysis (pLDA). Furthermore, by exploiting a dichotomic property of the 2D emotion model, we develop a novel scheme of emotion-specific multilevel dichotomous classification (EMDC) and compare its performance with direct multiclass classification using the pLDA. Improved recognition accuracy of 95\% and 70\% for subject-dependent and subject-independent classification, respectively, is achieved by using the EMDC scheme. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Respiration-Based Features <s> Repeated exposures to psychological stress can lead to or worsen diseases of slow accumulation such as heart diseases and cancer. The main challenge in addressing the growing epidemic of stress is a lack of robust methods to measure a person's exposure to stress in the natural environment. Periodic self-reports collect only subjective aspects, often miss stress episodes, and impose significant burden on subjects. Physiological sensors provide objective and continuous measures of stress response, but exhibit wide between-person differences and are easily confounded by daily activities (e.g., speaking, physical movements, coffee intake, etc.). In this paper, we propose, train, and test two models for continuous prediction of stress from physiological measurements captured by unobtrusive, wearable sensors. The first model is a physiological classifier that predicts whether changes in physiology represent stress. Since the effect of stress may persist in the mind longer than its acute effect on physiology, we propose a perceived stress model to predict perception of stress. It uses the output of the physiological classifier to model the accumulation and gradual decay of stress in the mind. To account for wide between-person differences, both models self-calibrate to each subject. Both models were trained using data collected from 21 subjects in a lab study, where they were exposed to cognitive, physical, and social stressors representative of that experienced in the natural environment. Our physiological classifier achieves 90% accuracy and our perceived stress model achieves a median correlation of 0.72 with self-reported rating. We also evaluate the perceived stress model on data collected from 17 participants in a two-day field study, and find that the average rating of stress obtained from our model has a correlation of 0.71 with that obtained from periodic self-reports. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Respiration-Based Features <s> MAHNOB-HCI is a multimodal database recorded in response to affective stimuli with the goal of emotion recognition and implicit tagging research. A multimodal setup was arranged for synchronized recording of face videos, audio signals, eye gaze data, and peripheral/central nervous system physiological signals. Twenty-seven participants from both genders and different cultural backgrounds participated in two experiments. In the first experiment, they watched 20 emotional videos and self-reported their felt emotions using arousal, valence, dominance, and predictability as well as emotional keywords. In the second experiment, short videos and images were shown once without any tag and then with correct or incorrect tags. Agreement or disagreement with the displayed tags was assessed by the participants. The recorded videos and bodily responses were segmented and stored in a database. The database is made available to the academic community via a web-based system. The collected data were analyzed and single modality and modality fusion results for both emotion recognition and implicit tagging experiments are reported. These results show the potential uses of the recorded modalities and the significance of the emotion elicitation protocol. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Respiration-Based Features <s> In order to improve intelligent Human-Computer Interaction it is important to create a personalized adaptive emotion estimator that is able to learn over time emotional response idiosyncrasies of individual person and thus enhance estimation accuracy. This paper, with the aim of identifying preferable methods for such a concept, presents an experiment-based comparative study of seven feature reduction and seven machine learning methods commonly used for emotion estimation based on physiological signals. The analysis was performed on data obtained in an emotion elicitation experiment involving 14 participants. Specific discrete emotions were targeted with stimuli from the International Affective Picture System database. The experiment was necessary to achieve the uniformity in the various aspects of emotion elicitation, data processing, feature calculation, self-reporting procedures and estimation evaluation, in order to avoid inconsistency problems that arise when results from studies that use different emotion-related databases are mutually compared. The results of the performed experiment indicate that the combination of a multilayer perceptron (MLP) with sequential floating forward selection (SFFS) exhibited the highest accuracy in discrete emotion classification based on physiological features calculated from ECG, respiration, skin conductance and skin temperature. Using leave-one-session-out crossvalidation method, 60.3% accuracy in classification of 5 discrete emotions (sadness, disgust, fear, happiness and neutral) was obtained. In order to identify which methods may be the most suitable for real-time estimator adaptation, execution and learning times of emotion estimators were also comparatively analyzed. Based on this analysis, preferred feature reduction method for real-time estimator adaptation was minimum redundancy - maximum relevance (mRMR), which was the fastest approach in terms of combined execution and learning time, as well as the second best in accuracy, after SFFS. In combination with mRMR, highest accuracies were achieved by k-nearest neighbor (kNN) and MLP with negligible difference (50.33% versus 50.54%); however, mRMR+kNN is preferable option for real-time estimator adaptation due to considerably lower combined execution and learning time of kNN versus MLP. <s> BIB005

Soleymani et al. BIB004 pointed out that slow respiration is linked to relaxation. In contrast, irregular and quickly varying breathing patterns correspond to more aroused states like, anger or fear BIB002 BIB001 . Therefore, different respiration patterns can provide valuable information for the detection of affective states. Plarre et al. BIB003 describe a number of time-domain features which aggregate information about breathing cycles: breathing rate, inhalation (I) and exhalation (E) duration, ratio between I/E, stretch (the difference between the peak and the minimum amplitude of a respiration cycle), and the volume of air inhaled/exhaled. Considering frequency-domain features, Kukolja et al. BIB005 used mean power values of four frequency subbands (0-0.1 Hz, 0.1-0.2 Hz, 0.2-0.3 Hz, and 0.3-0.4 Hz) in order to classify different types of emotions. As discussed previously features relation cardiac and respiratory activities (like RSA are frequently employed BIB003 .

Wearable-Based Affect Recognition—A Review <s> Classification <s> This article describes experiments in which happy or sad moods were induced in subjects by hyp- notic suggestion to investigate the influence of emo- tions on memory and thinking. One result was that subjects exhibited mood-state-dependent memory in recall of word lists, personal experiences recorded in a daily diary, and childhood experiences; people recalled a greater percentage of those experiences that were affectively congruent with the mood they were in dur- ing recall. Second, emotion powerfully influenced such cognitive processes as free associations, imaginative fantasies, social perceptions, and snap judgments about others' personalities (e.g., angry subjects generated an- gry associates, told hostile stories, and were prone to find fault with others). Third, when the feeling-tone of a narrative agreed with the reader's emotion, the salience and memorability of events in that narrative were increased. Thus, sad readers attended more to sad material, identified with a sad character from a story, and recalled more about that character. An associative network theory is proposed to account for these several results. In this theory, an emotion serves as a memory unit that can enter into associations with coincident events. Activation of this emotion unit aids retrieval of events associated with it; it also primes emotional themata for use in free association, fantasies, and per- ceptual categorization. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Objective. —This article defines stress and related concepts and reviews their historical development. The notion of a stress system as the effector of the stress syndrome is suggested, and its physiologic and pathophysiologic manifestations are described. A new perspective on human disease states associated with dysregulation of the stress system is provided. Data Sources. —Published original articles from human and animal studies and selected reviews. Literature was surveyed utilizing MEDLINE and the Index Medicus . Study Selection. —Original articles from the basic science and human literature consisted entirely of controlled studies based on verified methodologies and, with the exception of the most recent studies, replicated by more than one laboratory. Many of the basic science and clinical studies had been conducted in our own laboratories and clinical research units. Reviews cited were written by acknowledged leaders in the fields of neurobiology, endocrinology, and behavior. Data Extraction. —Independent extraction and cross-referencing by the authors. Data Synthesis. —Stress and related concepts can be traced as far back as written science and medicine. The stress system coordinates the generalized stress response, which takes place when a stressor of any kind exceeds a threshold. The main components of the stress system are the corticotropin-releasing hormone and locus ceruleus-norepinephrine/autonomic systems and their peripheral effectors, the pituitary-adrenal axis, and the limbs of the autonomic system. Activation of the stress system leads to behavioral and peripheral changes that improve the ability of the organism to adjust homeostasis and increase its chances for survival. There has been an exponential increase in knowledge regarding the interactions among the components of the stress system and between the stress system and other brain elements involved in the regulation of emotion, cognitive function, and behavior, as well as with the axes responsible for reproduction, growth, and immunity. This new knowledge has allowed association of stress system dysfunction, characterized by sustained hyperactivity and/or hypoactivity, to various pathophysiologic states that cut across the traditional boundaries of medical disciplines. These include a range of psychiatric, endocrine, and inflammatory disorders and/or susceptibility to such disorders. Conclusions. —We hope that knowledge from apparently disparate fields of science and medicine integrated into a working theoretical framework will allow generation and testing of new hypotheses on the pathophysiology and diagnosis of, and therapy for, a variety of human illnesses reflecting systematic alterations in the principal effectors of the generalized stress response. We predict that pharmacologic agents capable of altering the central apparatus that governs the stress response will be useful in the treatment of many of these illnesses. ( JAMA . 1992;267:1244-1252) <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Objective: This article presents a new formulation of the relationship between stress and the processes leading to disease. It emphasizes the hidden cost of chronic stress to the body over long time periods, which act as a predisposing factor for the effects of acute, stressful life events. It also presents a model showing how individual differences in the susceptibility to stress are tied to individual behavioral responses to environmental challenges that are coupled to physiologic and pathophysiologic responses. Data Sources: Published original articles from human and animal studies and selected reviews. Literature was surveyed using MEDLINE. Data Extraction: Independent extraction and cross-referencing by us. Data Synthesis: Stress is frequently seen as a significant contributor to disease, and clinical evidence is mounting for specific effects of stress on immune and cardiovascular systems. Yet, until recently, aspects of stress that precipitate disease have been obscure. The concept of homeostasis has failed to help us understand the hidden toll of chronic stress on the body. Rather than maintaining constancy, the physiologic systems within the body fluctuate to meet demands from external forces, a state termed allostasis . In this article, we extend the concept of allostasis over the dimension of time and we define allostatic load as the cost of chronic exposure to fluctuating or heightened neural or neuroendocrine response resulting from repeated or chronic environmental challenge that an individual reacts to as being particularly stressful. Conclusions: This new formulation emphasizes the cascading relationships, beginning early in life, between environmental factors and genetic predispositions that lead to large individual differences in susceptibility to stress and, in some cases, to disease. There are now empirical studies based on this formulation, as well as new insights into mechanisms involving specific changes in neural, neuroendocrine, and immune systems. The practical implications of this formulation for clinical practice and further research are discussed. (Arch Intern Med. 1993;153:2093-2101) <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Abdominal obesity, anxiety, and depression have been found to cluster in several studies. To further characterize these associations, the following study was performed. In a population of 51-year-old men (N = 284), measurements of obesity (body mass index [BMI]) and body fat distribution (waist to hip ratio [WHR] and sagittal trunk recumbent diameter [D]) were analyzed in relation to dexamethasone (0.5 mg) inhibition of cortisol secretion, measured as salivary cortisol. Symptoms of anxiety and depression were defined by a validated questionnaire. Furthermore, testosterone, insulin-like growth factor-I (IGF-I), insulin, glucose, and serum lipid levels were measured. Twenty-five men (8.8%) had symptoms of anxiety and depression. BMI, WHR, and D correlated negatively with testosterone, except for BMI in the anxio-depressive (ADP) group. IGF-I showed no significant relationship. Furthermore, fasting insulin and the insulin to glucose ratio correlated positively and high-density lipoprotein (HDL) cholesterol correlated negatively with BMI, WHR, and D in the total study population and in the subgroups. Total and low-density lipoprotein (LDL) cholesterol showed no significant relationships. Correlation coefficients tended to be higher in ADP men. Dexamethasone inhibition showed a negative significant relationship with BMI (rho = -.47, P = .025), WHR (borderline, rho = -.37, P = .086), and D (rho = -.43, P = .046) only in the ADP group. Comparing the ADP group versus the group without anxio-depression (ADO) and high or low BMI (P = .008), WHR (P = .026), and D (P = .012) showed blunted dexamethasone inhibition only in ADP men with high anthropometric measurements. These findings suggest there is a subgroup with elevated BMI, WHR, and D in whom a blunted dexamethasone response is found associated with traits of anxiety and depression, conditions characterized by such an abnormality. The reason for the association might be insufficient control of cortisol secretion, followed by visceral fat accumulation. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Variable and feature selection have become the focus of much research in areas of application for which datasets with tens or hundreds of thousands of variables are available. These areas include text processing of internet documents, gene expression array analysis, and combinatorial chemistry. The objective of variable selection is three-fold: improving the prediction performance of the predictors, providing faster and more cost-effective predictors, and providing a better understanding of the underlying process that generated the data. The contributions of this special issue cover a wide range of aspects of such problems: providing a better definition of the objective function, feature construction, feature ranking, multivariate feature selection, efficient search methods, and feature validity assessment methods. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Random forests are a combination of tree predictors such that each tree depends on the values of a random vector sampled independently and with the same distribution for all trees in the forest. The generalization error for forests converges a.s. to a limit as the number of trees in the forest becomes large. The generalization error of a forest of tree classifiers depends on the strength of the individual trees in the forest and the correlation between them. Using a random selection of features to split each node yields error rates that compare favorably to Adaboost (Y. Freund & R. Schapire, Machine Learning: Proceedings of the Thirteenth International conference, aaa, 148–156), but are more robust with respect to noise. Internal estimates monitor error, strength, and correlation and these are used to show the response to increasing the number of features used in the splitting. Internal estimates are also used to measure variable importance. These ideas are also applicable to regression. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> The detection of emotion is becoming an increasingly important field for human-computer interaction as the advantages emotion recognition offer become more apparent and realisable. Emotion recognition can be achieved by a number of methods, one of which is through the use of bio-sensors. Bio-sensors possess a number of advantages against other emotion recognition methods as they can be made both inobtrusive and robust against a number of environmental conditions which other forms of emotion recognition have difficulty to overcome. In this paper, we describe a procedure to train computers to recognise emotions using multiple signals from many different bio-sensors. In particular, we describe the procedure we adopted to elicit emotions and to train our system to recognise them. We also present a set of preliminary results which indicate that our neural net classifier is able to obtain accuracy rates of 96.6% and 89.9% for recognition of emotion arousal and valence respectively. <s> BIB007 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Little attention has been paid so far to physiological signals for emotion recognition compared to audiovisual emotion channels such as facial expression or speech. This paper investigates the potential of physiological signals as reliable channels for emotion recognition. All essential stages of an automatic recognition system are discussed, from the recording of a physiological dataset to a feature-based multiclass classification. In order to collect a physiological dataset from multiple subjects over many weeks, we used a musical induction method which spontaneously leads subjects to real emotional states, without any deliberate lab setting. Four-channel biosensors were used to measure electromyogram, electrocardiogram, skin conductivity and respiration changes. A wide range of physiological features from various analysis domains, including time/frequency, entropy, geometric analysis, subband spectra, multiscale entropy, etc., is proposed in order to find the best emotion-relevant features and to correlate them with emotional states. The best features extracted are specified in detail and their effectiveness is proven by classification results. Classification of four musical emotions (positive/high arousal, negative/high arousal, negative/low arousal, positive/low arousal) is performed by using an extended linear discriminant analysis (pLDA). Furthermore, by exploiting a dichotomic property of the 2D emotion model, we develop a novel scheme of emotion-specific multilevel dichotomous classification (EMDC) and compare its performance with direct multiclass classification using the pLDA. Improved recognition accuracy of 95\% and 70\% for subject-dependent and subject-independent classification, respectively, is achieved by using the EMDC scheme. <s> BIB008 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> We conducted a 19 participant study using a system comprised of wireless galvanic skin response (GSR), heart rate (HR), activity sensors and a mobile phone for aggregating sensor data and enabling affect logging by the user. Each participant wore the sensors daily for five days, generating approximately 900 hours of continuous data. We found that analysis of emotional events was highly dependent on correct windowing and report results on synthesized windows around annotated events. Where raters agreed on the timing and quality of the emotion we were able to recognize 85% of the high and low energy emotions and 70% of the positive and negative emotions. We also gained many insights regarding participant's perception of their emotional state and the complexity of emotion in real life. <s> BIB009 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Repeated exposures to psychological stress can lead to or worsen diseases of slow accumulation such as heart diseases and cancer. The main challenge in addressing the growing epidemic of stress is a lack of robust methods to measure a person's exposure to stress in the natural environment. Periodic self-reports collect only subjective aspects, often miss stress episodes, and impose significant burden on subjects. Physiological sensors provide objective and continuous measures of stress response, but exhibit wide between-person differences and are easily confounded by daily activities (e.g., speaking, physical movements, coffee intake, etc.). In this paper, we propose, train, and test two models for continuous prediction of stress from physiological measurements captured by unobtrusive, wearable sensors. The first model is a physiological classifier that predicts whether changes in physiology represent stress. Since the effect of stress may persist in the mind longer than its acute effect on physiology, we propose a perceived stress model to predict perception of stress. It uses the output of the physiological classifier to model the accumulation and gradual decay of stress in the mind. To account for wide between-person differences, both models self-calibrate to each subject. Both models were trained using data collected from 21 subjects in a lab study, where they were exposed to cognitive, physical, and social stressors representative of that experienced in the natural environment. Our physiological classifier achieves 90% accuracy and our perceived stress model achieves a median correlation of 0.72 with self-reported rating. We also evaluate the perceived stress model on data collected from 17 participants in a two-day field study, and find that the average rating of stress obtained from our model has a correlation of 0.71 with that obtained from periodic self-reports. <s> BIB010 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Emotion modeling and recognition has drawn extensive attention from disciplines such as psychology, cognitive science, and, lately, engineering. Although a significant amount of research has been done on behavioral modalities, less explored characteristics include the physiological signals. This work brings to the table the ECG signal and presents a thorough analysis of its psychological properties. The fact that this signal has been established as a biometric characteristic calls for subject-dependent emotion recognizers that capture the instantaneous variability of the signal from its homeostatic baseline. A solution based on the empirical mode decomposition is proposed for the detection of dynamically evolving emotion patterns on ECG. Classification features are based on the instantaneous frequency (Hilbert-Huang transform) and the local oscillation within every mode. Two experimental setups are presented for the elicitation of active arousal and passive arousal/valence. The results support the expectations for subject specificity, as well as demonstrating the feasibility of determining valence out of the ECG morphology (up to 89 percent for 44 subjects). In addition, this work differentiates for the first time between active and passive arousal, and advocates that there are higher chances of ECG reactivity to emotion when the induction method is active for the subject. <s> BIB011 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> This paper reports on a new methodology for the automatic assessment of emotional responses. More specifically, emotions are elicited in agreement with a bidimensional spatial localization of affective states, that is, arousal and valence dimensions. A dedicated experimental protocol was designed and realized where specific affective states are suitably induced while three peripheral physiological signals, i.e., ElectroCardioGram (ECG), ElectroDermal Response (EDR), and ReSPiration activity (RSP), are simultaneously acquired. A group of 35 volunteers was presented with sets of images gathered from the International Affective Picture System (IAPS) having five levels of arousal and five levels of valence, including a neutral reference level in both. Standard methods as well as nonlinear dynamic techniques were used to extract sets of features from the collected signals. The goal of this paper is to implement an automatic multiclass arousal/valence classifier comparing performance when extracted features from nonlinear methods are used as an alternative to standard features. Results show that, when nonlinearly extracted features are used, the percentages of successful recognition dramatically increase. A good recognition accuracy (>;90 percent) after 40-fold cross-validation steps for both arousal and valence classes was achieved by using the Quadratic Discriminant Classifier (QDC). <s> BIB012 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> More than 15 years after the early studies in Affective Computing (AC), [1] the problem of detecting and modeling emotions in the context of human-computer interaction (HCI) remains complex and largely unexplored. The detection and modeling of emotion is, primarily, the study and use of artificial intelligence (AI) techniques for the construction of computational models of emotion. The key challenges one faces when attempting to model emotion [2] are inherent in the vague definitions and fuzzy boundaries of emotion, and in the modeling methodology followed. In this context, open research questions are still present in all key components of the modeling process. These include, first, the appropriateness of the modeling tool employed to map emotional manifestations and responses to annotated affective states; second, the processing of signals that express these manifestations (i.e., model input); and third, the way affective annotation (i.e., model output) is handled. This paper touches upon all three key components of an affective model (i.e., input, model, output) and introduces the use of deep learning (DL) [3], [4], [5] methodologies for affective modeling from multiple physiological signals. <s> BIB013 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> We evaluate 179 classifiers arising from 17 families (discriminant analysis, Bayesian, neural networks, support vector machines, decision trees, rule-based classifiers, boosting, bagging, stacking, random forests and other ensembles, generalized linear models, nearest-neighbors, partial least squares and principal component regression, logistic and multinomial regression, multiple adaptive regression splines and other methods), implemented in Weka, R (with and without the caret package), C and Matlab, including all the relevant classifiers available today. We use 121 data sets, which represent the whole UCI data base (excluding the large-scale problems) and other own real problems, in order to achieve significant conclusions about the classifier behavior, not dependent on the data set collection. The classifiers most likely to be the bests are the random forest (RF) versions, the best of which (implemented in R and accessed via caret) achieves 94.1% of the maximum accuracy overcoming 90% in the 84.3% of the data sets. However, the difference is not statistically significant with the second best, the SVM with Gaussian kernel implemented in C using LibSVM, which achieves 92.3% of the maximum accuracy. A few models are clearly better than the remaining ones: random forest, SVM with Gaussian and polynomial kernels, extreme learning machine with Gaussian kernel, C5.0 and avNNet (a committee of multi-layer perceptrons implemented in R with the caret package). The random forest is clearly the best family of classifiers (3 out of 5 bests classifiers are RF), followed by SVM (4 classifiers in the top-10), neural networks and boosting ensembles (5 and 3 members in the top-20, respectively). <s> BIB014 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Sentilo is a model and a tool to detect holders and topics of opinion sentences. Sentilo implements an approach based on the neo-Davidsonian assumption that events and situations are the primary entities for contextualizing opinions, which makes it able to distinguish holders, main topics, and sub-topics of an opinion. It uses a heuristic graph mining approach that relies on FRED, a machine reader for the Semantic Web that leverages Natural Language Processing (NLP) and Knowledge Representation (KR) components jointly with cognitively-inspired frames. The evaluation results are excellent for holder detection (F1: 95%), very good for subtopic detection (F1: 78%), and good for topic detection (F1: 68%). <s> BIB015 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Automatic emotion recognition is one of the most challenging tasks. To detect emotion from nonstationary EEG signals, a sophisticated learning algorithm that can represent high-level abstraction is required. This study proposes the utilization of a deep learning network (DLN) to discover unknown feature correlation between input signals that is crucial for the learning task. The DLN is implemented with a stacked autoencoder (SAE) using hierarchical feature learning approach. Input features of the network are power spectral densities of 32-channel EEG signals from 32 subjects. To alleviate overfitting problem, principal component analysis (PCA) is applied to extract the most important components of initial input features. Furthermore, covariate shift adaptation of the principal components is implemented to minimize the nonstationary effect of EEG signals. Experimental results show that the DLN is capable of classifying three different levels of valence and arousal with accuracy of 49.52% and 46.03%, respectively. Principal component based covariate shift adaptation enhances the respective classification accuracy by 5.55% and 6.53%. Moreover, DLN provides better performance compared to SVM and naive Bayes classifiers. <s> BIB016 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> In this work, we present DECAF —a multimodal data set for dec oding user physiological responses to af fective multimedia content. Different from data sets such as DEAP [15] and MAHNOB-HCI [31] , DECAF contains (1) brain signals acquired using the Magnetoencephalogram (MEG) sensor, which requires little physical contact with the user’s scalp and consequently facilitates naturalistic affective response, and (2) explicit and implicit emotional responses of 30 participants to 40 one-minute music video segments used in [15] and 36 movie clips, thereby enabling comparisons between the EEG versus MEG modalities as well as movie versus music stimuli for affect recognition. In addition to MEG data, DECAF comprises synchronously recorded near-infra-red (NIR) facial videos, horizontal Electrooculogram (hEOG), Electrocardiogram (ECG), and trapezius-Electromyogram (tEMG) peripheral physiological responses. To demonstrate DECAF’s utility, we present (i) a detailed analysis of the correlations between participants’ self-assessments and their physiological responses and (ii) single-trial classification results for valence , arousal and dominance , with performance evaluation against existing data sets. DECAF also contains time-continuous emotion annotations for movie clips from seven users, which we use to demonstrate dynamic emotion prediction. <s> BIB017 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> This paper presents results of analysis performed on a physiologic time-series dataset that was collected from a wearable ECG monitoring system worn by individuals who suffer from panic disorder. Models are constructed and evaluated for distinguishing between pathologic and non-pathologic states, including panic (during panic attack), pre-panic (preceding panic attack) and non-panic (outside panic attack window). The models presented use data fusion to combine both traditional time and frequency domain heart rate variability analysis together with nonlinear/complexity analysis. The best performing model is shown to be a random forest classifier that achieves an accuracy of 97.2% and 90.7% for recognizing states of panic and pre-panic, respectively. The models presented have application in pervasive and ubiquitous mobile and wearable health management systems. <s> BIB018 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Stress remains a significant social problem for individuals in modern societies. This paper presents a machine learning approach for the automatic detection of stress of people in a social situation by combining two sensor systems that capture physiological and social responses. We compare the performance using different classifiers including support vector machine, AdaBoost, and k-nearest neighbour. Our experimental results show that by combining the measurements from both sensor systems, we could accurately discriminate between stressful and neutral situations during a controlled Trier social stress test (TSST). Moreover, this paper assesses the discriminative ability of each sensor modality individually and considers their suitability for real time stress detection. Finally, we present an study of the most discriminative features for stress detection. <s> BIB019 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Deep learning (DL) methods receive increasing attention within the field of human activity recognition (HAR) due to their success in other machine learning domains. Nonetheless, a direct transfer of these methods is often not possible due to domain specific challenges (e.g. handling of multi-modal sensor data, lack of large labeled datasets). In this paper, we address three key aspects for the future development of robust DL methods for HAR: (1) Is it beneficial to apply data specific normalization? (2) How to optimally fuse multimodal sensor data? (3) How robust are these approaches with respect to available training data? We evaluate convolutional neuronal networks (CNNs) on a new large real-world multimodal dataset (RBK) as well as the PAMAP2 dataset. Our results indicate that sensor specific normalization techniques are required. We present a novel pressure specific normalization method which increases the F1-score by ∼ 4.5 percentage points (pp) on the RBK dataset. Further, we show that late- and hybrid fusion techniques are superior compared to early fusion techniques, increasing the F1-score by up to 3.5 pp (RBK dataset). Finally, our results reveal that in particular CNNs based on a shared filter approach have a smaller dependency on the amount of available training data compared to other fusion techniques. <s> BIB020 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB021 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> While accurately predicting mood and wellbeing could have a number of important clinical benefits, traditional machine learning (ML) methods frequently yield low performance in this domain. We posit that this is because a one-size-fits-all machine learning model is inherently ill-suited to predicting outcomes like mood and stress, which vary greatly due to individual differences. Therefore, we employ Multitask Learning (MTL) techniques to train personalized ML models which are customized to the needs of each individual, but still leverage data from across the population. Three formulations of MTL are compared: i) MTL deep neural networks, which share several hidden layers but have final layers unique to each task; ii) Multi-task Multi-Kernel learning, which feeds information across tasks through kernel weights on feature types; and iii) a Hierarchical Bayesian model in which tasks share a common Dirichlet Process prior. We offer the code for this work in open source. These techniques are investigated in the context of predicting future mood, stress, and health using data collected from surveys, wearable sensors, smartphone logs, and the weather. Empirical results demonstrate that using MTL to account for individual differences provides large performance improvements over traditional machine learning methods and provides personalized, actionable insights. <s> BIB022 </s> Wearable-Based Affect Recognition—A Review <s> Classification <s> The detection and monitoring of emotions are important in various applications, e.g. to enable naturalistic and personalised human-robot interaction. Emotion detection often require modelling of various data inputs from multiple modalities, including physiological signals (e.g.EEG and GSR), environmental data (e.g. audio and weather), videos (e.g. for capturing facial expressions and gestures) and more recently motion and location data. Many traditional machine learning algorithms have been utilised to capture the diversity of multimodal data at the sensors and features levels for human emotion classification. While the feature engineering processes often embedded in these algorithms are beneficial for emotion modelling, they inherit some critical limitations which may hinder the development of reliable and accurate models. In this work, we adopt a deep learning approach for emotion classification through an iterative process by adding and removing large number of sensor signals from different modalities. Our dataset was collected in a real-world study from smart-phones and wearable devices. It merges local interaction of three sensor modalities: on-body, environmental and location into global model that represents signal dynamics along with the temporal relationships of each modality. Our approach employs a series of learning algorithms including a hybrid approach using Convolutional Neural Network and Long Short-term Memory Recurrent Neural Network (CNN-LSTM) on the raw sensor data, eliminating the needs for manual feature extraction and engineering. The results show that the adoption of deep-learning approaches is effective in human emotion classification when large number of sensors input is utilised (average accuracy 95% and F-Measure=%95) and the hybrid models outperform traditional fully connected deep neural network (average accuracy 73% and F-Measure=73%). Furthermore, the hybrid models outperform previously developed Ensemble algorithms that utilise feature engineering to train the model average accuracy 83% and F-Measure=82%) <s> BIB023

In AR the classification is either done using statistical approaches (e.g., ANOVA) or machine learning (ML) methods (e.g., support vector machine (SVM), k-nearest neighbour (kNN)). For both types of analyses, features similar to the ones described in Section 5.2 are combined into a feature vector, associated with a label and used as inputs. Since statistical analysis plays only a minor role in wearable-based AR literature, we focus in this section on classification approaches utilising ML techniques. In Table 9 the same studies are presented as in Table 4 . However, here we focus on the employed classification algorithms, number of target affective classes, setting of the study, number of participants, evaluation schemes, and achieved classification performance. The performance is, if possible, reported as accuracy, indicating the overall percentage of correctly classified instances. The rest of this section discusses and compares the different approaches and their performance. The algorithm column in Table 9 indicates that the SVM is the most common classification algorithm. It is employed in 48% of the considered studies. This is to some degree surprising as the SVM requires careful adjustment of the kernel size γ and the trade-off parameter C. For this adjustment the recorded data has to be split into training, validation, and test sets. The best set of hyperparameters can be found by performing a grid-search BIB019 , evaluating the current hyperparameter on the validation set. The performance of the final model is then evaluated on the test set. Hence, when using a SVM, it is important to report the final test error (and not the validation error). kNN and decision-tree (DT), are the second most popular classifiers both applied in 22% of the considered studies. kNN and DT require only little hyperparameter tuning and, hence, are applied (almost) in an off-the-shelf way. Concluding from Table 9 , ensemble methods (e.g., random-forest or AdaBoost) are employed less frequently. This is astonishing as ensemble methods have been proven to be strong classifiers. Fernández-Delgado et al. BIB014 evaluated 179 classifiers on more than hundred different datasets and found that the random-forest family 'is clearly the best family of classifiers'. In the wearable-based AR community, Rubin et al. BIB018 employed random-forests to detect panic and pre-panic states, reaching a 97% and 91% accuracy, respectively. In addition, boosting was found to be a strong classifier BIB014 , and Leo Breiman even considered it to be the 'best off-the-shelf classifier in the world' . Mozos et al. BIB019 applied the AdaBoost method to detect stress, reaching an accuracy of 94%. For a detailed description of random-forests we refer the reader to Breiman BIB006 and an introduction into boosting can be found in Freund et al. . Fernández-Delgado et al. BIB014 also found neural networks (NN) to be among the top-20 classifiers. Haag et al. BIB007 and Jaques et al. used NN, in the form of multi-layered perceptrons, to detect different affective states. Convolutional neural network (CNN) and long short-term memory-based classification techniques, which are becoming popular in the field of human activity recognition BIB020 , have not found broad application in the domain of wearable-based AR domain yet. Martinez et al. BIB013 compare the performance of learned and hand-crafted features to detect the affective states relaxation, anxiety, excitement, and fun. The learned features were extracted using a set of convolutional layers, and the final classification step was performed using a single-layer perceptron. The experiments of Martinez et al. BIB013 indicate that learned features lead to an improved classification performance (compared to the hand crafted features). 1 Given as pairwise preferences. BIB001 DT overfit, other classifiers performed worse than random guessing. BIB003 Correlation between self-reported and output of model. BIB002 No significant differences could be found between the affective states. BIB004 Mean absolute error of mood angle in circumplex model. 80/20% split of the entire data+5-fold CV. User specific models. Trained random on 70/30% splits with non-overlapping windows. Judging from Table 9 , a binary classification tasks were pursued in most (52%) presented studies. This holds even for the cases where the study protocol aimed at eliciting different emotions. A frequent task, following for instance Agrafioti et al. BIB011 or Abadi et al. BIB017 , is to distinguish between high/low valence/arousal using physiological data. Considering the setting, three different types of studies are distinguished: lab (L), field (F), and field with constraints (FC) studies. Studies conducted in a vehicle on public roads are referred to as FC studies, as subjects are constrained in their movement. In addition, studies were subjects followed a specific (outdoor) path, for example, Kanjo et al. BIB023 are referred to as FC studies. Most, 31 out of 46, studies presented in Table 9 , solely base their results on data recorded in a lab setting. The popularity of lab studies is easily explained: in lab studies the study protocol is designed to elicit a set of specific target affective states (see Section 4.1). Hence, the signal to noise ratio is much higher than in field studies. Furthermore, once the set of stimuli is chosen the same protocol is applied to multiple subjects, which makes lab studies very efficient. However, models trained on data gathered in constrained environments, are likely to exhibit a poor performance in an less constrained setting. In order to overcome this, field studies have become more frequent over the past years. This 'out of the lab and into the fray' BIB009 is also related to recent advances in mobile sensor technology and the broad acceptance of smart devices (watches, phones, et.) among users. As wearable-based AR clearly aims to detect the affective state users in unconstrained environments, this trend is certainly desirable. Recent work aspiring to detect stress in lab and real life scenarios has for instance been conducted by Gjoreski et al. BIB021 , Plarre et al. BIB010 , Hovsepian et al. , Taylor et al. BIB022 . Their results indicate that stress detection, based on wearable-based data and context information, is feasible, even in mostly unconstrained settings. Finally, considering the number of study participants there is a large variation: The results reported in Table 9 are based on data originating from a single subject up to 104 subjects. Clearly, a large and diversified subject pool is desirable. This would allow to develop generalized models for wearable-based AR. Judging from Table 9 , n-fold cross-validation (CV) (n ∈ BIB003 BIB004 BIB015 BIB016 ) is frequently employed as validation method (28%). Following this method, the dataset is randomly partitioned into n equally sized subsets. Then, n − 1 subsets are used for training and the remaining one for testing. This procedure is repeated n times. Hence, each of the n subsets is used exactly once as test set. In case the trained model requires hyperparameter tuning, part of the training data can serve as validation set in each iteration. If features are extracted on overlapping windows and n-fold CV is used as validation methods the results are often overoptimistic. This is due to the strong correlation between the features extracted from overlapping windows. Leave-One-Out (LOO) CV is also used in several studies listed in Table 9 . This is a specific version of the n-fold CV procedure, where n is equals to the total number of available feature vectors. In the LOO case each feature vector is used once for testing. A slightly different type of validation was performed by Abadi et al. BIB017 : Leave-One-Trial-Out (LOTO) CV. During LOTO CV, the model is trained on the data of all subjects but leaving one trial/stimulus (e.g., video) aside. The trained algorithm is then evaluated on the left-out data, and the procedure is repeated for each trial. LOO, LOTO, and n-fold CV lead to subject-dependent results. In order to obtain an subject independent score, corresponding to a more realistic results for real-life deployment, leave-one-subject-out (LOSO) CV should be applied. For this purpose, the algorithm under consideration is trained on the data of all but one subject. The data of the left-out subject is then used to evaluate the trained model. Repeating this procedure for all subjects in the dataset gives a realistic estimate of the model's generalisation properties on completely unseen data. As indicated by Table 9 , nowadays LOSO CV is widely accepted and applied. From the results shown here, it can be concluded that using the LOSO validation method leads to lower classification scores than applying n-fold or LOO CV. However, only LOSO provides the information on how good the trained model is able to perform on completely unseen data (e.g., data of a new user). Hence, we recommend using this validation scheme. The affect and stress recognition approaches presented in Table 9 report accuracies between 40% and 95%. Due to the lack of benchmarking datasets, the results obtained in different studies are hard to compare. On average, the classification accuracies obtained using lab data are higher than the ones obtained in field study data. Hovsepian et al. , who conducted both a lab and a field study, report on a 92 % mean accuracy in detecting stress based on lab data. However, when field data is considered, the accuracy drops to 62 %. Moreover, Healey et al. BIB009 conducted a field study and trained different classifiers on the collected data but none of them was able to perform better than random guessing. This indicates that wearable-based AR in the field is very challenging. As indicated in Table 4 , most studies were conducted recording multimodal datasets. This might be motivated by a recent review of D'mello and Kory , who pointed out that the classifiers relying on multimodal input reach on average higher classification scores than their unimodal counterparts. Considering the accuracy of classifiers detecting high/low arousal and high/low valence separately it becomes apparent, that arousal is classified more reliably BIB017 BIB007 BIB012 BIB011 . High arousal states are, from a physiological point of view, directed by the sympathetic nervous system (SNS) (see Section 3). Physiological changes directed by the SNS are quite distinct (e.g., increased heart rate, sweat production, etc.). Hence, detecting high arousal states using these physiological indicators is a feasible task. In contrast, detecting changes in a subject's valence based on physiological data is a more challenging. The performance of standard ML classifiers depend strongly on the employed features. Hence, the benefits of a careful feature selection can be threefold: 1. Feature selection can help to improve classification results. 2. Feature selection identifies cost-effective and yet strong predictors. 3. It provides a better understanding of the processes generating the data BIB005 . According to Guyon and Elisseeff BIB005 , feature selection methods are grouped into filter-based methods, wrappers, and embedded methods. Filter-based methods select a subset of features (e.g., based on statistical a criterion) and do not take the used classifier into account. Wrapper-based methods (e.g., sequential feature selection) treat the learning algorithm as black box and assess the quality of a subset of features based on the final classification score BIB005 . Finally, embedded methods perform variable selection during training. Hence, the selection is commonly specific to the used classifier BIB005 . Feature selection methods also find application in AR. Kim and André BIB008 , for instance, perform feature selection to improve the classification. Valenza et al. BIB012 used Principal Component Analysis to project the features onto a lower dimensional space. This linear method has the advantage that the features are condensed with only a minimal loss of information. For a detailed review of feature selection methods see Guyon and Elisseeff BIB005 .

Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Abstract —Work stress has repeatedly been associated with an increased risk for cardiovascular disease. This study tested whether this relationship could be explained by exaggerated cardiovascular reactivity to work or impaired recovery in leisure time. Vagal tone was assessed as a possible determinant of these work stress effects. Participants included 109 male white-collar workers (age, 47.2±5.3) who were monitored on 2 workdays and 1 nonworkday for ambulatory blood pressure, heart rate, and heart rate variability. Chronic work stress was defined according to Siegrist’s model as (1) high imbalance, a combination of high effort and low reward at work, or (2) high overcommitment, an exhaustive work-related coping style indexing the inability to unwind. All findings were adjusted for possible differences in posture and physical activity between the work stress groups. High imbalance was associated with a higher heart rate during work and directly after work, a higher systolic blood pressure during work and leisure time, and a lower 24-hour vagal tone on all 3 measurement days. Overcommitment was not associated with an unfavorable ambulatory profile. Logistic regression analysis revealed that heart rate [odds ratio 1-SD increase 1.95 (95% CI, 1.02 to 3.77)] and vagal tone [odds ratio 1-SD decrease 2.67 (95% CI, 1.24 to 5.75)] were independently associated with incident mild hypertension. Surprisingly, the values during sleep were more predictive for mild hypertension than the values during work. The results from the present study suggest that the detrimental effects of work stress are partly mediated by increased heart rate reactivity to a stressful workday, an increase in systolic blood pressure level, and lower vagal tone. <s> BIB001 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> It has been widely accepted by many studies that non-linearity exists in the financial markets and that neural networks can be effectively used to uncover this relationship. Unfortunately, many of these studies fail to consider alternative forecasting techniques, the relevance of input variables, or the performance of the models when using different trading strategies. This paper introduces an information gain technique used in machine learning for data mining to evaluate the predictive relationships of numerous financial and economic variables. Neural network models for level estimation and classification are then examined for their ability to provide an effective forecast of future values. A cross-validation technique is also employed to improve the generalization ability of several models. The results show that the trading strategies guided by the classification models generate higher risk-adjusted profits than the buy-and-hold strategy, as well as those guided by the level-estimation based forecasts of the neural network and linear regression models. <s> BIB002 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> This paper presents a method of extracting primary heart sound signals from chest-worn accelerometer data in the presence of motion artifacts. The proposed method outperforms noise removal techniques such as wavelet denoising and adaptive filtering. Results from six subjects show a primary heart signal detection rate of 99.36% with a false positive rate of 1.3%. <s> BIB003 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> We explore an original strategy for building deep networks, based on stacking layers of denoising autoencoders which are trained locally to denoise corrupted versions of their inputs. The resulting algorithm is a straightforward variation on the stacking of ordinary autoencoders. It is however shown on a benchmark of classification problems to yield significantly lower classification error, thus bridging the performance gap with deep belief networks (DBN), and in several cases surpassing it. Higher level representations learnt in this purely unsupervised fashion also help boost the performance of subsequent SVM classifiers. Qualitative experiments show that, contrary to ordinary autoencoders, denoising autoencoders are able to learn Gabor-like edge detectors from natural image patches and larger stroke detectors from digit images. This work clearly establishes the value of using a denoising criterion as a tractable unsupervised objective to guide the learning of useful higher level representations. <s> BIB004 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Pulse transit time (PTT) and pulse wave velocity (PWV), respectively, were shown to have a correlation with systolic blood pressure (SBP) and have been reported to be suitable for indirect BP measurements. The aim of this study was to create a function between SBP and PWV, and to test its reliability for the determination of absolute SBP using a non-linear algorithm and a one-point calibration. 63 volunteers performed exercise to induce rises in BP. Arterial PTT was measured between the R-spike of the ECG and the plethysmographic curve of finger pulse-oximetry. The reference BP was measured using a cuff-based sphygmomanometric aneroid device. Data from 13 of the 63 volunteers served for the detection of the PWV-BP relationship. The created non-linear function was used to calculate BP values after individual correction for the BP offset in a group of 50 volunteers. Individual correlation coefficients for SBP measured by PTT (SBP(PTT)) and by cuff (SBP(CUFF)) varied between r = 0.69 and r = 0.99. Taking all data together, we found r = 0.83 (276 measurements in 50 volunteers). In the Bland-Altman plot, the limits of agreement were [Formula: see text]± 19.8 mmHg. In conclusion, comparing SBP values using the PTT-based method and those measured by cuff resulted in a significant correlation. However, the Bland-Altman plot shows relevant differences between both methods, which are partly due to greater variability of the SBP(PTT) measurement during intensified exercise. Results suggest that PTT can be used for measuring absolute SBP when performing an individual correction for the offset of the BP-PWV relation. <s> BIB005 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> We present a multimodal data set for the analysis of human affective states. The electroencephalogram (EEG) and peripheral physiological signals of 32 participants were recorded as each watched 40 one-minute long excerpts of music videos. Participants rated each video in terms of the levels of arousal, valence, like/dislike, dominance, and familiarity. For 22 of the 32 participants, frontal face video was also recorded. A novel method for stimuli selection is proposed using retrieval by affective tags from the last.fm website, video highlight detection, and an online assessment tool. An extensive analysis of the participants' ratings during the experiment is presented. Correlates between the EEG signal frequencies and the participants' ratings are investigated. Methods and results are presented for single-trial classification of arousal, valence, and like/dislike ratings using the modalities of EEG, peripheral physiological signals, and multimedia content analysis. Finally, decision fusion of the classification results from different modalities is performed. The data set is made publicly available and we encourage other researchers to use it for testing their own affective state estimation methods. <s> BIB006 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> By 2025, when most of today’s psychology undergraduates will be in their mid-30s, more than 5 billion people on our planet will be using ultra-broadband, sensor-rich smartphones far beyond the abilities of today’s iPhones, Androids, and Blackberries. Although smartphones were not designed for psychological research, they can collect vast amounts of ecologically valid data, easily and quickly, from large global samples. If participants download the right “psych apps,” smartphones can record where they are, what they are doing, and what they can see and hear and can run interactive surveys, tests, and experiments through touch screens and wireless connections to nearby screens, headsets, biosensors, and other peripherals. This article reviews previous behavioral research using mobile electronic devices, outlines what smartphones can do now and will be able to do in the near future, explains how a smartphone study could work practically given current technology (e.g., in studying ovulatory cycle effects on w... <s> BIB007 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Automatically recognizing human emotions from spontaneous and non-prototypical real-life data is currently one of the most challenging tasks in the field of affective computing. This article presents our recent advances in assessing dimensional representations of emotion, such as arousal, expectation, power, and valence, in an audiovisual human-computer interaction scenario. Building on previous studies which demonstrate that long-range context modeling tends to increase accuracies of emotion recognition, we propose a fully automatic audiovisual recognition approach based on Long Short-Term Memory (LSTM) modeling of word-level audio and video features. LSTM networks are able to incorporate knowledge about how emotions typically evolve over time so that the inferred emotion estimates are produced under consideration of an optimal amount of context. Extensive evaluations on the Audiovisual Sub-Challenge of the 2011 Audio/Visual Emotion Challenge show how acoustic, linguistic, and visual features contribute to the recognition of different affective dimensions as annotated in the SEMAINE database. We apply the same acoustic features as used in the challenge baseline system whereas visual features are computed via a novel facial movement feature extractor. Comparing our results with the recognition scores of all Audiovisual Sub-Challenge participants, we find that the proposed LSTM-based technique leads to the best average recognition performance that has been reported for this task so far. <s> BIB008 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Mobile telemental health is defined as the use of mobile phones and other wireless devices as applied to psychiatric and mental health practice. Applications of such include treatment monitoring and adherence, health promotion, ecological momentary assessment, and decision support systems. Advantages of mobile telemental health are underscored by its interactivity, just-in-time interventions, and low resource requirements and portability. Challenges in realizing this potential of mobile telemental health include the low penetration rates of health applications on mobile devices in part due to health literacy, the delay in current published research in evaluating newer technologies, and outdated research methodologies. Despite such challenges, one immediate opportunity for mobile telemental health is utilizing mobile devices as videoconferencing mediums for psychotherapy and psychosocial interventions enhanced by novel sensor based monitoring and behavior-prediction algorithms. This paper provides an overview of mobile telemental health and its current trends, as well as future opportunities as applied to patient care in both academic research and commercial ventures. <s> BIB009 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> In this work, we present DECAF —a multimodal data set for dec oding user physiological responses to af fective multimedia content. Different from data sets such as DEAP [15] and MAHNOB-HCI [31] , DECAF contains (1) brain signals acquired using the Magnetoencephalogram (MEG) sensor, which requires little physical contact with the user’s scalp and consequently facilitates naturalistic affective response, and (2) explicit and implicit emotional responses of 30 participants to 40 one-minute music video segments used in [15] and 36 movie clips, thereby enabling comparisons between the EEG versus MEG modalities as well as movie versus music stimuli for affect recognition. In addition to MEG data, DECAF comprises synchronously recorded near-infra-red (NIR) facial videos, horizontal Electrooculogram (hEOG), Electrocardiogram (ECG), and trapezius-Electromyogram (tEMG) peripheral physiological responses. To demonstrate DECAF’s utility, we present (i) a detailed analysis of the correlations between participants’ self-assessments and their physiological responses and (ii) single-trial classification results for valence , arousal and dominance , with performance evaluation against existing data sets. DECAF also contains time-continuous emotion annotations for movie clips from seven users, which we use to demonstrate dynamic emotion prediction. <s> BIB010 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> We collected and analyzed subjective and objective data using surveys and wearable sensors worn day and night from 68 participants, for 30 days each, to address questions related to the relationships among sleep duration, sleep irregularity, self-reported Happy-Sad mood and other factors in college students. We analyzed daily and monthly behavior and physiology and identified factors that affect mood, including how accurately sleep duration and sleep regularity for the past 1-5 days classified the participants into high/low mood using support vector machines. We found statistically significant associations among sad mood and poor health-related factors. Behavioral factors such as the percentage of neutral social interactions and the total academic activity hours showed the best performance in separating the Happy-Sad mood groups. Sleep regularity was a more important discriminator of mood than sleep duration for most participants, although both variables predicted happy/sad mood with from 70-82% accuracy. The number of nights giving the best prediction of happy/sad mood varied for different groups of individuals. <s> BIB011 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Today's health care is difficult to imagine without the possibility to objectively measure various physiological parameters related to patients' symptoms (from temperature through blood pressure to complex tomographic procedures). Psychiatric care remains a notable exception that heavily relies on patient interviews and self-assessment. This is due to the fact that mental illnesses manifest themselves mainly in the way patients behave throughout their daily life and, until recently there were no "behavior measurement devices." This is now changing with the progress in wearable activity recognition and sensor enabled smartphones. In this paper, we introduce a system, which, based on smartphone-sensing is able to recognize depressive and manic states and detect state changes of patients suffering from bipolar disorder. Drawing upon a real-life dataset of ten patients, recorded over a time period of 12 weeks (in total over 800 days of data tracing 17 state changes) by four different sensing modalities, we could extract features corresponding to all disease-relevant aspects in behavior. Using these features, we gain recognition accuracies of 76% by fusing all sensor modalities and state change detection precision and recall of over 97%. This paper furthermore outlines the applicability of this system in the physician-patient relations in order to facilitate the life and treatment of bipolar patients. <s> BIB012 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> This paper focuses on human activity recognition (HAR) problem, in which inputs are multichannel time series signals acquired from a set of bodyworn inertial sensors and outputs are predefined human activities. In this problem, extracting effective features for identifying activities is a critical but challenging task. Most existing work relies on heuristic hand-crafted feature design and shallow feature learning architectures, which cannot find those distinguishing features to accurately classify different activities. In this paper, we propose a systematic feature learning method for HAR problem. This method adopts a deep convolutional neural networks (CNN) to automate feature learning from the raw inputs in a systematic way. Through the deep architecture, the learned features are deemed as the higher level abstract representation of low level raw time series signals. By leveraging the labelled information via supervised learning, the learned features are endowed with more discriminative power. Unified in one model, feature learning and classification are mutually enhanced. All these unique advantages of the CNN make it outperform other HAR algorithms, as verified in the experiments on the Opportunity Activity Recognition Challenge and other benchmark datasets. <s> BIB013 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> We report the first demonstration of a graphene-based epidermal sensor system (GESS) with total thickness below 500 nm. The GESS is manufactured by the cost-effective and rapid “cut-and-paste” method on tattoo paper and can be directly laminated on human skin like a temporary transfer tattoo. Without any tape or adhesive, the GESS completely conforms to the microscopic morphology of human skin via van der Waals interaction. The softness and transparency of the GESS, make it the world's first epidermal sensor system that is invisible both mechanically and optically. The GESS has been successfully applied to measure electrocardiogram (ECG), electroencephalogram (EEG) and electromyogram (EMG) with signal-to-noise ratio comparable with commercial electrodes, in addition to skin temperature and skin hydration. The thin and transparent graphene epidermal sensor can be used for the first time enable simultaneous electrical and optical epidermal sensing. <s> BIB014 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Wearable sensor technologies are essential to the realization of personalized medicine through continuously monitoring an individual’s state of health. Sampling human sweat, which is rich in physiological information, could enable non-invasive monitoring. Previously reported sweat-based and other non-invasive biosensors either can only monitor a single analyte at a time or lack on-site signal processing circuitry and sensor calibration mechanisms for accurate analysis of the physiological state. Given the complexity of sweat secretion, simultaneous and multiplexed screening of target biomarkers is critical and requires full system integration to ensure the accuracy of measurements. Here we present a mechanically flexible and fully integrated (that is, no external analysis is needed) sensor array for multiplexed in situ perspiration analysis, which simultaneously and selectively measures sweat metabolites (such as glucose and lactate) and electrolytes (such as sodium and potassium ions), as well as the skin temperature (to calibrate the response of the sensors). Our work bridges the technological gap between signal transduction, conditioning (amplification and filtering), processing and wireless transmission in wearable biosensors by merging plastic-based sensors that interface with the skin with silicon integrated circuits consolidated on a flexible circuit board for complex signal processing. This application could not have been realized using either of these technologies alone owing to their respective inherent limitations. The wearable system is used to measure the detailed sweat profile of human subjects engaged in prolonged indoor and outdoor physical activities, and to make a real-time assessment of the physiological state of the subjects. This platform enables a wide range of personalized diagnostic and physiological monitoring applications. <s> BIB015 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Flexible, wearable sensing devices can yield important information about the underlying physiology of a human subject for applications in real-time health and fitness monitoring. Despite significant progress in the fabrication of flexible biosensors that naturally comply with the epidermis, most designs measure only a small number of physical or electrophysiological parameters, and neglect the rich chemical information available from biomarkers. Here, we introduce a skin-worn wearable hybrid sensing system that offers simultaneous real-time monitoring of a biochemical (lactate) and an electrophysiological signal (electrocardiogram), for more comprehensive fitness monitoring than from physical or electrophysiological sensors alone. The two sensing modalities, comprising a three-electrode amperometric lactate biosensor and a bipolar electrocardiogram sensor, are co-fabricated on a flexible substrate and mounted on the skin. Human experiments reveal that physiochemistry and electrophysiology can be measured simultaneously with negligible cross-talk, enabling a new class of hybrid sensing devices. <s> BIB016 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> This paper presents results of analysis performed on a physiologic time-series dataset that was collected from a wearable ECG monitoring system worn by individuals who suffer from panic disorder. Models are constructed and evaluated for distinguishing between pathologic and non-pathologic states, including panic (during panic attack), pre-panic (preceding panic attack) and non-panic (outside panic attack window). The models presented use data fusion to combine both traditional time and frequency domain heart rate variability analysis together with nonlinear/complexity analysis. The best performing model is shown to be a random forest classifier that achieves an accuracy of 97.2% and 90.7% for recognizing states of panic and pre-panic, respectively. The models presented have application in pervasive and ubiquitous mobile and wearable health management systems. <s> BIB017 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> The use of deep learning for the activity recognition performed by wearables, such as smartwatches, is an understudied problem. To advance current understanding in this area, we perform a smartwatch-centric investigation of activity recognition under one of the most popular deep learning methods — Restricted Boltzmann Machines (RBM). This study includes a variety of typical behavior and context recognition tasks related to smartwatches (such as transportation mode, physical activities and indoor/outdoor detection) to which RBMs have previously never been applied. Our findings indicate that even a relatively simple RBM-based activity recognition pipeline is able to outperform a wide-range of common modeling alternatives for all tested activity classes. However, usage of deep models is also often accompanied by resource consumption that is unacceptably high for constrained devices like watches. Therefore, we complement this result with a study of the overhead of specifically RBM-based activity models on representative smartwatch hardware (the Snapdragon 400 SoC, present in many commercial smartwatches). These results show, contrary to expectation, RBM models for activity recognition have acceptable levels of resource use for smartwatch-class hardware already on the market. Collectively, these two experimental results make a strong case for more widespread adoption of deep learning techniques within smartwatch designs moving forward. <s> BIB018 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Spasticity is a condition that impairs voluntary muscle movements and physically debilitates persons across several neurological disorders, including stroke, multiple sclerosis and cerebral palsy. Assessing the progression of spasticity during clinical interventions and at home is key to rehabilitation efficacy and care management. Here we present electromyography (EMG) and motion data using skin-mounted, flexible and wireless sensors in a cohort of 13 individuals with stroke. We compute a set of 15 features from the EMG data and use machine learning to infer whether spasticity is present during movements of the knee and ankle joints. Using a Linear Discriminant Analysis (LDA) classifier, we show that our approach successfully discriminates voluntary contractions from spastic muscle contractions (AUC=0.94). These results show that continuous and non-invasive monitoring of spasticity symptoms could be applied to optimize and personalize rehabilitation regimens. <s> BIB019 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Traditional manufacturing methods and materials used to fabricate epidermal electronics for physiological monitoring, transdermal stimulation, and therapeutics are complex and expensive, preventing their adoption as single-use medical devices. This work describes the fabrication of epidermal, paper-based electronic devices (EPEDs) for wearable and implantable applications by combining the spray-based deposition of silanizing agents, highly conductive nanoparticles, and encapsulating polymers with laser micromachining. EPEDs are inexpensive, stretchable, easy to apply, and disposable by burning. The omniphobic character and fibrous structure of EPEDs make them breathable, mechanically stable upon stretching, and facilitate their use as electrophysiological sensors to record electrocardiograms, electromyograms, and electrooculograms, even under water. EPEDs can also be used to provide thermotherapeutic treatments to joints, map temperature spatially, and as wirelessly powered implantable devices for stimulation and therapeutics. This work makes epidermal electronic devices accessible to high-throughput manufacturing technologies and will enable the fabrication of a variety of wearable medical devices at a low cost. <s> BIB020 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> The ability to recognize emotion is one of the hallmarks of emotional intelligence, an aspect of human intelligence that has been argued to be even more important than mathematical and verbal intelligences. This paper proposes that machine intelligence needs to include emotional intelligence and demonstrates results toward this goal: developing a machine's ability to recognize the human affective state given four physiological signals. We describe difficult issues unique to obtaining reliable affective data and collect a large set of data from a subject trying to elicit and experience each of eight emotional states, daily, over multiple weeks. This paper presents and compares multiple algorithms for feature-based recognition of emotional state from this data. We analyze four physiological signals that exhibit problematic day-to-day variations: The features of different emotions on the same day tend to cluster more tightly than do the features of the same emotion on different days. To handle the daily variations, we propose new features and algorithms and compare their performance. We find that the technique of seeding a Fisher Projection with the results of sequential floating forward search improves the performance of the Fisher Projection and provides the highest recognition rates reported to date for classification of affect from physiology: 81 percent recognition accuracy on eight classes of emotion, including neutral. <s> BIB021 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> Abstract Being able to detect stress as it occurs can greatly contribute to dealing with its negative health and economic consequences. However, detecting stress in real life with an unobtrusive wrist device is a challenging task. The objective of this study is to develop a method for stress detection that can accurately, continuously and unobtrusively monitor psychological stress in real life. First, we explore the problem of stress detection using machine learning and signal processing techniques in laboratory conditions, and then we apply the extracted laboratory knowledge to real-life data. We propose a novel context-based stress-detection method. The method consists of three machine-learning components: a laboratory stress detector that is trained on laboratory data and detects short-term stress every 2 min; an activity recognizer that continuously recognizes the user’s activity and thus provides context information; and a context-based stress detector that uses the outputs of the laboratory stress detector, activity recognizer and other contexts, in order to provide the final decision on 20-min intervals. Experiments on 55 days of real-life data showed that the method detects (recalls) 70% of the stress events with a precision of 95%. <s> BIB022 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> In affective computing (AC) field studies it is impossible to obtain an objective ground truth. Hence, self-reports in form of ecological momentary assessments (EMAs) are frequently used in lieu of ground truth. Based on four paradigms, we formulate practical guidelines to increase the accuracy of labels generated via EMAs. In addition, we detail how these guidelines were implemented in a recent AC field study of ours. During our field study, 1081 EMAs were collected from 10 subjects over a duration of 148 days. Based on these EMAs, we perform a qualitative analysis of the effectiveness of our proposed guidelines. Furthermore, we present insights and lessons learned from the field study. <s> BIB023 </s> Wearable-Based Affect Recognition—A Review <s> Discussion And Outlook <s> An increase in the collection of physiological signals, whether done implicitly in wearable or IoT device or explicitly in experimental and laboratory environments, creates the need for development of smart systems and tools capable of data analysis with limited expert knowledge. For instances in which an expert is absolutely necessary, a method for filtering out uninteresting data is necessary to reduce the amount of data storage necessary as well as reduce the amount of data the expert will have to analyze. Anomaly detection, specifically unsupervised anomaly detection can be used to design a tool or even as a tool to help remedy these issues. This paper will focus on how unsupervised anomaly detection can be utilized for the development of such systems. A systematic, robust, and customizable approach will be presented and preliminary results will be shown that open the door for future research and algorithm development. <s> BIB024

Based on the previous sections, we would like to go one step further and identify key challenges and opportunities in wearable-based affect recognition (AR). We will focus on the following key challenges in this section: (a) valence detection; (b) hardware; (c) datasets; (d) algorithmic challenges; and (e) long-term reasoning. Valence detection: From Sections 2 and 3, the link between physiological changes and the arousal axis of the circumplex model became apparent. Hence, it is not surprising that approaches of stress detection and arousal assessment in Table 9 reach high accuracies. However, valence-related changes in human physiology are more subtle and, therefore, difficult to detect. This explains the lower accuracy of valence detection systems in Table 9 . In some studies BIB006 BIB010 , facial expressions, which are directly connected to valence (e.g., smiles), were recorded using facial electromyogram. However, this procedure is not applicable in everyday life due to practical considerations. One possibility to improve the assessment of valence is to incorporate contextual data into the classification process. This contextual information can range from audio samples (e.g., detection laughter), information about the sleep qualit, to calendar meta data or text (e.g., emails/chat). Following for instance Sano et al. BIB011 , the regularity of sleep and duration has a very strong impact on the mood of a person and is a strong feature to predict the morning mood. Hardware: The setups used to record physiological data in affect recognition studies are often either watch-like (e.g., Empatica E4 [168] ), chest-belt (e.g., AutoSense ) or stationary devices (e.g., BioPac systems [170] ). Recent progress in flexible electronics enabled the development of sensor patches (e.g., Vivalnk [171] ) and epidermal electronics. In recent research, the potential of epidermal electronics measuring different electrophysiological signals, like electrocardiogram, electromyogram, and even electroencephalogram has been demonstrated BIB020 BIB014 . Up to now patches and epidermal electronics have found little application in affect recognition (field) studies. Furthermore, sensors and processing units can be integrated into fabric (for a comprehensive summary see Reiss and Amft ). These technologies offer an increased wearing comfort, potential new measurement positions BIB019 , and are, similar to smartwatches, only minimally intrusive. Hence, they certainly deserve more attention in wearable-based affect recognition. In addition to the traditionally employed set of modalities (electrocardiogram, electrodermal activity, etc.), the merits of other sensors should be explored. First, considering the cardiac system, stress has been related to changes in blood pressure BIB001 . Hence, incorporating data representing a blood pressure correlate (e.g., pulse wave transit time BIB005 ) could enable more reliable stress detectors. Second, body microphones placed on the subject's chest or abdomen could provide further insights into the cardiac BIB003 , respiration, and digestive system. Third, the chemical composition of perspiration could provide further information about the physiological state of a person. Hence, integrating chemical-electrophysiological sensors BIB015 BIB016 in affect recognition studies has the potential to create new insights into the physiology of affective states. Finally, as already mentioned above, contextual information about the user could help to improve the classification. The sources of contextual data are nearly unlimited and range from ambient audio data to video streams provided by devices like smartglasses (e.g., Google Glass). These sources could be used to classify the surroundings of a user and the affective state of other nearby persons as well. Datasets: The wearable-based affect recognition community lacks publicly available datasets, frequently used for benchmarking. In order to generate statistically meaningful results, a representative cohort of subjects is desirable. However, most affect recognition studies target students or research staff, which are likely to represent a homogeneous group . In order to mitigate this selection bias, studies could recruit subjects from different social groups (gender, age, etc.). In addition, cross-cultural data collection would be very interesting in order to facilitate universal wearable-based affect recognition. Furthermore, as shown by Grünerbl et al. BIB012 or Rubin et al. BIB017 , affect recognition systems can find application in clinical settings. However, in order to facilitate this type of application, datasets containing data from patients with specific health conditions are required. The available datasets (see Section 4.3) already feature multiple modalities. Measuring physiological changes in a redundant fashion (e.g., using electrocardiogram and photoplethysmogram) or using the same modality on various locations (e.g., wrist and torso) would facilitate a direct comparison of the signals. Studies on wearable/based emotion detection commonly elicit and detect multiple emotional states BIB021 BIB006 . In contrast, stress detection systems mainly target binary problems (stress versus no-stress). In our opinion, robust affect recognition systems should be trained on datasets like WESAD BIB023 , which include redundant data streams and different affective state (stress, amusement, and neutral). Up-to-date affect recognition research based on wearables mainly focus on lab studies. For benchmarking and exploitative studies, lab data is a good starting point. Hence, we hope that the observed trend towards field studies (see Table 9 ) continues. To support this trend, we provided in Section 4.2.1 practical guidelines on ground truth generation in field studies. Algorithmic challenges: The way humans perceive and react to an affective stimulus is subject dependent. This highlights the importance of personalisation. However, the current state-of-the-art in wearable-based affect recognition makes little use of personalisation methods. One way to account for the subjective nature of affective states is to utilise online learning. Following this idea, a general model could be deployed, which is then customised. Customisation could happen, for instance, via an active labelling approach, where the user is occasionally asked to provide labels. In addition, semi-supervised or even unsupervised training methods could be used. To the best of our knowledge, these methods have not found application in wearable-based affect recognition research yet. Healthy subjects are unlikely to exhibit strong swings over the entire affective spectrum. Hence, in order to identify the rare extreme cases methods from anomaly detection could be applied BIB024 . In most studies presented in Table 9 classical feature-based machine learning algorithms (e.g., support vector machine, k-nearest neighbour, etc.) were employed. In human activity recognition BIB013 , audio analysis BIB008 or stock return forecasting BIB002 , which all deal with time series data, (deep) neural networks (NN) proved to be powerful classifiers. Using convolutional neural network (CNN) makes feature engineering obsolete, as via backpropagation features are learned. From a methodical point of view CNN offer interesting approaches to transfer or semi-supervised BIB004 learning. Deep NN require a large amount of training data and are known to be resource intensive. Hence, deployment on an embedded device is an open research question. However, first approaches to deploy such models on embedded devices were presented in Reference BIB018 . Hence, due to strong interest in NN from both academia and industry, we are confident that resource-related issues will be solved in near future. Long-term reasoning: Image-based affect recognition systems can only perform a temporal-and spatial-limited assessment of the user's state (e.g., while driving [16] ). In contrast, wearable-based affect recognition systems detect the user's affective state continuously and ubiquitously. This can be used for a deeper analysis, providing reasoning for certain affective states or behavioural patterns. First approaches of long-term reasoning were presented by Gjoreski et al. BIB022 and in the HappyMeter App . The latter investigated correlations between affective states and environmental conditions (e.g., temperature, wind, humidity) or persons nearby. Visualizing this information can increase awareness of specific situations (e.g., showing locations where the user is stressed). Essential for this correlation analysis, is contextual information. We see a large potential for this research direction, as the reasoning methods presented above are still in an early stage. The aim of this review was to provide a broad overview and in-depth understanding of the theoretical background, methods, and best practices of wearable-based affect recognition. Currently, there is a strong trend to small, lightweight, affordable, and wearable electronic gadgets. These devices can be used for sensing, storing, and data processing BIB007 . Hence, they offer an ideal platform for enhanced affect recognition systems. There is a wide range of applications for such systems, in particular in the consumer and healthcare domain. From a healthcare point of view, wearable-based affect recognition systems could, for instance, help to ubiquitously monitor the state of patients with mental disorders (e.g., depression). This data could provide valuable insights for therapists, promoting behaviour change interventions . Furthermore, these systems could facilitate the development of tele-mental BIB009 and tele-medical applications. Wearable-based affect recognition systems could improve self monitoring, provide users with a better understanding of their affective states, and support behavioural changes. Beyond these health-related applications, affect recognition systems could be used in urban planning or to improve human-machine interfaces. Despite the impressive progress made in recent years, the applications mentioned above are still under research and not available for customers. We are convinced that robust and personalised affect recognition systems applicable in everyday life could provide many users with an added value. Hence, we encourage the community to support and address the remaining challenges. Author Contributions: Conceptualization, P.S., A.R., R.D., and K.V.L.; formal analysis, P.S., A.R.; resources, P.S.; data curation, P.S.; writing-original draft, P.S., A.R., R.D., and K.V.L.; writing-review and editing, P.S., A.R, and K.V.L.; Supervision, K.V.L.; project administration, P.S.; funding acquisition, K.V.L.; Funding: This research received no external funding.

Survey of Large-Scale MIMO Systems <s> W <s> The signal detection algorithm of the vertical BLAST (Bell Laboratories Layered Space-Time) wireless communications architecture is briefly described. Using this joint space-time approach, spectral efficiencies ranging from 20-40 bit/s/Hz have been demonstrated in the laboratory under flat fading conditions at indoor fading rates. Early results are presented. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> W <s> Wireless data traffic is expected to grow over the next few years and the technologies that will provide data services are still being debated. One possibility is to use multiple antennas at base stations and terminals to get very high spectral efficiencies in rich scattering environments. Such multiple-input/multiple-output (MIMO) channels can then be used in conjunction with scheduling and rate-feedback algorithms to further increase channel throughput. This paper provides an analysis of the expected gains due to scheduling and bits needed for rate feedback. Our analysis requires an accurate approximation of the distribution of the MIMO channel mutual information. Because the exact distribution of the mutual information in a Rayleigh-fading environment is difficult to analyze, we prove a central limit theorem for MIMO channels with a large number of antennas. While the growth in average mutual information (capacity) of a MIMO channel with the number of antennas is well understood, it turns out that the variance of the mutual information can grow very slowly or even shrink as the number of antennas grows. We discuss implications of this "channel-hardening" result for data and voice services, scheduling, and rate feedback. We also briefly discuss the implications when shadow fading effects are included. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> W <s> Coding for MIMO Communication Systems is a comprehensive introduction and overview to the various emerging coding techniques developed for MIMO communication systems. The basics of wireless communications and fundamental issues of MIMO channel capacity are introduced and the space-time block and trellis coding techniques are covered in detail. Other signaling schemes for MIMO channels are also considered, including spatial multiplexing, concatenated coding and iterative decoding for MIMO systems, and space-time coding for non-coherent MIMO channels. Practical issues including channel correlation, channel estimation and antenna selection are also explored, with problems at the end of each chapter to clarify many important topics. ::: ::: ::: A comprehensive book on coding for MIMO techniques covering main strategies ::: Theories and practical issues on MIMO communications are examined in detail ::: Easy to follow and accessible for both beginners and experienced practitioners in the field ::: References at the end of each chapter for further reading ::: Can be used with ease as a research book, or a textbook on a graduate or advanced undergraduate level course ::: ::: ::: ::: This book is aimed at advanced undergraduate and postgraduate students, researchers and practitioners in industry, as well as individuals working for government, military, science and technology institutions who would like to learn more about coding for MIMO communication systems. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> W <s> A cellular base station serves a multiplicity of single-antenna terminals over the same time-frequency interval. Time-division duplex operation combined with reverse-link pilots enables the base station to estimate the reciprocal forward- and reverse-link channels. The conjugate-transpose of the channel estimates are used as a linear precoder and combiner respectively on the forward and reverse links. Propagation, unknown to both terminals and base station, comprises fast fading, log-normal shadow fading, and geometric attenuation. In the limit of an infinite number of antennas a complete multi-cellular analysis, which accounts for inter-cellular interference and the overhead and errors associated with channel-state information, yields a number of mathematically exact conclusions and points to a desirable direction towards which cellular wireless could evolve. In particular the effects of uncorrelated noise and fast fading vanish, throughput and the number of terminals are independent of the size of the cells, spectral efficiency is independent of bandwidth, and the required transmitted energy per bit vanishes. The only remaining impairment is inter-cellular interference caused by re-use of the pilot sequences in other cells (pilot contamination) which does not vanish with unlimited number of antennas. <s> BIB004 </s> Survey of Large-Scale MIMO Systems <s> W <s> Energy efficiency in cellular networks is a growing concern for cellular operators to not only maintain profitability, but also to reduce the overall environment effects. This emerging trend of achieving energy efficiency in cellular networks is motivating the standardization authorities and network operators to continuously explore future technologies in order to bring improvements in the entire network infrastructure. In this article, we present a brief survey of methods to improve the power efficiency of cellular networks, explore some research issues and challenges and suggest some techniques to enable an energy efficient or "green" cellular network. Since base stations consume a maximum portion of the total energy used in a cellular system, we will first provide a comprehensive survey on techniques to obtain energy savings in base stations. Next, we discuss how heterogeneous network deployment based on micro, pico and femto-cells can be used to achieve this goal. Since cognitive radio and cooperative relaying are undisputed future technologies in this regard, we propose a research vision to make these technologies more energy efficient. Lastly, we explore some broader perspectives in realizing a "green" cellular network technology <s> BIB005 </s> Survey of Large-Scale MIMO Systems <s> W <s> Traditional mobile wireless network mainly design focuses on ubiquitous access and large capacity. However, as energy saving and environmental protection become global demands and inevitable trends, wireless researchers and engineers need to shift their focus to energy-efficiency-oriented design, that is, green radio. In this article, we propose a framework for green radio research and integrate the fundamental issues that are currently scattered. The skeleton of the framework consists of four fundamental tradeoffs: deployment efficiency-energy efficiency, spectrum efficiency-energy efficiency, bandwidth-power, and delay-power. With the help of the four fundamental trade-offs, we demonstrate that key network performance/cost indicators are all strung together. <s> BIB006 </s> Survey of Large-Scale MIMO Systems <s> W <s> Multi-user MIMO offers big advantages over conventional point-to-point MIMO: it works with cheap single-antenna terminals, a rich scattering environment is not required, and resource allocation is simplified because every active terminal utilizes all of the time-frequency bins. However, multi-user MIMO, as originally envisioned, with roughly equal numbers of service antennas and terminals and frequency-division duplex operation, is not a scalable technology. Massive MIMO (also known as large-scale antenna systems, very large MIMO, hyper MIMO, full-dimension MIMO, and ARGOS) makes a clean break with current practice through the use of a large excess of service antennas over active terminals and time-division duplex operation. Extra antennas help by focusing energy into ever smaller regions of space to bring huge improvements in throughput and radiated energy efficiency. Other benefits of massive MIMO include extensive use of inexpensive low-power components, reduced latency, simplification of the MAC layer, and robustness against intentional jamming. The anticipated throughput depends on the propagation environment providing asymptotically orthogonal channels to the terminals, but so far experiments have not disclosed any limitations in this regard. While massive MIMO renders many traditional research problems irrelevant, it uncovers entirely new problems that urgently need attention: the challenge of making many low-cost low-precision components that work effectively together, acquisition and synchronization for newly joined terminals, the exploitation of extra degrees of freedom provided by the excess of service antennas, reducing internal power consumption to achieve total energy efficiency reductions, and finding new deployment scenarios. This article presents an overview of the massive MIMO concept and contemporary research on the topic. <s> BIB007 </s> Survey of Large-Scale MIMO Systems <s> W <s> Presents a list of articles published by the IEEE Signal Processing Society (SPS) that ranked among the top 100 most downloaded IEEE Xplore articles. <s> BIB008 </s> Survey of Large-Scale MIMO Systems <s> W <s> We consider the uplink (UL) and downlink (DL) of non-cooperative multi-cellular time-division duplexing (TDD) systems, assuming that the number N of antennas per base station (BS) and the number K of user terminals (UTs) per cell are large. Our system model accounts for channel estimation, pilot contamination, and an arbitrary path loss and antenna correlation for each link. We derive approximations of achievable rates with several linear precoders and detectors which are proven to be asymptotically tight, but accurate for realistic system dimensions, as shown by simulations. It is known from previous work assuming uncorrelated channels, that as N→∞ while K is fixed, the system performance is limited by pilot contamination, the simplest precoders/detectors, i.e., eigenbeamforming (BF) and matched filter (MF), are optimal, and the transmit power can be made arbitrarily small. We analyze to which extent these conclusions hold in the more realistic setting where N is not extremely large compared to K. In particular, we derive how many antennas per UT are needed to achieve η% of the ultimate performance limit with infinitely many antennas and how many more antennas are needed with MF and BF to achieve the performance of minimum mean-square error (MMSE) detection and regularized zero-forcing (RZF), respectively. <s> BIB009 </s> Survey of Large-Scale MIMO Systems <s> W <s> A multiplicity of autonomous terminals simultaneously transmits data streams to a compact array of antennas. The array uses imperfect channel-state information derived from transmitted pilots to extract the individual data streams. The power radiated by the terminals can be made inversely proportional to the square-root of the number of base station antennas with no reduction in performance. In contrast if perfect channel-state information were available the power could be made inversely proportional to the number of antennas. Lower capacity bounds for maximum-ratio combining (MRC), zero-forcing (ZF) and minimum mean-square error (MMSE) detection are derived. An MRC receiver normally performs worse than ZF and MMSE. However as power levels are reduced, the cross-talk introduced by the inferior maximum-ratio receiver eventually falls below the noise level and this simple receiver becomes a viable option. The tradeoff between the energy efficiency (as measured in bits/J) and spectral efficiency (as measured in bits/channel use/terminal) is quantified for a channel model that includes small-scale fading but not large-scale fading. It is shown that the use of moderately large antenna arrays can improve the spectral and energy efficiency with orders of magnitude compared to a single-antenna system. <s> BIB010 </s> Survey of Large-Scale MIMO Systems <s> W <s> This paper investigates the performance for the downlink massive multiple-input multiple-output system when the base station serves multiple user terminals UTs using zero-forcing precoding. The tight lower bound of average area spectrum efficiency A2SE is derived as a function of the number of transmission antennas, the number of UTs and the equivalent transmission signal-to-noise ratio. Regarding the lower bound as the approximate A2SE, the optimal number of UTs maximising the A2SE is attained for given the number of transmission antennas and the equivalent transmission signal-to-noise ratio. After formulating the realistic power consumption model in consideration of both the transmission power and circuit power, the trade-off between energy efficiency EE and A2SE is established, and the optimal EE with respect to A2SE is deduced. Moreover, both the outage probability and bit error ratio expressions are derived related to the degrees of freedom, defined as the difference between the number of transmission antennas and the number of UTs. Simulation results coincide to the analysis well, and they indicate that deploying more transmission antennas or multiplexing a rational number of UTs can improve the A2SE and EE, increasing the degrees of freedom will better both the outage probability and bit error ratio. Copyright © 2013 John Wiley & Sons, Ltd. <s> BIB011 </s> Survey of Large-Scale MIMO Systems <s> W <s> Massive multiple-input multiple-output (MIMO) wireless communications refers to the idea equipping cellular base stations (BSs) with a very large number of antennas, and has been shown to potentially allow for orders of magnitude improvement in spectral and energy efficiency using relatively simple (linear) processing. In this paper, we present a comprehensive overview of state-of-the-art research on the topic, which has recently attracted considerable attention. We begin with an information theoretic analysis to illustrate the conjectured advantages of massive MIMO, and then we address implementation issues related to channel estimation, detection and precoding schemes. We particularly focus on the potential impact of pilot contamination caused by the use of non-orthogonal pilot sequences by users in adjacent cells. We also analyze the energy efficiency achieved by massive MIMO systems, and demonstrate how the degrees of freedom provided by massive MIMO systems enable efficient single-carrier transmission. Finally, the challenges and opportunities associated with implementing massive MIMO in future wireless communications systems are discussed. <s> BIB012

ITH the evolution of smart terminals and their applications, the need for multimedia services rapidly increases recently. Thus, the capacity of wireless communications networks has to be increased in order to guarantee the Quality of Service (QoS) requirements of mobile applications. Meanwhile, telecommunication manufactures and operators have also foreseen that the load of wireless communications networks is increasing exponentially . Therefore, it is necessary to introduce new technologies to meet the demands of explosive traffic for next-generation wireless communications networks. Bandwidth Efficiency (BE) is usually one of the most important metrics to select candidate technologies for next-generation wireless communications systems. Meanwhile, with excessive power consumption in wireless communications networks, both carbon emissions and operator expenditure increase year by year BIB005 , . As a result, Energy Efficiency (EE) has become another significant metric for evaluating the performances of wireless communications systems with some given BE constraints BIB006 - . Multiple-Input and Multiple-Output (MIMO) technology has attracted much attention in wireless communications, because it offers significant increases in data throughput and link range without an additional increase in bandwidth or transmit power. In 1993 and 1994, a MIMO approach was proposed and the corresponding patent was issued , where multiple transmit antennas are co-located at one transmitter with the objective of improving the attainable link throughput. Then, the first laboratory prototype of spatial multiplexing was implemented to demonstrate the practical feasibility of MIMO technology BIB001 . Nowadays, MIMO has been accepted as one of key technologies in the Fourth Generation (4G) wireless communications systems. When an evolved Node B (eNB) equipped with multiple antennas communicates with several User Terminals (UEs) at the same time-frequency resources, it is referred to as Multi-User MIMO (MU-MIMO). MU-MIMO is capable of improving either the BE or the reliability by improving either the multiplexing gains or diversity gains BIB003 . In order to scale up these gains, the Large-Scale MIMO (LS-MIMO) concept, which is also known as massive MIMO scheme often also associated with the terminologies of large-scale antenna systems, very large MIMO, very large multi-user-MIMO, fulldimensional MIMO, hyper MIMO, etc, was proposed by Marzetta in BIB004 . More explicitly, a LS-MIMO refers to the system that uses hundreds of antennas to simultaneously serve dozens of UEs. Both theoretical and measurement results indicate that a LS-MIMO is capable of significantly improving the BE, which simultaneously reducing the transmit power BIB007 , BIB008 . As a result, a LS-MIMO is regarded as a candidate technique for next-generation wireless communications systems conceived for the sake of improving both their BE and EE. As the down tilt of an Antenna Array (AA) is fixed, traditional MIMO technology can only adjust signal transmission in the horizontal dimension. In order to exploit the vertical dimension of signal propagation, AAs, such as rectangular, spherical and cylindrical AAs, were studied by the 3rd Generation Partnership Project (3GPP) - . MIMO with these arrays can adjust both azimuth and elevation angles, and propagate signals in Three-Dimensional (3D) space, thus termed 3D MIMO. To further increase capacity, 3D MIMO deploys more antennas to achieve larger multiplexing gains. Meanwhile, LS-MIMO adopts rectangular, spherical or cylindrical AAs in practical systems considering the space of AAs. Therefore, 3D MIMO with massive antennas can be seen as a practical deployment means of LS-MIMO, and both of them are investigated in this paper. LS-MIMO can improve BE since it can achieve large multiplexing gains when serving tens of UEs simultaneously BIB004 , BIB009 . The significant increase in EE is due to the fact that the use of more antennas helps focus energy with an extremely narrow beam on small regions where the UEs are located BIB010 . Apart from these advantages, LS-MIMO can enhance transmission reliability owing to the excessive Degrees of Freedom (DoF) BIB011 . Inter-User Interference (IUI) can also be alleviated because of the extreme narrow beam BIB007 . In an LS-MIMO system, individual element failure of the AA is not detrimental to the performance of the entire system BIB007 . Simple low-complexity signal processing algorithms are capable of approximating the performance achieved by optimal methods, such as MaximumLikelihood (ML) multiuser detection and Dirty Paper Coding (DPC) BIB008 . The latency of the air interface can be reduced and LS-MIMO the protocols at the Media Access Control (MAC) layer can be simplified because of the channel harden phenomenon and sufficient capacity BIB002 . Certainly, the complexity of signal processing, including Transmit Precoding (TPC), channel estimation and detection, increases with the number of antennas. On the other hand, the maximum number of orthogonal pilot sequences is limited by the coherence interval and coherence bandwidth. Therefore, the performance of LS-MIMO systems is constrained by pilot contamination due to pilot reuse in multi-cell scenarios BIB004 . Moreover, compared to the Physical Downlink Shared Channel (PDSCH) employing either precoding or beamforming, the Signal-to-Interference-plus-Noise Ratio (SINR) of the Physical Broadcasting Channel (PBCH) is lower due to the omnidirectional signal transmission . Currently, fundamental theoretical problems and several physical layer techniques of LS-MIMO have been already widely investigated. For example, the capacity and the realistic performance of precoding and detection for LS-MIMO have been analyzed from the viewpoint of information theory BIB008 , BIB012 . Additionally, the advantages and disadvantages, the potential applications and limitations of LS-MIMO were generally given in BIB007 . However, the performance of LS-MIMO is much more affected by practical factors, which are not well summarized so far. Therefore, besides a comprehensive investigation on the theoretical performance of LS-MIMO, this paper pays more attention to the discussions of issues from the system point of view, such as practical channel models under different AAs, practical application scenarios for LS-MIMO, networking techniques, and so on. According to the current literature related to LS-MIMO, the major research directions about LS-MIMO are listed in Table I , some of which have been investigated in BIB007 , BIB008 and BIB012 while others are not. The remainder of this paper is organized as shown in Fig. 1 . The measurement and modeling of LS-MIMO channels are discussed in Section II. Section III introduces main scenarios and applications in wireless communications networks. The theoretical and measured performances of LS-MIMO with different precoders and detectors are discussed for both single-cell and multi-cell scenarios in consideration of perfect and imperfect Channel State Information (CSI) in Section IV. Section V investigates related networking techniques, such as Inter-Cell Interference Coordination (ICIC) and scheduling. Section VI identifies challenges and research directions, and Section VII concludes this paper.

Survey of Large-Scale MIMO Systems <s> A. Antenna Configurations <s> Preface. 1 Antennas and Antenna Arrays. 1.1 Introduction. 1.2 Antenna Array Factor. 1.3 Elements and Array Types. 1.4 Antenna Parameters and Indices. 1.5 Antenna Input Impedance. 1.6 Antenna Arrays Classification. 1.7 Array Factor Classification. References. 2 Arrays: Linear, Planar, 3D and Conformal. 2.1 Introduction. 2.2 Linear Arrays. 2.3 Uniform Linear Arrays. 2.4 Chebyshev Linear Arrays. 2.5 Linear Arrays from Sampling or Root Matching of Line Sources. 2.6 Planar Arrays. 2.7 3-D Arrays. 2.8 Conformal Arrays. References. 3 Pattern Synthesis for Arrays. 3.1 Introduction. 3.2 Uniform Linear Array Synthesis. 3.3 Chebyshev Array Synthesis. 3.4 Synthesis by Sampling or by Root Matching. 3.5 Synthesis by Fourier Transform. 3.6 The Woodward - Lawson (WL) Method. 3.7 Array Synthesis as an Optimization Problem. 3.8 Synthesis by Convolution of Linear, Planar and 3-D Arrays. References. 4 The Orthogonal Methods. 4.1 Introduction. 4.2 Synthesis of Non-uniformly Spaced Linear Arrays: The Matrix Inversion Method. 4.3 Synthesis of Non-uniformly Spaced Linear Arrays: The Orthogonal Method. 4.5 Quantized Excitation and Geometry Synthesis of a Linear Array: The Orthogonal Perturbation Method. 4.6 Synthesis of Non-uniformly Spaced Planar Arrays: The Orthogonal Method. 4.7 Synthesis of Non-uniformly Spaced 3-D Arrays: The Orthogonal Method. 4.8 Synthesis of Non-uniformly Spaced 3-D Arrays with Arbitrarily Oriented Dipoles: The Non-parallel Orthogonal Method. 4.9 Synthesis of Arrays of Wire Antennas: The MoM Orthogonal Method. 4.10 Synthesis of General Antenna Arrays: The Orthogonal Compensation Method. 4.11 Synthesis of Conformal Arrays: The Conformal Orthogonal Method. References. 5 The Orama Computer Tool (George S. Miaris and John N. Sahalos). 5.1 Introduction. 5.2 Description of the ORAMA Program. 5.3 Element Types. 5.4 Design Examples. 5.5 Conclusion. References. Index. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> A. Antenna Configurations <s> Multi-user MIMO offers big advantages over conventional point-to-point MIMO: it works with cheap single-antenna terminals, a rich scattering environment is not required, and resource allocation is simplified because every active terminal utilizes all of the time-frequency bins. However, multi-user MIMO, as originally envisioned, with roughly equal numbers of service antennas and terminals and frequency-division duplex operation, is not a scalable technology. Massive MIMO (also known as large-scale antenna systems, very large MIMO, hyper MIMO, full-dimension MIMO, and ARGOS) makes a clean break with current practice through the use of a large excess of service antennas over active terminals and time-division duplex operation. Extra antennas help by focusing energy into ever smaller regions of space to bring huge improvements in throughput and radiated energy efficiency. Other benefits of massive MIMO include extensive use of inexpensive low-power components, reduced latency, simplification of the MAC layer, and robustness against intentional jamming. The anticipated throughput depends on the propagation environment providing asymptotically orthogonal channels to the terminals, but so far experiments have not disclosed any limitations in this regard. While massive MIMO renders many traditional research problems irrelevant, it uncovers entirely new problems that urgently need attention: the challenge of making many low-cost low-precision components that work effectively together, acquisition and synchronization for newly joined terminals, the exploitation of extra degrees of freedom provided by the excess of service antennas, reducing internal power consumption to achieve total energy efficiency reductions, and finding new deployment scenarios. This article presents an overview of the massive MIMO concept and contemporary research on the topic. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> A. Antenna Configurations <s> This book unites two different technologies: parasitic antenna arrays driven via analogue circuits that control the electromagnetic waves generated by the antenna array; and MIMO technology for multi-antenna arrays, typically driven by digital techniques in the baseband domain. The combination of these two technologies has revealed a novel functionality that breaks through the conventional MIMO paradigm, allowing MIMO transmission over the air with the use of antenna arrays that may consist of only a single active element, that is surrounded by a number of passive neighboring antennas. The contributions in the book show the capability of such systems to also perform MIMO transmission. This fact holds the potential of revolutionizing the way small-form wireless terminals operate and seems to set the scene for a win-win situation, achieving MIMO transmission with very small and cheap antenna arrays. The book is structured to provide a well-rounded treatment of the various facets of this newly discovered wireless communication capability. All relevant technical angles, ranging from information theoretic to electromagnetic considerations; from analogue circuit to digital baseband control for signal generation; and from channel modeling to communication theoretic aspects are taken into account. A good balance between theory, practical considerations and over-the-air experimentation is proposed and reflected in the chapter outline. Finally, a discussion and early evidence related to potential applications as well as the relevance to current and upcoming wireless standards is provided. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> A. Antenna Configurations <s> Sparse antenna arrays with irregularly spaced elements are increasingly being used in large array applications such as radio astronomy and radar. These antenna arrays may consist of hundreds of elements and therefore optimisation of these arrays becomes computationally expensive. This study proposes a novel solution to the problem of null synthesis in such arrays, where nulls are formed in prescribed locations for two-dimensional irregular sparse antenna arrays. The technique is based on a subarray and hybrid synthesis solution, where each subarray spans the entire antenna aperture. Single and multiple null formations with array gain maximisation are considered. The technique's effectiveness is demonstrated using evolutionary optimisation algorithms; however, the basic technique can be used with other optimisation methods. The simulation results presented compare the null synthesis performance with and without subarraying and it has shown that the use of subarray results in improved convergence compared with a direct full array synthesis. Furthermore the technique is appropriate for ready application to parallel computer architectures. <s> BIB004

In a traditional passive AA, the Radio-Frequency (RF) circuit is usually connected to its physical antennas through an RF cable. In order to reduce the loss imposed by the RF cable and to save the costs of installation and maintenance, a Remote Radio Unit (RRU) in conjunction with a Baseband Unit (BBU) has become a preferred configuration recently . The baseband digital signal generated by the BBU is sent to the RRUs through an optical fiber. The RF circuit is placed as close as possible to the physical AA. Furthermore, active AAs operating without RF cables are now also commercially available, which enabling engineers to carefully configure a LS AA . In an active AA, the RF circuit and the AA are integrated into a single circuit board, which is an important milestone in the development of AA. Fig. 2 illustrates several typical LS AAs, namely the linear AA, spherical AA, cylindrical AA, rectangular AA, and distributed AA BIB002 . The linear AA is an example of TwoDimensional (2D) AAs, whereas the spherical AA, cylindrical AA and rectangular AAs belong to the family of 3D AAs. Considering the space limitations at both the eNBs and UEs, the spherical, cylindrical and rectangular AAs are more realistic for practical systems. The distributed AA is mainly used either inside buildings or for outdoor cooperation, and the linear AA is mostly assumed in theoretical analysis and realistic measurements. Moreover, due to the associated aspects of aesthetics and potential health issues, commercial deployments of LS AAs have been partially opposed both by the public and by the organizations. By integrating the AEs into the environment, LS AAs can be rendered virtually invisible. An aesthetically pleasing method is to deploy LS AA as part of the building's facade or signage in an irregular fashion , e.g. the black AEs of a rectangular AA may assume the shape of the Chinese character "Zhong," as shown in Fig. 2 . On the other hand, in order to reduce the side lobes of the irregular AA, advanced algorithms relying on subarrays BIB004 , on orthogonal placement BIB001 , or on parasitic AAs BIB003 can be introduced for improving the beamforming performance of these irregular AAs.

Survey of Large-Scale MIMO Systems <s> B. Channel Measurements <s> Wireless communication using very-large multiple-input multiple-output (MIMO) antennas is a new research field, where base stations are equipped with a very large number of antennas as compared to previously considered systems. In theory, as the number of antennas increases, propagation properties that were random before start to become deterministic. Theoretical investigations with independent identically distributed (i.i.d.) complex Gaussian (Rayleigh fading) channels and unlimited number of antennas have been done, but in practice we need to know what benefits we can get from very large, but limited, number of antenna elements in realistic propagation environments. In this study we evaluate properties of measured residential-area channels, where the base station is equipped with 128 antenna ports. An important property to consider is the orthogonality between channels to different users, since this property tells us how advanced multi-user MIMO (MU-MIMO) pre-coding schemes we need in the downlink. We show that orthogonality improves with increasing number of antennas, but for two single-antenna users there is very little improvement beyond 20 antennas. We also evaluate sum-rate performance for two linear pre-coding schemes, zero-forcing (ZF) and minimum mean squared error (MMSE), as a function of the number of base station antennas. Already at 20 base station antennas these linear pre-coding schemes reach 98% of the optimal dirty-paper coding (DPC) capacity for the measured channels. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> B. Channel Measurements <s> Very large MIMO is a technique that potentially can offer large network capacities in multi-user scenarios where the users are equipped only with single antennas. In this paper we are investigating channel properties for a realistic, though somewhat extreme, outdoor base station scenario using a large array. We present measurement results using a 128 element linear array base station and 26 different user position in line-of-sight (LOS) and 10 different user position in non line-of-sight (NLOS). We analyze the Ricean K-factor, received power levels over the array, antenna correlation and eigenvalue distributions. We show that the statistical properties of the received signal vary significantly over the large array. Near field effects and the non-stationarities over the array help decorrelating the channel for different users, thereby providing a favorable channel conditions with stable channels and low interference for the considered single antenna users. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> B. Channel Measurements <s> Very-large multiple-input multiple-output (MIMO), also called massive MIMO, is a new technique that potentially can offer large network capacities in multi-user scenarios, where the base stations are equipped with a large number of antennas simultaneously serving multiple single-antenna users on the same frequency. We investigate channel behavior for a realistic outdoor base station scenario using large arrays. Specifically we compare dirty-paper coding (DPC) capacities and zero-forcing (ZF) sum-rates when using a physically large linear array and a compact cylindrical array, both having 128 antenna elements. As a base-line reference, we use the DPC capacity and ZF sum-rate for the ideal case with independent and identically distributed (i.i.d.) channel coefficients. The investigation shows that the measured channels, for both array types, often allow us to harvest most of the capacities/sum-rates achieved in the i.i.d. case, already at about 10 base station antennas per user. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> B. Channel Measurements <s> Equipping base stations (BSs) with very large antenna arrays is a promising way to increase the spectral and energy efficiency of mobile communication systems without the need for new cell sites. However, the prominently theoretical works on this topic are based on several crucial assumptions about the wireless channel which have not been sufficiently validated by measurements. In this paper, we report on an outdoor measurement campaign with a scalable virtual antenna array consisting of up to 112 elements. The large amount of acquired data allows us to study several important aspects of large-scale MIMO systems. For example, we partially confirm the theoretical results based on uncorrelated channels which predict that the channels at different positions become more and more orthogonal as the number of antennas grows. However, for the measured channels, the marginal gain of an additional antenna quickly diminishes. Nevertheless, our results indicate that most of the theoretical benefits of large-scale MIMO could be realized also over the measured channels. <s> BIB004

Realistic channel measurements have been carried out in BIB002 , BIB003 in an effort to identify the main characteristics of LS-MIMO channels. As shown in Table II , different antenna configurations may be considered under different scenarios at a carrier frequency of 2.6 GHz. The outdoor measurements in BIB002 focus mainly on the impact of the number of antennas imposed on the small-scale fading characteristics. When a linear AA is employed at the eNB, both the non-stationary nature of the fading and the near-field AA effects have been studied in order to capture the main properties of a realistic channel model BIB002 . However, it requires further investigations to ascertain whether these properties are valid for both spherical as well as cylindrical and rectangular arrays. The main results of these measurements may be summarized as follows: • Since different Antenna Elements (AEs) of the AA at the eNB may encounter different multi-path clusters and the AA is also often subjected to shadow fading, the accurate modeling of LS-MIMO systems in practical nonstationary propagation scenarios remains to a large extent an open challenge BIB002 ; • The Channel Impulse Responses (CIRs) experienced by UEs become more de-correlated from each other in the case of large AAs, because having more AEs allows one to more accurately distinguish both their CIRs and their angles of arrival BIB001 , BIB004 . In other words, having more AEs at the eNB is capable of achieving improved orthogonality amongst different UEs in comparison to their traditional small-scale MIMO counterparts. It is particularly important in Spatial Division Multiplexing (SDM) or Spatial Division Multiple Access (SDMA) systems, where the unique and user-specific CIRs are used for distinguishing the UEs and the transmission streams; and • The linear AA has the better angular resolution in azimuth than the cylindrical array. However, the latter is capable of achieving a beneficial resolution in both azimuth and elevation, which may be more useful in high-rise urban environments BIB003 .

Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> This paper presents a spatio-temporal model of multipath propagation channels based on a physically defined propagation process for testing and validation of the antenna array systems operating in multipath propagation environments. The idea behind the model is to generate a set of spatially distributed reflectors or scatterers which will act as sources for the signals arriving at the receiver in accordance with the distribution of excess delay times. The large scale and small scale signal fadings are incorporated into the model, which allows an easy and flexible simulation of multipath vector channels. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> Analysis and design of multielement antenna systems in mobile fading channels require a model for the space-time cross correlation among the links of the underlying multiple-input multiple-output (MIMO) channel. In this paper, we propose a general space-time cross-correlation function for mobile frequency nonselective Rice fading MIMO channels, in which various parameters of interest such as the angle spreads at the base station and the user, the distance between the base station and the user, mean directions of the signal arrivals, array configurations, and Doppler spread are all taken into account. The new space-time cross-correlation function includes all the relevant parameters of the MIMO fading channel in a clean compact form, suitable for both mathematical analysis and numerical calculations/simulations. It also covers many known correlation models as special cases. We demonstrate the utility of the new space-time correlation model by clarifying the limitations of a widely accepted correlation model for MIMO fading channels. As another application, we quantify the impact of nonisotropic scattering around the user, on the capacity of a MIMO fading channel. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> Wireless communication systems employing multiple antennas at both the transmitter and receiver have been shown to offer significant gains over single-antenna systems. Recent studies on the capacity of multiple-input-multiple-output (MIMO) channels have focused on the effect of spatial correlation. The joint effect of spatial and temporal correlation has not been well studied. In this paper, a geometric MIMO channel model is presented, which considers motion of the receiver and nonisotropic scattering at both ends of the radio link. A joint space-time cross-correlation function is derived from this model and variates with this joint correlation are generated by using the vector autoregressive stochastic model. The outage capacity of this channel is considered where the effects of antenna spacing, antenna array angle, degree of nonisotropic scattering, and receiver motion are investigated. When n transmit and n receive antennas are employed, it is shown that the outage capacity still increases linearly with respect to n, despite the presence of spatial and temporal correlation. Furthermore, analytical expressions are derived for the ergodic capacity of a MIMO channel for the cases of spatial correlation at one end and at both ends of the radio link. The latter case does not lend itself to numerical evaluation, but the former case is shown to be accurate by comparison with simulation results. The proposed analysis is very general, as it is based on the transmit and receive antenna correlations matrices. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> Very large MIMO is a technique that potentially can offer large network capacities in multi-user scenarios where the users are equipped only with single antennas. In this paper we are investigating channel properties for a realistic, though somewhat extreme, outdoor base station scenario using a large array. We present measurement results using a 128 element linear array base station and 26 different user position in line-of-sight (LOS) and 10 different user position in non line-of-sight (NLOS). We analyze the Ricean K-factor, received power levels over the array, antenna correlation and eigenvalue distributions. We show that the statistical properties of the received signal vary significantly over the large array. Near field effects and the non-stationarities over the array help decorrelating the channel for different users, thereby providing a favorable channel conditions with stable channels and low interference for the considered single antenna users. <s> BIB004 </s> Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> Very-large multiple-input multiple-output (MIMO), also called massive MIMO, is a new technique that potentially can offer large network capacities in multi-user scenarios, where the base stations are equipped with a large number of antennas simultaneously serving multiple single-antenna users on the same frequency. We investigate channel behavior for a realistic outdoor base station scenario using large arrays. Specifically we compare dirty-paper coding (DPC) capacities and zero-forcing (ZF) sum-rates when using a physically large linear array and a compact cylindrical array, both having 128 antenna elements. As a base-line reference, we use the DPC capacity and ZF sum-rate for the ideal case with independent and identically distributed (i.i.d.) channel coefficients. The investigation shows that the measured channels, for both array types, often allow us to harvest most of the capacities/sum-rates achieved in the i.i.d. case, already at about 10 base station antennas per user. <s> BIB005 </s> Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> Wireless communication using very-large multiple-input multiple-output (MIMO) antennas is a new research field, where base stations are equipped with a very large number of antennas as compared to previously considered systems. Unlike the conventional small and compact antenna arrays, the very-large arrays may span many wavelengths in space, thus they can experience large-scale fading over the array. This power variation over the antenna array may be critical to algorithm design and performance evaluations for very-large MIMO systems, and it is thus important that it is included in channel models. Based on channel measurements using a 128-antenna linear array in a semi-urban area, we characterize and model the large-scale fading properties. We focus on cluster-based modeling which can be seen as an extension of the current COST 2100 channel model. In the paper, we report our latest channel behavior modeling progress for very-large MIMO systems. <s> BIB006 </s> Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> This paper proposes a non-stationary wideband geometry-based stochastic model (GBSM) for multiple-input multiple-output (MIMO) high-speed train (HST) channels. The proposed model has the ability to investigate the non-stationarity of HST environment caused by the high speed movement of the receiver. Based on the proposed model, the space-time-frequency (STF) correlation function (CF) and STF local scattering function (LSF) are derived for different taps. Numerical results show the non-stationarity of the proposed channel model. <s> BIB007 </s> Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> In this contribution, the 3-dimensional (3D) channel characteristics, particularly in the elevation domains, are extracted through measurements in typical urban macro and micro environments in Xi'an China. Stochastic channel model parameters are obtained based on the high-resolution multi-path parameter estimates. In addition, a modified spatial channel model (SCM) for the line-of-sight (LoS) scenario is proposed where the LoS polarization matrix is parameterized in accordance with the reality. Measurement results justify the reasonability of the proposed model. These works significantly improve the applicability of the ITU SCM models in realistic 3D channel simulations. <s> BIB008 </s> Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> This article considers a practical implementation of massive MIMO systems [1]. Although the best performance can be achieved when a large number of active antennas are placed only in the horizontal domain, BS form factor limitation often makes horizontal array placement infeasible. To cope with this limitation, this article introduces full-dimension MIMO (FD-MIMO) cellular wireless communication system, where active antennas are placed in a 2D grid at BSs. For analysis of the FD-MIMO systems, a 3D spatial channel model is introduced, on which system-level simulations are conducted. The simulation results show that the proposed FD-MIMO system with 32 antenna ports achieves 2-3.6 times cell average throughput gain and 1.5-5 times cell edge throughput gain compared to the 4G LTE system of two antenna ports at the BS. <s> BIB009 </s> Survey of Large-Scale MIMO Systems <s> 2) GBSM: <s> This paper proposes a novel non-stationary wideband multi-confocal ellipse two dimensional (2-D) channel model for massive multiple-input multiple-output (MIMO) communication systems. Spherical wavefront is assumed in the proposed channel model, instead of the plane wavefront assumption used in conventional MIMO channel models. In addition, the birth-death process is incorporated into the proposed model to capture the dynamic properties of clusters on both the array and time axes. Statistical properties of the channel model such as the space-time-frequency correlation function and power imbalance on the antenna array are studied. The impact of the spherical wavefront assumption on the statistical properties of the channel model is investigated. Furthermore, numerical analysis shows that the proposed channel model is able to capture specific characteristics of massive MIMO channel as observed in measurements. <s> BIB010

According to the modeling of scattering propagation environments, the GBSM can be classified into singlering BIB002 , twin-ring BIB003 and elliptical models BIB001 . However, depending on whether the elevation angle is considered or not, the GBSM involved for an LS-MIMO system is mainly used in the context of elliptical models and can be divided into 2D and 3D channel models. When an AA is employed at the eNB, the angle of elevation is fixed and the 2D channel model is adequate for accurately evaluating the performance. However, if a practical spherical, cylindrical or rectangular AA is adopted at the eNB, the 3D channel model with an adjustable angle of elevation has to be considered. • 2D channel model: When adopting a linear AA, a nonstationary spherical wavefront has been observed by LS-MIMO channel measurements BIB004 , BIB005 . Similarly, a non-stationary wavefront has also been observed for a linear 128-element AA in a semi-urban area BIB006 . However, due to the fact that the non-stationary spherical wavefront critically affects both the receiver design and its performance, it is important to model this property of the LS-MIMO channel. So, the COST 2100 channel model of has been extended to include the effect of the wavefront's non-stationary BIB006 ; When considering both the non-stationary wavefront propagation phenomenon and the spherical nature of the waveform, an elliptical GBSM is proposed for LS-MIMO systems based on a linear AA in BIB010 , BIB007 . Also, the birthdeath process of clusters appearing and disappearing is discussed in BIB010 for the sake of characterizing its nonstationary nature. The results in BIB010 show that the phases of the linear AA responses are no longer linear, and the AoAs impinging on the AA gradually shift, in agreement with the measurement results in BIB004 ; and • 3D channel model: The model methodology and parameters of some existing models, such as the 3D channel model of the WINNER+ project , depend primarily on literatures, rather than on realistic measurements BIB008 . Thus, it needs further verification by realistic channel measurements, including the elevation characteristics of the AA as well as the cross-correlation matrix of the large-scale fading parameters. Therefore, the existing 3D channel models cannot be directly applied in realistic scenarios. In the WINNER II, WINNER+ and COST273 projects, the procedures of modeling the 3D MIMO channel have been proposed as detailed in Fig. 3 BIB009 , . The main parameters of the 3D channel model consist of the Shadow Fading (SF), the Delay Spread (DS), the Ricean K-factor, the AoA, Azimuth of Departure (AoD), the Elevation angle of Arrival (EoA) and the Elevation angle of Departure (EoD). The EoA and EoD are of particular interest during the modeling. Moreover, the cross correlation matrix of the channel has been extended from a 5-dimensional to a 7-dimensional matrix, and the correlations between any two parameters have also been characterized in detail. In 3GPP Release 12 (R12) , there are three scenarios at present, namely the urban micro cell associated with a high UE density, the urban macro-cell having a high UE density and the urban macro cell associated with one high-rise AA per sector and the Inter Site Distance (ISD) of 300 meter. The distribution of the related parameters and the cross correlation matrix of the 3D channel model associated with a rectangular AA have been measured, and documented in . According to the procedures of modeling the 3D MIMO channel , realistic channel coefficients can be generated for characterizing the abovementioned three scenarios. However, the non-stationary nature of LS-MIMO channels has not been taken into account in BIB005 , . Therefore, the properties and the modeling of the 3D MIMO channel requires further study.

Survey of Large-Scale MIMO Systems <s> 3) PSM: <s> Space-time receivers for wireless communication systems offer the possibility to have both Tx- and Rx-antennas. For a realistic simulation of such systems, a multiple input multiple output (MIMO) spatial channel model is required which reasonably characterizes the space- and time-variant effects of the mobile radio channel. This paper describes a space-time vector channel model with realistic fading simulation for different scenarios. Mutual correlation between the fading coefficients is considered. This allows an estimation of the diversity gain, that can be achieved with space-time receivers in different scenarios. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> 3) PSM: <s> We introduce the concept of the double-directional mobile radio channel. It is called this because it includes angular information at both link ends, e.g., at the base station and at the mobile station. We show that this angular information can be obtained with synchronized antenna arrays at both link ends. In wideband high-resolution measurements, we use a switched linear array at the receiver and a virtual-cross array at the transmitter. We evaluate the raw measurement data with a technique that alternately used estimation and beamforming, and that relied on ESPRIT (estimation of signal parameters via rotational invariance techniques) to obtain superresolution in both angular domains and in the delay domain. In sample microcellular scenarios (open and closed courtyard, line-of-sight and obstructed line-of-sight), up to 50 individual propagation paths are determined. The major multipath components are matched precisely to the physical environment by geometrical considerations. Up to three reflection/scattering points per propagation path are identified and localized, lending insight into the multipath spreading properties in a microcell. The extracted multipath parameters allow unambiguous scatterer identification and channel characterization, independently of a specific antenna, its configuration (single/array), and its pattern. The measurement results demonstrate a considerable amount of power being carried via multiply reflected components, thus suggesting revisiting the popular single-bounce propagation models. It turns out that the wideband double-directional evaluation is a most complete method for separating multipath components. Due to its excellent spatial resolution, the double-directional concept provides accurate estimates of the channel's multipath-richness, which is the important parameter for the capacity of multiple-input multiple-output (MIMO) channels. <s> BIB002

The PSM describes the signals impinging on the receivers as a superposition of waves. A common form of these models employs the structure of a tapped delay line, where each tap reflects a specific propagation path. There are some PSMs for traditional MIMO, such as the Double Directional Channel Model (DDCM) of BIB002 and the Virtual Ray Model (VRM) of BIB001 , where different methods are used for modeling each tap of the channel. But again, the PSMs are less well studied in the context of LS-MIMOs since they are more complex in terms of their theoretical analysis than the CBSMs. Nevertheless, they constitute viable design alternatives, which are capable of reducing the complexity of the GBSMs for LS-MIMOs. Therefore, the development of the PSM for LS-MIMO systems requires further research efforts, before it becomes a reality.

Survey of Large-Scale MIMO Systems <s> 3) Case 1C-Large-scale cooperation: <s> The idea of implementing an indoor radio communications system serving an entire building from a single central antenna appears to be an attractive proposition. However, based on various indoor propagation measurements of the signal attenuation and the multipath delay spread, such a centralized approach appears to be limited to small buildings and to narrow-band FDMA-type systems with limited reliability and flexibility. In this paper, we present the results of indoor radio propagation measurements of two signal distribution approaches that improve the picture dramatically. In the first, the building is divided into many small cells, each served from an antenna located in its own center, and with adjacent cells operating in different frequency bands. In the second approach, the building is divided into one or more large cells, each served from a distributed antenna system or a "leaky feeder" that winds its way through the hallways. This approach eliminates the frequency cell handoff problem that is bound to exist in the first approach, while still preserving the dramatic reductions in multipath delay spread and signal attenuation compared to a centralized system. For example, the measurements show that, with either approach, the signal attenuation can be reduced by as much as a few tens of decibels and the rms delay spread becomes limited to 20 to 50 us, even in large buildings. This can make possible the implementation of sophisticated broad-band TDMA-type systems that are flexible, robust, and virtually building-independent. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> 3) Case 1C-Large-scale cooperation: <s> Distributed antenna systems (DAS) augment the base station's transmit capability by adding multiple remote radio units, connected to the base station via a high bandwidth and low latency link. With DAS, the base station operates as if it had multiple antennas, but the antennas happen to be in different geographic locations. DAS have been shown to enhance coverage and capacity in cellular systems, in a variety of different configurations. This paper proposes, analyzes, and compares several downlink multiuser multiple input multiple output (MIMO) DAS strategies in terms of per-user throughput and area spectral efficiency. Zero-forcing transmit beamforming is used for transmission, the remote radio units may have one or more antennas, and the subscriber has a single receive antenna. Techniques considered include beamforming across all remote radio units (full transmission), using the same beamforming vector for each remote radio unit (simplified transmission), and selecting a subset of remote radio units. To facilitate rapid simulation and design space exploration, approximations of the ergodic rate are proposed for each technique assuming path-loss, small-scale Rayleigh fading, and out-of-cell interference. Simulations accounting for multiple interfering cells are used to compare the different transmission techniques. Full transmission is found to have the best performance even accounting for out-of-cell interference, though gains diminish for higher numbers of active users. Simplified transmission improves over no DAS but performance degrades with more active remote radio units. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> 3) Case 1C-Large-scale cooperation: <s> Massive multiple-input multiple-output (MIMO) is a breakthrough communication technique for providing high spectral efficiency. The idea is to deploy a very large number of antennas at each base station and to use multiuser MIMO transmission to serve a smaller number of users. In this paper, the viability of using distributed antennas for massive MIMO on the uplink is investigated for a particular spatial correlation channel model. Both maximal ratio combining (MRC) and minimum mean squared error (MMSE) beamforming are found to provide higher performance in several distributed settings. An algorithm for remote radio head selection is proposed that allows MRC to approach the performance achieved by the MMSE beamforming solution yet retaining its lower complexity. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> 3) Case 1C-Large-scale cooperation: <s> Massive MIMO has been identified as one of the promising disruptive air interface technologies to address the massive capacity requirement demanded by 5G wireless communications. Most of the existing works showed different benefits of massive MIMO system but for the co-located deployment scenario. In this paper, we are interested to find out extra benefits that may be brought by distributing the massive number of antennas to different levels. Specifically, we compare the performance of co-located and distributed deployment scenarios in the framework of area spectrum efficiency (ASE) and area energy efficiency (AEE) tradeoff. Closed-form expressions are derived to help discover important design insights. For instance, it is interesting to see the distribution of the massive antennas always helps to improve ASE, but not the case for AEE. Moreover, with the large number of transmit antennas available and the practical cost of channel estimation considered, collaboration between cells is not always welcomed in each deployment scenario and the positive use cases depend on system parameters such as the number of antennas and the way they are distributed. <s> BIB004 </s> Survey of Large-Scale MIMO Systems <s> 3) Case 1C-Large-scale cooperation: <s> Despite the common belief that substantial capacity gains can be achieved by using more antennas at the base-station (BS) side in cellular networks, the effect of BS antenna topology on the capacity scaling behavior is little understood. In this paper, we present a comparative study on the ergodic capacity of a downlink single-user multiple-input-multiple-output (MIMO) system where BS antennas are either co-located at the center or grouped into uniformly distributed antenna clusters in a circular cell. By assuming that the number of BS antennas and the number of user antennas go to infinity with a fixed ratio $L\gg 1$, the asymptotic analysis reveals that the average per-antenna capacities in both cases logarithmically increase with $L$, but in the orders of $\log_2 L$ and $\tfrac{\alpha}{2}\log_2 L$, for the co-located and distributed BS antenna layouts, respectively, where $\alpha>2$ denotes the path-loss factor. The analysis is further extended to the multi-user case where a 1-tier (7-cell) MIMO cellular network with $K\gg 1$ uniformly distributed users in each cell is considered. By assuming that the number of BS antennas and the number of user antennas go to infinity with a fixed ratio $L\gg K$, an asymptotic analysis is presented on the downlink rate performance with block diagonalization (BD) adopted at each BS. It is shown that the average per-antenna rates with the co-located and distributed BS antenna layouts scale in the orders of $\log_2 \tfrac{L}{K}$ and $\log_2 \frac{(L-K+1)^{\alpha/2}}{K}$, respectively. The rate performance of MIMO cellular networks with small cells is also discussed, which highlights the importance of employing a large number of distributed BS antennas for the next-generation cellular networks. <s> BIB005

Most of the existing contributions on LS-MIMOs show different benefits in a colocated deployment scenario, where there is a large number of antennas installed at a single cell site. However, such co-located deployments impose challenges both on their hardware design and on their field deployment. On the other hand, Distributed Antenna Systems (DASs) associated with spatially separated antennas have been conceived for improving the indoor coverage using a moderate number of antennas BIB001 . Recent studies have shown that apart from its improved coverage, a DAS is capable of significantly increasing the network's BE, even in the presence of ICI BIB002 . This motivates researchers to identify specific scenarios as illustrated in Fig. 6 , where the LS-MIMO system associated with a distributed architecture outperforms the one relying on a co-located deployment BIB004 , BIB003 . The advantage of distributed LS-MIMOs is plausible, because the signals arriving from the distributed antennas to each UE are subject to independent random levels of large-scale fading, thereby leading to potential capacity gains over their co-located counterpart BIB005 . However, it may be a challenge to achieve these gains by coordinating the intra-cell interferences, especially in scenarios having dozens or even hundreds of RRUs in a cell. Although full cooperation constitutes an efficient method of eliminating the intra-cell interference, it is not practical due to its high reliance on full CSI sharing. To strike an elegant trade-off between the performance attained and the overhead imposed, efficient large-scale cooperation schemes are of high importance under this scenario. Moreover, distributed LS-MIMO and small cell deployments may be viewed as being complementary rather than competitive. For example, a cooperative cellular architecture composed of a DAS and a femtocell-macrocell underlay system is proposed in BIB005 , which may be extended to operate in conjunction with distributed LS-MIMOs.

Survey of Large-Scale MIMO Systems <s> A. Performance of Precoders/Detectors in Single-Cell Environments 1) Theoretical Performance: <s> Presents a list of articles published by the IEEE Signal Processing Society (SPS) that ranked among the top 100 most downloaded IEEE Xplore articles. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> A. Performance of Precoders/Detectors in Single-Cell Environments 1) Theoretical Performance: <s> A multiplicity of autonomous terminals simultaneously transmits data streams to a compact array of antennas. The array uses imperfect channel-state information derived from transmitted pilots to extract the individual data streams. The power radiated by the terminals can be made inversely proportional to the square-root of the number of base station antennas with no reduction in performance. In contrast if perfect channel-state information were available the power could be made inversely proportional to the number of antennas. Lower capacity bounds for maximum-ratio combining (MRC), zero-forcing (ZF) and minimum mean-square error (MMSE) detection are derived. An MRC receiver normally performs worse than ZF and MMSE. However as power levels are reduced, the cross-talk introduced by the inferior maximum-ratio receiver eventually falls below the noise level and this simple receiver becomes a viable option. The tradeoff between the energy efficiency (as measured in bits/J) and spectral efficiency (as measured in bits/channel use/terminal) is quantified for a channel model that includes small-scale fading but not large-scale fading. It is shown that the use of moderately large antenna arrays can improve the spectral and energy efficiency with orders of magnitude compared to a single-antenna system. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> A. Performance of Precoders/Detectors in Single-Cell Environments 1) Theoretical Performance: <s> Large-Scale Antenna Systems (LSAS) is a form of multi-user MIMO technology in which unprecedented numbers of antennas serve a significantly smaller number of autonomous terminals. We compare the two most prominent linear pre-coders, conjugate beamforming and zero-forcing, with respect to net spectral-efficiency and radiated energy-efficiency in a simplified single-cell scenario where propagation is governed by independent Rayleigh fading, and where channel-state information (CSI) acquisition and data transmission are both performed during a short coherence interval. An effective-noise analysis of the pre-coded forward channel yields explicit lower bounds on net capacity which account for CSI acquisition overhead and errors as well as the sub-optimality of the pre-coders. In turn the bounds generate trade-off curves between radiated energy-efficiency and net spectral-efficiency. For high spectral-efficiency and low energy-efficiency zero-forcing outperforms conjugate beamforming, while at low spectral-efficiency and high energy-efficiency the opposite holds. Surprisingly, in an optimized system, the total LSAS-critical computational burden of conjugate beamforming may be greater than that of zero-forcing. Conjugate beamforming may still be preferable to zero-forcing because of its greater robustness, and because conjugate beamforming lends itself to a de-centralized architecture and de-centralized signal processing. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> A. Performance of Precoders/Detectors in Single-Cell Environments 1) Theoretical Performance: <s> We explore the performance of multiple input multiple output (MIMO) transmitters in correlated channels where increasing numbers of antenna elements are fitted in a fixed physical space. As well investigated in the literature, two main effects emerge in such a design: transmit spatial correlation and mutual antenna coupling. In contrast to the literature however, here we investigate the combined effect of reducing the distance between the antenna elements with increasing the number of elements in a fixed transmitter space. In other words, towards the implementation of large-scale MIMO transmitters in limited physical spaces, we investigate the joint effect of two contradicting phenomena: the reduction of spatial diversity due to reducing the separation between antennas and the increase in transmit diversity by increasing the number of elements. Within this context, we analytically approximate the performance of two distinct linear precoding designs. The theoretical analysis and simulations show the somewhat surprising result that for a given number of receivers the improved transmit diversity dominates the performance of practical linear precoders. Consequently, important benefits in the system sum rate can be gleaned by fitting more antenna elements in a fixed space by employing separations smaller than the wavelength of the transmit frequency. <s> BIB004

Employing the CBSM, including both the i.i.d. Rayleigh channel model and correlated Rayleigh channel model, the performance of different TPCs or detectors designed for LS-MIMO has been widely analyzed in single-cell environments BIB001 , BIB002 , BIB003 , BIB004 , etc. (

Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> Dirty-paper coding makes an analogy to the problem of writing on dirty paper, where the reader cannot nominally distinguish dirt from ink. There are many scenarios where this result may be applied. In the field of information hiding (or watermarking), theoretical bounds as well as practical watermarking schemes have been found. Another important application of dirty-paper coding is for a multiuser channel wherein a multiple-antenna transmitter is communicating with multiple users. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> Recent theoretical results describing the sum-capacity when using multiple antennas to communicate with multiple users in a known rich scattering environment have not yet been followed with practical transmission schemes that achieve this capacity. We introduce a simple encoding algorithm that achieves near-capacity at sum-rates of tens of bits/channel use. The algorithm is a variation on channel inversion that regularizes the inverse and uses a "sphere encoder" to perturb the data to reduce the energy of the transmitted signal. The paper is comprised of two parts. In this second part, we show that, after the regularization of the channel inverse introduced in the first part, a certain perturbation of the data using a "sphere encoder" can be chosen to further reduce the energy of the transmitted signal. The performance difference with and without this perturbation is shown to be dramatic. With the perturbation, we achieve excellent performance at all signal-to-noise ratios. The results of both uncoded and turbo-coded simulations are presented. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> Recent theoretical results describing the sum capacity when using multiple antennas to communicate with multiple users in a known rich scattering environment have not yet been followed with practical transmission schemes that achieve this capacity. We introduce a simple encoding algorithm that achieves near-capacity at sum rates of tens of bits/channel use. The algorithm is a variation on channel inversion that regularizes the inverse and uses a "sphere encoder" to perturb the data to reduce the power of the transmitted signal. This work is comprised of two parts. In this first part, we show that while the sum capacity grows linearly with the minimum of the number of antennas and users, the sum rate of channel inversion does not. This poor performance is due to the large spread in the singular values of the channel matrix. We introduce regularization to improve the condition of the inverse and maximize the signal-to-interference-plus-noise ratio at the receivers. Regularization enables linear growth and works especially well at low signal-to-noise ratios (SNRs), but as we show in the second part, an additional step is needed to achieve near-capacity performance at all SNRs. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> This paper studies the problem of signal detection for multiple-input multiple-output (MIMO) channels with large signal dimensions. We propose a block-iterative generalized decision feedback equalization (BI-GDFE) receiver to recover the transmitted symbols in a block-iterative manner. By exploiting the input-decision correlation, a measure for the reliability of the earlier-made decisions, we design the feed-forward equalizers (FFEs) and feedback equalizers (FBEs) in such a way that maximized signal-to-interference-plus-noise ratio (SINR) is achieved for each of the iterations. Novel implementations are also introduced to simplify the complexity of the receiver, which requires only one-tap filters for FFE and FBE. The proposed receiver also works when the signal dimension is greater than the observation dimension. The asymptotic performance of the proposed receiver is analyzed and its convergence has been confirmed through numerical evaluations for various parameters. Computer simulations are presented to illustrate the capability of the proposed receiver to achieve single user matched-filter bound (MFB) for large random MIMO channels when the received SNR is high enough. <s> BIB004 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> In this paper, tabu search (TS) is applied to the detection algorithm of MIMO system, which is originally designed to find the approximate solution of combinatorial optimization problems. Resorting to an original two embedded cubes stereograph, we give a simple illustration of TS process. As a simple add-on to traditional suboptimal detectors, TS provides the potential of moving toward the best ML solution. The method of skipping out of the search under a certain condition in advance is proposed, leading to the flexible tradeoff between the performance and the complexity. Simulation results show that the advantage of TS detection on performance comparing with the traditional detection is great, especially when the system has more transmit-antennas. <s> BIB005 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> A minimum mean-square error (MMSE)-based iterative soft interference cancellation (MMSE-SIC) receiver has been proposed to mitigate the interferences of the multiple-input multiple-output (MIMO) channels, with reduced complexity as compared to maximum-likelihood (ML) detection. On the other hand, the block-iterative generalized decision-feedback equalizer (BI-GDFE) attains close to the performance of the MMSE-SIC receivers with further reduced complexity. The BI-GDFE, however, requires an accurate estimate of the input-decision correlation (IDC), which is a statistical reliability metric of earlier-made decisions. To date, the BI-GDFE receiver is applicable only to phase-shift-keying (PSK) modulations due to the absence of a method to estimate the IDC for higher order quadrature amplitude modulations (QAMs). In this paper, we establish the relationship between the MMSE-SIC and BI-GDFE receivers and propose an algorithm to determine the IDC for BI-GDFE from the unconditional MMSE-SIC (U-MMSE-SIC). We further analyze and compare the asymptotic performances of the two receivers for large random MIMO channels and prove that for the limiting case, the output signal-to-interference-plus-noise ratios (SINRs) at each iteration for both receivers converge in probability to their respective deterministic limits. Our simulation results have shown that the bit error rate (BER) performance of the BI-GDFE receiver with the proposed IDC selection method achieves close to that of the U-MMSE-SIC receiver with similar convergence behavior and reaches the single-user matched filter bound (MFB) with several iterations for high enough signal-to-noise ratio (SNR). <s> BIB006 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> In this paper, we study the sum rate performance of zero-forcing (ZF) and regularized ZF (RZF) precoding in large MISO broadcast systems under the assumptions of imperfect channel state information at the transmitter and per-user channel transmit correlation. Our analysis assumes that the number of transmit antennas $M$ and the number of single-antenna users $K$ are large while their ratio remains bounded. We derive deterministic approximations of the empirical signal-to-interference plus noise ratio (SINR) at the receivers, which are tight as $M,K\to\infty$. In the course of this derivation, the per-user channel correlation model requires the development of a novel deterministic equivalent of the empirical Stieltjes transform of large dimensional random matrices with generalized variance profile. The deterministic SINR approximations enable us to solve various practical optimization problems. Under sum rate maximization, we derive (i) for RZF the optimal regularization parameter, (ii) for ZF the optimal number of users, (iii) for ZF and RZF the optimal power allocation scheme and (iv) the optimal amount of feedback in large FDD/TDD multi-user systems. Numerical simulations suggest that the deterministic approximations are accurate even for small $M,K$. <s> BIB007 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> A simple linear precoding technique is proposed for multiple input multiple output (MIMO) broadcast systems using phase shift keying (PSK) modulation. The proposed technique is based on the fact that, on an instantaneous basis, the interference between spatial links in a MIMO system can be constructive and can contribute to the power of the useful signal to improve the performance of signal detection. In MIMO downlinks this co-channel interference (CCI) can be predicted and characterised prior to transmission. Contrary to common practice where knowledge of the interference is used to eliminate it, the main idea proposed here is to use this knowledge to influence the interference and benefit from it, thus gaining advantage from energy already existing in the communication system that is left unexploited otherwise. The proposed precoding aims at adaptively rotating, rather than zeroing, the correlation between the MIMO substreams depending on the transmitted data, so that the signal of interfering transmissions is aligned to the signal of interest at each receive antenna. By doing so, the CCI is always kept constructive and the received signal to interference-plus-noise ratio (SINR) delivered to the mobile units (MUs) is enhanced without the need to invest additional signal power per transmitted symbol at the MIMO base station (BS). It is shown by means of theoretical analysis and simulations that the proposed MIMO precoding technique offers significant performance and throughput gains compared to its conventional counterparts. <s> BIB008 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> Wireless communication using very-large multiple-input multiple-output (MIMO) antennas is a new research field, where base stations are equipped with a very large number of antennas as compared to previously considered systems. In theory, as the number of antennas increases, propagation properties that were random before start to become deterministic. Theoretical investigations with independent identically distributed (i.i.d.) complex Gaussian (Rayleigh fading) channels and unlimited number of antennas have been done, but in practice we need to know what benefits we can get from very large, but limited, number of antenna elements in realistic propagation environments. In this study we evaluate properties of measured residential-area channels, where the base station is equipped with 128 antenna ports. An important property to consider is the orthogonality between channels to different users, since this property tells us how advanced multi-user MIMO (MU-MIMO) pre-coding schemes we need in the downlink. We show that orthogonality improves with increasing number of antennas, but for two single-antenna users there is very little improvement beyond 20 antennas. We also evaluate sum-rate performance for two linear pre-coding schemes, zero-forcing (ZF) and minimum mean squared error (MMSE), as a function of the number of base station antennas. Already at 20 base station antennas these linear pre-coding schemes reach 98% of the optimal dirty-paper coding (DPC) capacity for the measured channels. <s> BIB009 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> Multi-user multiple-input multiple-output theory predicts manyfold capacity gains by leveraging many antennas on wireless base stations to serve multiple clients simultaneously through multi-user beamforming (MUBF). However, realizing a base station with a large number antennas is non-trivial, and has yet to be achieved in the real-world. We present the design, realization, and evaluation of Argos, the first reported base station architecture that is capable of serving many terminals simultaneously through MUBF with a large number of antennas (M >> 10). Designed for extreme flexibility and scalability, Argos exploits hierarchical and modular design principles, properly partitions baseband processing, and holistically considers real-time requirements of MUBF. Argos employs a novel, completely distributed, beamforming technique, as well as an internal calibration procedure to enable implicit beamforming with channel estimation cost independent of the number of base station antennas. We report an Argos prototype with 64 antennas and capable of serving 15 clients simultaneously. We experimentally demonstrate that by scaling from 1 to 64 antennas the prototype can achieve up to 6.7 fold capacity gains while using a mere 1/64th of the transmission power. <s> BIB010 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> Equipping base stations (BSs) with very large antenna arrays is a promising way to increase the spectral and energy efficiency of mobile communication systems without the need for new cell sites. However, the prominently theoretical works on this topic are based on several crucial assumptions about the wireless channel which have not been sufficiently validated by measurements. In this paper, we report on an outdoor measurement campaign with a scalable virtual antenna array consisting of up to 112 elements. The large amount of acquired data allows us to study several important aspects of large-scale MIMO systems. For example, we partially confirm the theoretical results based on uncorrelated channels which predict that the channels at different positions become more and more orthogonal as the number of antennas grows. However, for the measured channels, the marginal gain of an additional antenna quickly diminishes. Nevertheless, our results indicate that most of the theoretical benefits of large-scale MIMO could be realized also over the measured channels. <s> BIB011 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> Very-large multiple-input multiple-output (MIMO), also called massive MIMO, is a new technique that potentially can offer large network capacities in multi-user scenarios, where the base stations are equipped with a large number of antennas simultaneously serving multiple single-antenna users on the same frequency. We investigate channel behavior for a realistic outdoor base station scenario using large arrays. Specifically we compare dirty-paper coding (DPC) capacities and zero-forcing (ZF) sum-rates when using a physically large linear array and a compact cylindrical array, both having 128 antenna elements. As a base-line reference, we use the DPC capacity and ZF sum-rate for the ideal case with independent and identically distributed (i.i.d.) channel coefficients. The investigation shows that the measured channels, for both array types, often allow us to harvest most of the capacities/sum-rates achieved in the i.i.d. case, already at about 10 base station antennas per user. <s> BIB012 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> A multiplicity of autonomous terminals simultaneously transmits data streams to a compact array of antennas. The array uses imperfect channel-state information derived from transmitted pilots to extract the individual data streams. The power radiated by the terminals can be made inversely proportional to the square-root of the number of base station antennas with no reduction in performance. In contrast if perfect channel-state information were available the power could be made inversely proportional to the number of antennas. Lower capacity bounds for maximum-ratio combining (MRC), zero-forcing (ZF) and minimum mean-square error (MMSE) detection are derived. An MRC receiver normally performs worse than ZF and MMSE. However as power levels are reduced, the cross-talk introduced by the inferior maximum-ratio receiver eventually falls below the noise level and this simple receiver becomes a viable option. The tradeoff between the energy efficiency (as measured in bits/J) and spectral efficiency (as measured in bits/channel use/terminal) is quantified for a channel model that includes small-scale fading but not large-scale fading. It is shown that the use of moderately large antenna arrays can improve the spectral and energy efficiency with orders of magnitude compared to a single-antenna system. <s> BIB013 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> Large-Scale Antenna Systems (LSAS) is a form of multi-user MIMO technology in which unprecedented numbers of antennas serve a significantly smaller number of autonomous terminals. We compare the two most prominent linear pre-coders, conjugate beamforming and zero-forcing, with respect to net spectral-efficiency and radiated energy-efficiency in a simplified single-cell scenario where propagation is governed by independent Rayleigh fading, and where channel-state information (CSI) acquisition and data transmission are both performed during a short coherence interval. An effective-noise analysis of the pre-coded forward channel yields explicit lower bounds on net capacity which account for CSI acquisition overhead and errors as well as the sub-optimality of the pre-coders. In turn the bounds generate trade-off curves between radiated energy-efficiency and net spectral-efficiency. For high spectral-efficiency and low energy-efficiency zero-forcing outperforms conjugate beamforming, while at low spectral-efficiency and high energy-efficiency the opposite holds. Surprisingly, in an optimized system, the total LSAS-critical computational burden of conjugate beamforming may be greater than that of zero-forcing. Conjugate beamforming may still be preferable to zero-forcing because of its greater robustness, and because conjugate beamforming lends itself to a de-centralized architecture and de-centralized signal processing. <s> BIB014 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> Presents a list of articles published by the IEEE Signal Processing Society (SPS) that ranked among the top 100 most downloaded IEEE Xplore articles. <s> BIB015 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> We consider the multi-user MIMO broadcast channel with M single-antenna users and N transmit antennas under the constraint that each antenna emits signals having constant envelope (CE). The motivation for this is that CE signals facilitate the use of power-efficient RF power amplifiers. Analytical and numerical results show that, under certain mild conditions on the channel gains, for a fixed M, an array gain is achievable even under the stringent per-antenna CE constraint. Essentially, for a fixed M, at sufficiently large N the total transmitted power can be reduced with increasing N while maintaining a fixed information rate to each user. Simulations for the i.i.d. Rayleigh fading channel show that the total transmit power can be reduced linearly with increasing N (i.e., an O(N) array gain). We also propose a precoding scheme which finds near-optimal CE signals to be transmitted, and has O(MN) complexity. Also, in terms of the total transmit power required to achieve a fixed desired information sum-rate, despite the stringent per-antenna CE constraint, the proposed CE precoding scheme performs close to the sum-capacity achieving scheme for an average-only total transmit power constrained channel. <s> BIB016 </s> Survey of Large-Scale MIMO Systems <s> a) i.i.d. Rayleigh channel: <s> We explore the performance of multiple input multiple output (MIMO) transmitters in correlated channels where increasing numbers of antenna elements are fitted in a fixed physical space. As well investigated in the literature, two main effects emerge in such a design: transmit spatial correlation and mutual antenna coupling. In contrast to the literature however, here we investigate the combined effect of reducing the distance between the antenna elements with increasing the number of elements in a fixed transmitter space. In other words, towards the implementation of large-scale MIMO transmitters in limited physical spaces, we investigate the joint effect of two contradicting phenomena: the reduction of spatial diversity due to reducing the separation between antennas and the increase in transmit diversity by increasing the number of elements. Within this context, we analytically approximate the performance of two distinct linear precoding designs. The theoretical analysis and simulations show the somewhat surprising result that for a given number of receivers the improved transmit diversity dominates the performance of practical linear precoders. Consequently, important benefits in the system sum rate can be gleaned by fitting more antenna elements in a fixed space by employing separations smaller than the wavelength of the transmit frequency. <s> BIB017

• Linear precoder/detector: When employing a linear detector, the BE performance lower bounds of the Uplink (UL) of LS-MIMO systems have been studied both with perfect CSI and with the aid of realistically estimated CSI BIB013 . A Maximum Ratio Combining (MRC) receiver performs worse than its Zero-Forcing (ZF) and Minimum Mean Square Error (MMSE) criterion based counterparts in the high Signal-to-Noise Ratio (SNR) region BIB013 . However, when the SNR is low, the MRC detector outperforms the ZF and MMSE detectors, since the IUI imposed by the MRC detector falls below the noise level BIB013 . Moreover, the radiated power of the UE can be made inversely proportional to the number of AEs at the eNB under the idealized conditions of having perfect CSI. However, in the presence of realistically estimated CSI, it is only inversely proportional to the square-root of the number of AEs BIB013 . The relationship between the radiated EE and BE has also been investigated. Upon increasing the BE, the EE of MRC is initially better, but beyond a crossover point it becomes worse than that of ZF BIB013 . Similar to BIB013 , the BE and radiated EE recorded for the DL of LS-MIMO systems are also evaluated by considering both the realistically estimated CSI and the CSI overhead imposed by the pilots BIB014 . The lower bounds of BE and the optimal number of UEs supported are studied for both the Maximum Ratio Transmission (MRT) based TPC as well as for ZF TPC. Based on the lower bound, the radiated EE as a function of the BE is quantified BIB014 , which is found to be a monotonically decreasing function of the BE for both detectors. Upon increasing the BE, the EE of MRT is initially better than that of ZF, and then the opposite trend holds. In the high-EE regime, the total computational complexity of the MRT TPC may be higher than that of the ZF TPC, since its optimal number of UEs is higher. However, it may still be preferable to ZF, since MRT TPCs can be realized with the aid of a de-centralized architecture BIB010 . • Non-linear precoder: Apart from the above-mentioned linear TPCs, non-linear TPCs, such as the DPC BIB001 , the Vector Perturbation (VP) BIB002 and the lattice-aided TPC methods are also investigated and compared to linear precoders for transmission over the i.i.d. Rayleigh channel BIB015 . The performance of the ZF TPC is shown to approach that of the DPC TPC, when the number of AEs increases. Similarly, diverse non-linear detectors, such as the MMSE based Soft Interference Cancellation (MMSE-SIC) scheme BIB006 , the Block-Iterative Generalized Decision Feedback Equalizer (BI-GDFE) BIB004 , Tabu Search (TS) BIB005 and the Maximum Posterior Probability (MAP) algorithm are also evaluated in terms of the Bit Error Ratio (BER) versus complexity BIB015 . Moreover, the so-called per-antenna Constant Envelope Precoding (CEP) technique implemented with the aid of a sub-optimal algorithm is also invoked for the DL of LS-MIMO systems in an attempt to improve the efficiency of their Power Amplifiers (PAs) BIB016 . It is stated in BIB016 that an eNB relying on CEP needs about 4 dB lower transmit power than a conventional technique having a high Peak to Average Power Ratio (PAPR), which is an explicit benefit of using an efficient PA. (b) Correlated Rayleigh channel: When considering both a realistic Doppler-induced correlation and imperfect CSI, the sum rates of the ZF and Regularized ZF (RZF) BIB003 TPCs were studied in the DL of an LS-MIMO system in BIB007 . Since the sum rate of ZF precoding first increases and then decreases with the number of UEs, hence the optimal number of UEs is obtained in BIB007 . Similarly, the optimal parameters of the RZF TPC may be determined by finding the optimal sum rate. Moreover, the optimal power allocation scheme of both ZF and RZF TPCs, as well as the optimal amount of feedback required in Frequency Division Duplex (FDD)/Time Division Duplex (TDD) systems are also determined in BIB007 for LS-MIMO systems. Both the ZF and Correlation Rotation (CR) aided TPC techniques in BIB008 are evaluated in terms of both the BE and the Symbol Error Ratio (SER), considering the effects of both transmit correlation as well as mutual coupling BIB017 , BIB008 . According to the approximate lower bounds and to the simulation results characterizing both TPCs, CR-aided precoding outperforms ZF precoding. Upon increasing the number of AEs within a given antenna dimension, the increased transmit diversity achieved dominates the attainable performance, rather than the reduction in spatial diversity imposed by shrinking the spacing among the antennas. 2) Measurement Performances: All the above-mentioned results are based on the use of theoretical channel models, such as the i.i.d. Rayleigh and correlated channel models. However, there exist some differences between the theoretical channel models and their practical counterparts. Therefore, the performances of different TPCs are also measured in practice. • Effects of correlation BIB009 , BIB011 : Both the theoretical and realistic sum-rates versus the SNRs of the DPC, RZF and MRT precoders are measured, when the eNB employs a 112-element AA in an outdoor scenario BIB011 . The theoretical performance bound can be derived, despite significant differences between the i.i.d. Rayleigh channel model and the realistic channel. The correlation amongst the CIRs of the different UEs decreases upon increasing the number of AEs employed at the eNB, because larger AAs are capable of resolving smaller AoA and CIR differences amongst their channels. Furthermore, when encountering different channel correlations amongst the UE channels, the measured and theoretical sum-rates of the DPC, ZF and RZF TPCs are compared in conjunction with an eNB equipped with a 128-antenna indoor cylindrical array operated in residential areas BIB009 . Upon increasing the number of antennas at the eNB, the channel correlation decreases, and the measured sum-rates approach their theoretical limits. When the eNB employs 20 antennas, about 98% of the sum-rate of the ideal DPC scheme is achieved for a pair of single-antenna aided UEs by the ZF or RZF TPCs BIB009 . • Effects of the propagation environment BIB012 : Considering realistic environments, both a 128-antenna cylindrical and a linear array are employed at the eNB BIB012 . Then, their realistic sum-rates are compared for both DPC and ZF TPC to the theoretical sum-rates in the i.i.d. Rayleigh channel BIB012 . Even for the worst combination of the cylindrical array and a dense population supported in a Line of Sight (LoS) environment, ZF precoding is capable of achieving about 55% of the DPC scheme's sum-rate in the i.i.d. Rayleigh channel, when the number of antennas exceeded 40. By contrast, in a Non Line of Sight (NLoS) environment, despite encountering a dense user population, both cylindrical and linear arrays relying on ZF TPC are capable of attaining about 80-90% of the sum-rate of the DPC scheme. Regardless of the propagation environment, most of the theoretical sum-rate of LS-MIMO is achievable by linear precoding, if the eNB employs a sufficiently high number of antennas BIB012 . • Real time operation BIB010 : To implement real-time MRT based TPC, de-centralized MRT precoding weights can be locally calculated at each antenna BIB010 . A prototype using a 64-antenna rectangular array is employed in BIB010 for simultaneously serving 15 UEs. The sum-rate of decentralized MRT precoding is similar to that of traditional MRT precoding. Adopting 64 antennas was shown to achieve 6.7-fold sum-rate gains, while consuming only 1/64 of the transmit power compared to a single antenna BIB010 .

Survey of Large-Scale MIMO Systems <s> 3) Simulation Results: <s> A cellular base station serves a multiplicity of single-antenna terminals over the same time-frequency interval. Time-division duplex operation combined with reverse-link pilots enables the base station to estimate the reciprocal forward- and reverse-link channels. The conjugate-transpose of the channel estimates are used as a linear precoder and combiner respectively on the forward and reverse links. Propagation, unknown to both terminals and base station, comprises fast fading, log-normal shadow fading, and geometric attenuation. In the limit of an infinite number of antennas a complete multi-cellular analysis, which accounts for inter-cellular interference and the overhead and errors associated with channel-state information, yields a number of mathematically exact conclusions and points to a desirable direction towards which cellular wireless could evolve. In particular the effects of uncorrelated noise and fast fading vanish, throughput and the number of terminals are independent of the size of the cells, spectral efficiency is independent of bandwidth, and the required transmitted energy per bit vanishes. The only remaining impairment is inter-cellular interference caused by re-use of the pilot sequences in other cells (pilot contamination) which does not vanish with unlimited number of antennas. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> 3) Simulation Results: <s> We consider the uplink of a multicell multiuser MIMO system. The data detection is done by using the zero-forcing (ZF) technique, assuming the base station has perfect channel state information. We derive an exact closed-form expression for the uplink rate per user. We further study the asymptotic performance of the system. We show that, at high signal-to-noise ratio, the system is interference-limited and hence, we cannot improve the system performance by increasing the transmit power at each user. Instead, by increasing the number of base station antennas, the effects of interference and noise can be reduced, thereby improving the system performance. In particular, we show that, with very large antenna arrays at the base station, the transmit power of each user can be made inversely proportional to the number of base station antennas while maintaining a desired quality-of-service. Numerical results verify our analysis. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> 3) Simulation Results: <s> We consider the uplink (UL) and downlink (DL) of non-cooperative multi-cellular time-division duplexing (TDD) systems, assuming that the number N of antennas per base station (BS) and the number K of user terminals (UTs) per cell are large. Our system model accounts for channel estimation, pilot contamination, and an arbitrary path loss and antenna correlation for each link. We derive approximations of achievable rates with several linear precoders and detectors which are proven to be asymptotically tight, but accurate for realistic system dimensions, as shown by simulations. It is known from previous work assuming uncorrelated channels, that as N→∞ while K is fixed, the system performance is limited by pilot contamination, the simplest precoders/detectors, i.e., eigenbeamforming (BF) and matched filter (MF), are optimal, and the transmit power can be made arbitrarily small. We analyze to which extent these conclusions hold in the more realistic setting where N is not extremely large compared to K. In particular, we derive how many antennas per UT are needed to achieve η% of the ultimate performance limit with infinitely many antennas and how many more antennas are needed with MF and BF to achieve the performance of minimum mean-square error (MMSE) detection and regularized zero-forcing (RZF), respectively. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> 3) Simulation Results: <s> We consider multicell multiuser MIMO systems with a very large number of antennas at the base station (BS). We assume that the channel is estimated by using uplink training. We further consider a physical channel model where the angular domain is separated into a finite number of distinct directions. We analyze the so-called pilot contamination effect discovered in previous work, and show that this effect persists under the finite-dimensional channel model that we consider. In particular, we consider a uniform array at the BS. For this scenario, we show that when the number of BS antennas goes to infinity, the system performance under a finite-dimensional channel model with P angular bins is the same as the performance under an uncorrelated channel model with P antennas. We further derive a lower bound on the achievable rate of uplink data transmission with a linear detector at the BS. We then specialize this lower bound to the cases of maximum-ratio combining (MRC) and zero-forcing (ZF) receivers, for a finite and an infinite number of BS antennas. Numerical results corroborate our analysis and show a comparison between the performances of MRC and ZF in terms of sum-rate. <s> BIB004

The quantitative results of several linear precoders are illustrated in Fig. 10 . Unless otherwise specified, the default simulation configurations in this survey are as follows: a) The number of transmit antennas is 128; d is the distance between the eNB and the given UE in kilometers, the log-normal shadow fading is also considered with the standard variance of 8 dB, and the i.i.d. Rayleigh channel is assumed for the small-scale fading; e) The transmit power at the eNB is 14 dBm and the noise power spectral density is −174 dBm/Hz; f) In the single-cell scenario, apart from the inner circle associated with a radius of 10 meter, the UEs are uniformly distributed in hexagonal cells, whose ISD is 500 meter. As there is no performance difference between the CR precoder and the ZF precoder for transmission over the i.i.d. Rayleigh channel, only one of them has to be discussed, say the ZF precoder. As shown in Fig. 10(a) , the BE per UE is improved upon increasing the number of antennas at the eNB. However, the gains of the different precoders are not the same for different numbers of antennas. On the other hand, when multiplexing a large number of UEs in the system equipped with a given number of antennas, say 128, the average array gain of each UE is reduced, which impairs the BE in Fig. 10(b) . However, the total BE of all UEs is increased as a benefit of the multiplexing gains. The properties of a variety of LS-MIMO precoders and detectors operating in single-cell environments are summarized in Table IV . Generally, the precoders or detectors associated with a higher complexity offer an improved BE performance. Deploying more number of antennas at the eNB is capable of improving both the BE and EE. The performance of linear precoders or detectors is capable of approximating those of the high-complexity non-linear precoders or detectors. Precoders operating in realistic channels are capable of achieving a BE close to that estimated for idealized theoretical channel models. Assuming ideal CSI or estimated CSI, the transmit power of each antenna can be made inversely proportional to the number of AEs or to the square-root of the number of AEs deployed at the eNB, respectively. BIB001 , BIB003 , BIB002 , BIB004 : This section discussestheperformance of LS-MIMO in non-cooperative multi-cell multi-user systems. Fig. 11 illustrates the UL and DL interference encountered in such a multi-cell system. The affordable number of orthogonal pilots used for channel estimation is limited, because increasing the number and variety of pilots is only possible through enlarging the length of the pilot sequences, which ultimately increases the transmission overhead and/or the bandwidth. In practice the number of pilots may only be sufficient for supporting a limited number of UEs roaming in a single cell, which results in inevitable pilot reuse for the UEs in adjacent cells. As shown in Fig. 11 , employing the TDD mode, when the UEs of different cells send the same pilot sequence to their donor eNBs on the UL, each eNB has to rely on an interference-contaminated UL received signal, which inevitably contaminates the resultant channel estimate. Firstly, the interference-contaminated UL CSI is used for detecting the UL data. On the other hand, the DL channels may be assumed to be identical to the UL ones in the case of TDD-based reciprocity, provided that the bandwidth remains sufficiently narrow to encounter the non-frequencyselective propagation. Therefore, the UL CIR can be used by the DL transmit precoder. The precoded DL signals impinge on both the desired UEs and the UEs in other cells. Therefore, both the UL and DL sum rates are constrained by the adjacent cells interference imposed by pilot reuse. This phenomenon is known as pilot contamination and has been widely recognized as the main performance constraint of LS-MIMOs, as outlined below in more detail.

Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> A cellular base station serves a multiplicity of single-antenna terminals over the same time-frequency interval. Time-division duplex operation combined with reverse-link pilots enables the base station to estimate the reciprocal forward- and reverse-link channels. The conjugate-transpose of the channel estimates are used as a linear precoder and combiner respectively on the forward and reverse links. Propagation, unknown to both terminals and base station, comprises fast fading, log-normal shadow fading, and geometric attenuation. In the limit of an infinite number of antennas a complete multi-cellular analysis, which accounts for inter-cellular interference and the overhead and errors associated with channel-state information, yields a number of mathematically exact conclusions and points to a desirable direction towards which cellular wireless could evolve. In particular the effects of uncorrelated noise and fast fading vanish, throughput and the number of terminals are independent of the size of the cells, spectral efficiency is independent of bandwidth, and the required transmitted energy per bit vanishes. The only remaining impairment is inter-cellular interference caused by re-use of the pilot sequences in other cells (pilot contamination) which does not vanish with unlimited number of antennas. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> We consider the uplink of a multicell multiuser MIMO system. The data detection is done by using the zero-forcing (ZF) technique, assuming the base station has perfect channel state information. We derive an exact closed-form expression for the uplink rate per user. We further study the asymptotic performance of the system. We show that, at high signal-to-noise ratio, the system is interference-limited and hence, we cannot improve the system performance by increasing the transmit power at each user. Instead, by increasing the number of base station antennas, the effects of interference and noise can be reduced, thereby improving the system performance. In particular, we show that, with very large antenna arrays at the base station, the transmit power of each user can be made inversely proportional to the number of base station antennas while maintaining a desired quality-of-service. Numerical results verify our analysis. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> This paper considers a multi-cell multiple antenna system with precoding used at the base stations for downlink transmission. For precoding at the base stations, channel state information (CSI) is essential at the base stations. A popular technique for obtaining this CSI in time division duplex (TDD) systems is uplink training by utilizing the reciprocity of the wireless medium. This paper mathematically characterizes the impact that uplink training has on the performance of such multi-cell multiple antenna systems. When non-orthogonal training sequences are used for uplink training, the paper shows that the precoding matrix used by the base station in one cell becomes corrupted by the channel between that base station and the users in other cells in an undesirable manner. This paper analyzes this fundamental problem of pilot contamination in multi-cell systems. Furthermore, it develops a new multi-cell MMSE-based precoding method that mitigate this problem. In addition to being a linear precoding method, this precoding method has a simple closed-form expression that results from an intuitive optimization problem formulation. Numerical results show significant performance gains compared to certain popular single-cell precoding methods. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> This paper addresses the problem of channel estimation in multi-cell interference-limited cellular networks. We consider systems employing multiple antennas and are interested in both the finite and large-scale antenna number regimes (so-called "massive MIMO"). Such systems deal with the multi-cell interference by way of per-cell beamforming applied at each base station. Channel estimation in such networks, which is known to be hampered by the pilot contamination effect, constitutes a major bottleneck for overall performance. We present a novel approach which tackles this problem by enabling a low-rate coordination between cells during the channel estimation phase itself. The coordination makes use of the additional second-order statistical information about the user channels, which are shown to offer a powerful way of discriminating across interfering users with even strongly correlated pilot sequences. Importantly, we demonstrate analytically that in the large-number-of-antennas regime, the pilot contamination effect is made to vanish completely under certain conditions on the channel covariance. Gains over the conventional channel estimation framework are confirmed by our simulations for even small antenna array sizes. <s> BIB004 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> This paper considers multicell multiuser MIMO systems with very large antenna arrays at the base station. We propose an eigenvalue-decomposition-based approach to channel estimation, that estimates the channel blindly from the received data. The approach exploits the asymptotic orthogonality of the channel vectors in very large MIMO systems. We show that the channel to each user can be estimated from the covariance matrix of the received signals, up to a remaining scalar multiplicative ambiguity. A short training sequence is required to resolve this ambiguity. Furthermore, to improve the performance of our approach, we combine it with the iterative least-square with projection (ILSP) algorithm. Numerical results verify the effectiveness of our channel estimation approach. <s> BIB005 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> An LSAS entails a large number (tens or hundreds) of base station antennas serving a much smaller number of terminals, with large gains in spectral-efficiency and energy-efficiency compared with conventional MIMO technology. Until recently it was believed that in multi-cellular LSAS, even in the asymptotic regime, as the number of service antennas tends to infinity, the performance is limited by directed inter-cellular interference. The interference results from unavoidable re-use of reverse-link pilot sequences (pilot contamination) by terminals in different cells. We devise a new concept that leads to the effective elimination of inter-cell interference in TDD LSAS systems. This is achieved by outer multi-cellular pre-coding, which we call pilot contamination pre-coding (PCP). The main idea of PCP is that each base station linearly combines messages aimed to terminals from different cells that re-use the same pilot sequence. Crucially, the combining coefficients depend only on the slow-fading coefficients between the terminals and the base stations. Each base station independently transmits its PCP-combined symbols using conventional linear pre-coding that is based on estimated fast-fading coefficients. Further we derive estimates for SINRs and a capacity lower bound for the case of LSASs with PCP and finite number of antennas M. <s> BIB006 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> We consider the uplink (UL) and downlink (DL) of non-cooperative multi-cellular time-division duplexing (TDD) systems, assuming that the number N of antennas per base station (BS) and the number K of user terminals (UTs) per cell are large. Our system model accounts for channel estimation, pilot contamination, and an arbitrary path loss and antenna correlation for each link. We derive approximations of achievable rates with several linear precoders and detectors which are proven to be asymptotically tight, but accurate for realistic system dimensions, as shown by simulations. It is known from previous work assuming uncorrelated channels, that as N→∞ while K is fixed, the system performance is limited by pilot contamination, the simplest precoders/detectors, i.e., eigenbeamforming (BF) and matched filter (MF), are optimal, and the transmit power can be made arbitrarily small. We analyze to which extent these conclusions hold in the more realistic setting where N is not extremely large compared to K. In particular, we derive how many antennas per UT are needed to achieve η% of the ultimate performance limit with infinitely many antennas and how many more antennas are needed with MF and BF to achieve the performance of minimum mean-square error (MMSE) detection and regularized zero-forcing (RZF), respectively. <s> BIB007 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> We consider multicell multiuser MIMO systems with a very large number of antennas at the base station (BS). We assume that the channel is estimated by using uplink training. We further consider a physical channel model where the angular domain is separated into a finite number of distinct directions. We analyze the so-called pilot contamination effect discovered in previous work, and show that this effect persists under the finite-dimensional channel model that we consider. In particular, we consider a uniform array at the BS. For this scenario, we show that when the number of BS antennas goes to infinity, the system performance under a finite-dimensional channel model with P angular bins is the same as the performance under an uncorrelated channel model with P antennas. We further derive a lower bound on the achievable rate of uplink data transmission with a linear detector at the BS. We then specialize this lower bound to the cases of maximum-ratio combining (MRC) and zero-forcing (ZF) receivers, for a finite and an infinite number of BS antennas. Numerical results corroborate our analysis and show a comparison between the performances of MRC and ZF in terms of sum-rate. <s> BIB008 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> Large-scale multi-cell multiuser multiple-input multiple-output (LS-MIMO) systems can greatly increase the spectral efficiency. But the performance of these systems is deteriorated by pilot contamination. In this paper, first, a pilot design criterion is proposed by exploiting the orthogonality of channel vectors of LS-MIMO systems. Second, following this criterion, Chu sequences based pilots are designed. Because of the proposed pilots, the channel estimate of most terminals of a cell is only interfered by the partial cells rather than all the other cells, where the latter is caused by traditional pilots. As a result, pilot contamination is mitigated. Numerical results verify the effectiveness of the proposed pilots. <s> BIB009 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> We investigate pilot contamination problem for very large MIMO multi-cell TDD system. The asymptotic sum rate of two typical precoding schemes, i.e., the single-cell zero forcing (ZF) precoding and the multi-cell minimum mean square error (MMSE) based coordinated precoding are first derived. Results show that these two schemes have the same asymptotic sum rate expression as the number of the base station antenna going to infinity, revealing that coordinated precoding based on local channel state information (CSI) only provides marginal gain in the presence of pilot contamination. Based on our derivations, a pilot scheduling scheme is further proposed to mitigate pilot contamination, which could provide much better performance. Numerical results finally verify our derivations and the proposed scheme. <s> BIB010 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> Pilot contamination is known to severely limit the performance of large-scale antenna (“massive MIMO”) systems due to degraded channel estimation. This paper proposes a twofold approach to this problem. First we show analytically that pilot contamination can be made to vanish asymptotically in the number of antennas for a certain class of channel fading statistics. The key lies in setting a suitable condition on the second order statistics for desired and interference signals. Second we show how a coordinated user-to-pilot assignment method can be devised to help fulfill this condition in practical networks. Large gains are illustrated in our simulations for even small antenna array sizes. <s> BIB011 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> In this paper we study the performance of cellular networks when their base stations have an unlimited number of antennas. In previous work, the asymptotic behavior of the signal to interference plus nose ratio (SINR) was obtained. We revisit these results by deriving the rigorous expression for the SINR of both downlink and uplink in the scenario of infinite number of antennas. We show that the contamination of the channel estimates happens whenever a pilot sequence is received at a base station simultaneously with non-orthogonal signals coming from other users. We propose a method to avoid such simultaneous transmissions from adjacent cells, thus significantly decreasing interference. We also investigate the effects of power allocation in this interference-limited scenario, and show that it results in gains of over 15dB in the signal to interference ratio for the scenario simulated here. The combination of these two techniques results in rate gains of about 18 times in our simulations. <s> BIB012 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> Multi-user MIMO offers big advantages over conventional point-to-point MIMO: it works with cheap single-antenna terminals, a rich scattering environment is not required, and resource allocation is simplified because every active terminal utilizes all of the time-frequency bins. However, multi-user MIMO, as originally envisioned, with roughly equal numbers of service antennas and terminals and frequency-division duplex operation, is not a scalable technology. Massive MIMO (also known as large-scale antenna systems, very large MIMO, hyper MIMO, full-dimension MIMO, and ARGOS) makes a clean break with current practice through the use of a large excess of service antennas over active terminals and time-division duplex operation. Extra antennas help by focusing energy into ever smaller regions of space to bring huge improvements in throughput and radiated energy efficiency. Other benefits of massive MIMO include extensive use of inexpensive low-power components, reduced latency, simplification of the MAC layer, and robustness against intentional jamming. The anticipated throughput depends on the propagation environment providing asymptotically orthogonal channels to the terminals, but so far experiments have not disclosed any limitations in this regard. While massive MIMO renders many traditional research problems irrelevant, it uncovers entirely new problems that urgently need attention: the challenge of making many low-cost low-precision components that work effectively together, acquisition and synchronization for newly joined terminals, the exploitation of extra degrees of freedom provided by the excess of service antennas, reducing internal power consumption to achieve total energy efficiency reductions, and finding new deployment scenarios. This article presents an overview of the massive MIMO concept and contemporary research on the topic. <s> BIB013 </s> Survey of Large-Scale MIMO Systems <s> B. Performance of Precoders/Detectors in Multi-Cell Environments 1) Pilot Contamination in Multi-Cell Scenarios <s> A subspace projection to improve channel estimation in massive multi-antenna systems is proposed and analyzed. Together with power-controlled hand-off, it can mitigate the pilot contamination problem without the need for coordination among cells. The proposed method is blind in the sense that it does not require pilot data to find the appropriate subspace. It is based on the theory of large random matrices that predicts that the eigenvalue spectra of large sample covariance matrices can asymptotically decompose into disjoint bulks as the matrix size grows large. Random matrix and free probability theory are utilized to predict under which system parameters such a bulk decomposition takes place. Simulation results are provided to confirm that the proposed method outperforms conventional linear channel estimation if bulk separation occurs. <s> BIB014

• i.i.d. Rayleigh channel: Assuming that the i.i.d. Rayleigh channel can be ideally estimated at the eNB, the performance of the ZF detector is analyzed in a UL LS-MIMO system in the multi-cell scenario BIB002 . Both the exact closed-form and lower bound of the UL BE indicate that the system is interference-limited at high SNRs. Hence, boosting the transmit power at the UEs cannot further improve the attainable performance. The effects of interference and noise can be reduced by employing more antennas at the eNB. However, in the presence of realistically estimated CSI, the received SINRs of both the MRT precoder and of the MRC detector demonstrate that the effects of the Additive White Gaussian Noise (AWGN) and IUI disappear, and that the only remaining constraint is imposed by pilot contamination BIB001 . Moreover, the results in BIB001 , BIB002 demonstrate that the required radiated energy per bit is reduced upon increasing the number of AEs. • Correlated Rayleigh channel: Under the correlated channel model, the approximate sum rate is studied in BIB007 , where the MRT/RZF precoder is employed for the DL or MRC/MMSE detection is adopted for the UL BIB007 . The BE attained depends mainly on the effective SNR as well as the DoF, which is defined as the ratio of the rank of the correlation matrix to the number of UEs. Similar conclusions in BIB007 can be found in BIB001 , i.e., when the number of antennas tends to infinity, both the thermal noise and the IUI can be averaged out. Hence, pilot contamination remains the main constraint of LS-MIMOs, and the radiated power can be kept low. Moreover, the number of antennas required for different TPCs or detectors has to be deduced on the basis of aiming for a given percentage of the BE attained with the aid of an infinite number of antennas. The RZF/MMSE precoder/reciever is capable of achieving the same BE using a reduced number of antennas compared to the MRT/MRC schemes. • Dispersive multi-path channel: The performances of both the MRC and ZF detectors are studied in the UL of a LS-MIMO system in a multi-cell environment, where a dispersive multi-path channel model is assumed BIB008 . The lower performance bounds derived in BIB008 demonstrate that pilot contamination remains the dominant performance limiting factor for the LS-MIMO with a realistic finite-dimensional channel model. The ZF receiver performs better than the MRC receiver in terms of the achievable sum-rate, when pilot contamination is not severe, and vice versa. A rich scattering propagation environment may benefit the ZF detector. However, the MRC detector performs better than ZF in poor scattering propagation environments. If the AoAs of the UEs roaming in the specific cell that uses the same pilots are not identical, pilot contamination can be completely eliminated with the aid of Bayesian channel estimation BIB004 . In other words, if the covariances of the desired signal and interference span different subspaces, pilot contamination can be eliminated, provided that there are an unlimited number of antennas. In order to explicitly show the effects of pilot contamination imposed on the different precoders in the multi-cell scenario, the BE per UE under the i.i.d. Rayleigh channel is depicted in Fig. 12 . The default simulation configurations are the same as those of the single-cell scenario, except that the number of cells is 7. It can be seen that the performance trends of all precoders involved in the multi-cell scenario are similar to those in the single-cell scenario. However, the BE performances are degraded due to the pilot contamination compared to those seen in Fig. 10 . Moreover, the CEP is more sensitive to pilot contamination in contrast to the other precoders. 2) Remedies of Pilot Contamination: As discussed before, when realistic imperfect CSI is acquired at the eNB, the pilot contamination caused by pilot reuse in multi-cell scenarios imposes the ultimate limitation on the attainable performance of LS-MIMO. This subsection studies the main techniques of eliminating or at least alleviating pilot contamination. As Table V , the main counter-measures can be classified as follows. • Pilot design: When the classic Matched Filter (MF) is invoked for estimating the UL channel through the use of UL pilots in the i.i.d. Rayleigh channel, a specific pilot design criterion, aiming to minimize the inner product of the pilots for different cells is proposed in BIB009 for mitigating pilot contamination. Based on this criterion, Chu-sequence-based pilots are designed, which makes the pilots for the UEs in the same cell orthogonal, and reuses the same pilots in the neighboring cells after suboptimal phase shifts. Therefore, the accuracy of the estimated channels can be substantially improved and pilot contamination can be beneficially mitigated. • Pilot allocation: When the eNB employs an MMSE channel estimator in the i.i.d. Rayleigh channel, pilot contamination is proved to also be the dominant constraint of BE even with the assistance of multi-cell MMSE TPC BIB010 . However, if we allow the UEs benefitting from low ICI to reuse the same pilots, pilot contamination may be mitigated and substantial BE improvements can be achieved BIB010 . In BIB004 , BIB011 , a Bayesian channel estimator is first developed for the UL channel in the multi-cell scenario. Then, it is shown that pilot contamination is commensurately reduced, when using more antennas at the eNB, provided that the covariances of the desired signal and interference span different subspaces. Furthermore, a pilot allocation scheme is proposed for suppressing the pilot contamination by carefully shaping the covariance in order to satisfy this condition. • Frame structure: A time-shifted frame structure is designed in BIB012 for mitigating pilot contamination based on MF-assisted channel estimation. In this scheme, all the cells are divided into different groups, which transmit their UL pilots in different time slots. When a specific group transmits its UL pilots, the other groups transmit their DL data. An example of the frame structure of three groups is depicted in Fig. 13 . The theoretical proof provided in BIB012 shows that the ICI imposed by the different groups is gradually mitigated upon employing more antennas at the eNB. In order to efficiently suppress the ICI arriving from the inner group, optimum power allocation may be employed according to the specific received SINRs on both the UL and DL BIB012 , . • Channel estimation: Taking advantage of the asymptotic orthogonality of different UE channels in the i.i.d. Rayleigh channel, an eigenvalue-decomposition-based approach is proposed in BIB005 to improve the channel estimation accuracy. Hence, the fast fading channel coefficients may be estimated with the aid of joint estimation of the channels and data by blind techniques, whilst mitigating pilot contamination BIB013 , BIB014 . • Precoding scheme: A pilot contamination mitigating precoding scheme is designed in BIB006 for communicating over an i.i.d. Rayleigh channel. This two-stage precoding scheme consists of an outer multi-cell TPC arrangement based on the knowledge of large-scale fading features, and conventional inner linear TPC based on the estimated fast fading coefficients. The essential idea behind this scheme is that the eNBs linearly combine their signals intended for all the UEs by reusing the same pilots. The combined symbols are then transmitted with the aid of traditional precoding. As a result, the ICI can be completely eliminated, as shown by the theoretical derivations in BIB006 . A further optimization problem is formulated in BIB003 to mitigate pilot contamination. The objective function used is constituted by the Mean Square Error (MSE) of the received signals of the UEs in the same cell, plus the mean-square interference power imposed on the UEs in other cells. Then, the optimal closed-form expression of precoder for this MMSE-based multi-cell precoding optimization problem is derived analytically. The proposed precoding scheme is capable of simultaneously reducing both the intra-cell and inter-cell interferences. It is implemented at each eNB without information exchange among eNBs, which has the benefit of lower overhead than the joint-multi-cell precoding in BIB006 .

Survey of Large-Scale MIMO Systems <s> 1) Imperfect CSI: <s> Feedback delay can severely affect the performance of transmit beamforming (TB) and the analytical quantification of the performance degradation has attracted much research interest recently. In this letter, we study the effect of delayed and limited-rate codebook index feedback on the error rate performance of TB systems over Rayleigh fading channels. We derive closed-form expressions for the moment generating function (MGF) and the probability density function (PDF) of the receiver output signal-to-noise ratio (SNR) including the effects of outdated and finite-rate feedback and further provide accurate analytical error rate expressions, which are verified by simulation results. The coding gain gap between the full-rate and limited-rate feedback and the coding gain advantage of multiple transmit antennas to the single antenna are analyzed, for limited and delayed feedback. These results are simple and concise and provide new analytical insight into the achievable diversity and combining gains and the loss caused by feedback delay for different system parameters and modulation formats. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> 1) Imperfect CSI: <s> This paper proposes an adaptive multi-mode transmission strategy to improve the spectral efficiency achieved in the multiple-input multiple-output (MIMO) broadcast channel with delayed and quantized channel state information. The adaptive strategy adjusts the number of active users, denoted as the transmission mode, to balance transmit array gain, spatial division multiplexing gain, and residual inter-user interference. Accurate closed-form approximations are derived for the achievable rates for different modes, which help identify the active mode that maximizes the average sum throughput for given feedback delay and channel quantization error. The proposed transmission strategy can be easily combined with round-robin scheduling to serve a large number of users. As instantaneous channel information is not exploited, the proposed algorithm cannot provide multiuser diversity gain, but it is still able to provide throughput gain over single-user MIMO at moderate signal-to-noise ratio. In addition, it has a light feedback overhead and only requires feedback of instantaneous channel state information from a small number of users. In the system with a feedback load constraint, it is shown that the proposed algorithm provides performance close to that achieved by opportunistic scheduling with instantaneous feedback from a large number of users. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> 1) Imperfect CSI: <s> An LSAS entails a large number (tens or hundreds) of base station antennas serving a much smaller number of terminals, with large gains in spectral-efficiency and energy-efficiency compared with conventional MIMO technology. Until recently it was believed that in multi-cellular LSAS, even in the asymptotic regime, as the number of service antennas tends to infinity, the performance is limited by directed inter-cellular interference. The interference results from unavoidable re-use of reverse-link pilot sequences (pilot contamination) by terminals in different cells. We devise a new concept that leads to the effective elimination of inter-cell interference in TDD LSAS systems. This is achieved by outer multi-cellular pre-coding, which we call pilot contamination pre-coding (PCP). The main idea of PCP is that each base station linearly combines messages aimed to terminals from different cells that re-use the same pilot sequence. Crucially, the combining coefficients depend only on the slow-fading coefficients between the terminals and the base stations. Each base station independently transmits its PCP-combined symbols using conventional linear pre-coding that is based on estimated fast-fading coefficients. Further we derive estimates for SINRs and a capacity lower bound for the case of LSASs with PCP and finite number of antennas M. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> 1) Imperfect CSI: <s> This paper considers multicell multiuser MIMO systems with very large antenna arrays at the base station. We propose an eigenvalue-decomposition-based approach to channel estimation, that estimates the channel blindly from the received data. The approach exploits the asymptotic orthogonality of the channel vectors in very large MIMO systems. We show that the channel to each user can be estimated from the covariance matrix of the received signals, up to a remaining scalar multiplicative ambiguity. A short training sequence is required to resolve this ambiguity. Furthermore, to improve the performance of our approach, we combine it with the iterative least-square with projection (ILSP) algorithm. Numerical results verify the effectiveness of our channel estimation approach. <s> BIB004 </s> Survey of Large-Scale MIMO Systems <s> 1) Imperfect CSI: <s> We consider the uplink (UL) and downlink (DL) of non-cooperative multi-cellular time-division duplexing (TDD) systems, assuming that the number N of antennas per base station (BS) and the number K of user terminals (UTs) per cell are large. Our system model accounts for channel estimation, pilot contamination, and an arbitrary path loss and antenna correlation for each link. We derive approximations of achievable rates with several linear precoders and detectors which are proven to be asymptotically tight, but accurate for realistic system dimensions, as shown by simulations. It is known from previous work assuming uncorrelated channels, that as N→∞ while K is fixed, the system performance is limited by pilot contamination, the simplest precoders/detectors, i.e., eigenbeamforming (BF) and matched filter (MF), are optimal, and the transmit power can be made arbitrarily small. We analyze to which extent these conclusions hold in the more realistic setting where N is not extremely large compared to K. In particular, we derive how many antennas per UT are needed to achieve η% of the ultimate performance limit with infinitely many antennas and how many more antennas are needed with MF and BF to achieve the performance of minimum mean-square error (MMSE) detection and regularized zero-forcing (RZF), respectively. <s> BIB005 </s> Survey of Large-Scale MIMO Systems <s> 1) Imperfect CSI: <s> Multiple-input multiple-output (MIMO) communication may provide high spectral efficiency through the deployment of a very large number of antenna elements at the base stations. The gains from massive MIMO communication come from the use of multiuser MIMO on the uplink and downlink, but with a large excess of antennas at the base station compared to the number of served users. Initial work on massive MIMO did not fully address several practical issues associated with its deployment. This paper considers the impact of channel aging on the performance of massive MIMO systems. The effects of channel variation are characterized as a function of different system parameters assuming a simple model for the channel time variations at the transmitter. Channel prediction is proposed to overcome channel aging effects. The analytical results on aging show how capacity is lost due to time variation in the channel. Numerical results in a multiceli network show that massive MIMO works even with some channel variation and that channel prediction could partially overcome channel aging effects. <s> BIB006 </s> Survey of Large-Scale MIMO Systems <s> 1) Imperfect CSI: <s> To fully utilize the spatial multiplexing gains or array gains of massive MIMO, the channel state information must be obtained at the transmitter side (CSIT). However, conventional CSIT estimation approaches are not suitable for FDD massive MIMO systems because of the overwhelming training and feedback overhead. In this paper, we consider multi-user massive MIMO systems and deploy the compressive sensing (CS) technique to reduce the training as well as the feedback overhead in the CSIT estimation. The multi-user massive MIMO systems exhibits a hidden joint sparsity structure in the user channel matrices due to the shared local scatterers in the physical propagation environment. As such, instead of naively applying the conventional CS to the CSIT estimation, we propose a distributed compressive CSIT estimation scheme so that the compressed measurements are observed at the users locally, while the CSIT recovery is performed at the base station jointly. A joint orthogonal matching pursuit recovery algorithm is proposed to perform the CSIT recovery, with the capability of exploiting the hidden joint sparsity in the user channel matrices. We analyze the obtained CSIT quality in terms of the normalized mean absolute error, and through the closed-form expressions, we obtain simple insights into how the joint channel sparsity can be exploited to improve the CSIT recovery performance. <s> BIB007

• Imperfect channel estimation: Since in practice both the precoder and the detector have to rely on realistically estimated CSI, accurate channel estimation is vital for the performance of the LS-MIMO system. Under the TDD mode, due to pilot reuse in multi-cell scenarios, the CSI obtained by linear channel estimators, such as the MMSE and MF estimators, is contaminated by interference BIB005 , BIB003 . In order to eliminate or mitigate the pilot overhead, blind channel estimation is considered in BIB004 . Under the FDD mode, the feedback from UEs to eNBs may become overwhelming for practical systems, which can be mitigated by compressive sensing BIB007 . However, channel estimation for LS-MIMO, such as a 100 × 10 element system remains a challenging problem, since the complexity of estimating or recovering 1,000 channels is excessive. Therefore, non-coherent MIMO also has to be studied. • Channel ageing: Adopting an autoregressive model for assessing the predictability of a realistic channel, the effect of channel ageing was investigated for LS-MIMO system in BIB006 . Channel ageing results in a mismatch between the current channel to be encountered by the next transmission and the channel detected at the DL receiver, quantized and sent back for precoding. In order to mitigate the effects of channel ageing on the sum rate of LS-MIMO, an optimal causal linear Finite Impulse Response (FIR) Wiener channel predictor is employed in BIB006 and as a benefit, its sum rate is substantially improved. In a practical scenario having an imperfect CSI can be characterized for example by the correlation coefficient between the estimated CSI and the perfect CSI. At the time of writing, there are two main methods of modeling the correlation coefficient, namely the Gaussian model BIB001 and the Clarke model BIB002 . They describe the correlation coefficient as the function of Doppler frequency shift and channel delay in the form of either an exponential function or the zeroth-order Bessel function, respectively. When the correlation coefficient calculated by the Gaussian model is 0.9, the quantitative performances of the different linear precoders are shown in Fig. 14. As expected, the BE is reduced due to the inaccurate CSI compared to that seen in Fig. 10 . Therefore, advanced techniques are required for mitigating the effects of imperfect CSI.

Survey of Large-Scale MIMO Systems <s> 2) Non-Ideal Hardware: <s> In order to support multiple antennas on compact wireless devices, transceivers are often designed with matching networks that compensate for mutual coupling. Some works have suggested that when optimal matching is applied to such a system, performance at the center frequency can be improved at the expense of an apparent reduction in the system bandwidth. This paper addresses the question of how coupling impacts bandwidth in the context of circular arrays. It will be shown that mutual coupling creates eigen-modes (virtual antennas) with diverse frequency responses, using the standard matching techniques. We shall also demonstrate how common communications techniques such as Diversity-OFDM would need to be optimized in order to compensate for these effects. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> 2) Non-Ideal Hardware: <s> Multiuser multiple-input–multiple-output (MIMO) cellular systems with an excess of base station (BS) antennas (Massive MIMO) offer unprecedented multiplexing gains and radiated energy efficiency. Oscillator phase noise is introduced in the transmitter and receiver radio frequency chains and severely degrades the performance of communication systems. We study the effect of oscillator phase noise in frequency-selective Massive MIMO systems with imperfect channel state information. In particular, we consider two distinct operation modes, namely, when the phase noise processes at the $M$ BS antennas are identical (synchronous operation) and when they are independent (nonsynchronous operation) . We analyze a linear and low-complexity time-reversal maximum-ratio combining reception strategy. For both operation modes, we derive a lower bound on the sum-capacity, and we compare their performance. Based on the derived achievable sum-rates, we show that with the proposed receive processing, an $O(\sqrt{M} ) $ array gain is achievable. Due to the phase noise drift, the estimated effective channel becomes progressively outdated. Therefore, phase noise effectively limits the length of the interval used for data transmission and the number of scheduled users. The derived achievable rates provide insights into the optimum choice of the data interval length and the number of scheduled users. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> 2) Non-Ideal Hardware: <s> The use of large-scale antenna arrays can bring substantial improvements in energy and/or spectral efficiency to wireless systems due to the greatly improved spatial resolution and array gain. Recent works in the field of massive multiple-input multiple-output (MIMO) show that the user channels decorrelate when the number of antennas at the base stations (BSs) increases, thus strong signal gains are achievable with little interuser interference. Since these results rely on asymptotics, it is important to investigate whether the conventional system models are reasonable in this asymptotic regime. This paper considers a new system model that incorporates general transceiver hardware impairments at both the BSs (equipped with large antenna arrays) and the single-antenna user equipments (UEs). As opposed to the conventional case of ideal hardware, we show that hardware impairments create finite ceilings on the channel estimation accuracy and on the downlink/uplink capacity of each UE. Surprisingly, the capacity is mainly limited by the hardware at the UE, while the impact of impairments in the large-scale arrays vanishes asymptotically and interuser interference (in particular, pilot contamination) becomes negligible. Furthermore, we prove that the huge degrees of freedom offered by massive MIMO can be used to reduce the transmit power and/or to tolerate larger hardware impairments, which allows for the use of inexpensive and energy-efficient antenna elements. <s> BIB003

• Mutual coupling: Upon increasing the number of antennas at the eNB, either the spacing between the AEs is reduced, or alternatively, if the spacing is fixed, then the total AA dimension is increased. Having a reduced AE spacing leads to increased mutual coupling, which impairs the performance of LS-MIMO. In order to mitigate the effects of mutual coupling, sophisticated radiofrequency matching techniques have to be employed between the AA and the radio-frequency chain. In BIB001 , the optimal matching network is designed for circular arrays, which substantially improves attainable system capacity. • Phase noise: The phase noise is caused mainly by the upconversion stage used at the transmitter and by the downconversion circuit of the receiver. When considering the UL of an LS-MIMO system, the phase noise of the UE transmitter is independent of the phase noise of all other non-cooperative UEs. By contrast, the phase noise of different antennas at the eNB is dependent on whether a single-oscillator centralized eNB or a distinct-oscillator based distributed eNB is considered. When adopting MRC detection, the lower bounds of the sum-rate found for both type of solutions suggest that the achievable array gain is proportional to the square-root of the number of antennas BIB002 . Due to the progressive phase noise drift at the oscillators, there exists a fundamental trade-off between the time interval used for data transmission and the sum-rate. This trade-off offers an insight into the impact of transmission parameters. • Additive distortion of the baseband: The amplifier nonlinearity, the I/Q imbalance and phase noise, as well as the additive distortion terms encountered at the transceivers of both the eNB and UEs were introduced into the system model in BIB003 . Both the channel estimation accuracy and the sum rate of UEs are predominantly limited by the distortion encountered by the UEs, not by the eNBs. Both the impact of the distortion imposed on the eNB and pilot contamination were mitigated by deploying more number of antennas in BIB003 . Moreover, the attainable EE can be improved by employing more number of transmit antennas, while simultaneously reducing the radiated power. Therefore, the performance degradation imposed by lowcost AEs can be mitigated by increasing the number of transmit antennas.

Survey of Large-Scale MIMO Systems <s> D. Summary of Physical Layer Issues <s> This paper considers a multi-cell multiple antenna system with precoding used at the base stations for downlink transmission. For precoding at the base stations, channel state information (CSI) is essential at the base stations. A popular technique for obtaining this CSI in time division duplex (TDD) systems is uplink training by utilizing the reciprocity of the wireless medium. This paper mathematically characterizes the impact that uplink training has on the performance of such multi-cell multiple antenna systems. When non-orthogonal training sequences are used for uplink training, the paper shows that the precoding matrix used by the base station in one cell becomes corrupted by the channel between that base station and the users in other cells in an undesirable manner. This paper analyzes this fundamental problem of pilot contamination in multi-cell systems. Furthermore, it develops a new multi-cell MMSE-based precoding method that mitigate this problem. In addition to being a linear precoding method, this precoding method has a simple closed-form expression that results from an intuitive optimization problem formulation. Numerical results show significant performance gains compared to certain popular single-cell precoding methods. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> D. Summary of Physical Layer Issues <s> An LSAS entails a large number (tens or hundreds) of base station antennas serving a much smaller number of terminals, with large gains in spectral-efficiency and energy-efficiency compared with conventional MIMO technology. Until recently it was believed that in multi-cellular LSAS, even in the asymptotic regime, as the number of service antennas tends to infinity, the performance is limited by directed inter-cellular interference. The interference results from unavoidable re-use of reverse-link pilot sequences (pilot contamination) by terminals in different cells. We devise a new concept that leads to the effective elimination of inter-cell interference in TDD LSAS systems. This is achieved by outer multi-cellular pre-coding, which we call pilot contamination pre-coding (PCP). The main idea of PCP is that each base station linearly combines messages aimed to terminals from different cells that re-use the same pilot sequence. Crucially, the combining coefficients depend only on the slow-fading coefficients between the terminals and the base stations. Each base station independently transmits its PCP-combined symbols using conventional linear pre-coding that is based on estimated fast-fading coefficients. Further we derive estimates for SINRs and a capacity lower bound for the case of LSASs with PCP and finite number of antennas M. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> D. Summary of Physical Layer Issues <s> Presents a list of articles published by the IEEE Signal Processing Society (SPS) that ranked among the top 100 most downloaded IEEE Xplore articles. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> D. Summary of Physical Layer Issues <s> A multiplicity of autonomous terminals simultaneously transmits data streams to a compact array of antennas. The array uses imperfect channel-state information derived from transmitted pilots to extract the individual data streams. The power radiated by the terminals can be made inversely proportional to the square-root of the number of base station antennas with no reduction in performance. In contrast if perfect channel-state information were available the power could be made inversely proportional to the number of antennas. Lower capacity bounds for maximum-ratio combining (MRC), zero-forcing (ZF) and minimum mean-square error (MMSE) detection are derived. An MRC receiver normally performs worse than ZF and MMSE. However as power levels are reduced, the cross-talk introduced by the inferior maximum-ratio receiver eventually falls below the noise level and this simple receiver becomes a viable option. The tradeoff between the energy efficiency (as measured in bits/J) and spectral efficiency (as measured in bits/channel use/terminal) is quantified for a channel model that includes small-scale fading but not large-scale fading. It is shown that the use of moderately large antenna arrays can improve the spectral and energy efficiency with orders of magnitude compared to a single-antenna system. <s> BIB004 </s> Survey of Large-Scale MIMO Systems <s> D. Summary of Physical Layer Issues <s> Large-scale multi-cell multiuser multiple-input multiple-output (LS-MIMO) systems can greatly increase the spectral efficiency. But the performance of these systems is deteriorated by pilot contamination. In this paper, first, a pilot design criterion is proposed by exploiting the orthogonality of channel vectors of LS-MIMO systems. Second, following this criterion, Chu sequences based pilots are designed. Because of the proposed pilots, the channel estimate of most terminals of a cell is only interfered by the partial cells rather than all the other cells, where the latter is caused by traditional pilots. As a result, pilot contamination is mitigated. Numerical results verify the effectiveness of the proposed pilots. <s> BIB005 </s> Survey of Large-Scale MIMO Systems <s> D. Summary of Physical Layer Issues <s> The use of large-scale antenna arrays can bring substantial improvements in energy and/or spectral efficiency to wireless systems due to the greatly improved spatial resolution and array gain. Recent works in the field of massive multiple-input multiple-output (MIMO) show that the user channels decorrelate when the number of antennas at the base stations (BSs) increases, thus strong signal gains are achievable with little interuser interference. Since these results rely on asymptotics, it is important to investigate whether the conventional system models are reasonable in this asymptotic regime. This paper considers a new system model that incorporates general transceiver hardware impairments at both the BSs (equipped with large antenna arrays) and the single-antenna user equipments (UEs). As opposed to the conventional case of ideal hardware, we show that hardware impairments create finite ceilings on the channel estimation accuracy and on the downlink/uplink capacity of each UE. Surprisingly, the capacity is mainly limited by the hardware at the UE, while the impact of impairments in the large-scale arrays vanishes asymptotically and interuser interference (in particular, pilot contamination) becomes negligible. Furthermore, we prove that the huge degrees of freedom offered by massive MIMO can be used to reduce the transmit power and/or to tolerate larger hardware impairments, which allows for the use of inexpensive and energy-efficient antenna elements. <s> BIB006 </s> Survey of Large-Scale MIMO Systems <s> D. Summary of Physical Layer Issues <s> Multiple-input multiple-output (MIMO) communication may provide high spectral efficiency through the deployment of a very large number of antenna elements at the base stations. The gains from massive MIMO communication come from the use of multiuser MIMO on the uplink and downlink, but with a large excess of antennas at the base station compared to the number of served users. Initial work on massive MIMO did not fully address several practical issues associated with its deployment. This paper considers the impact of channel aging on the performance of massive MIMO systems. The effects of channel variation are characterized as a function of different system parameters assuming a simple model for the channel time variations at the transmitter. Channel prediction is proposed to overcome channel aging effects. The analytical results on aging show how capacity is lost due to time variation in the channel. Numerical results in a multiceli network show that massive MIMO works even with some channel variation and that channel prediction could partially overcome channel aging effects. <s> BIB007 </s> Survey of Large-Scale MIMO Systems <s> D. Summary of Physical Layer Issues <s> In order to support multiple antennas on compact wireless devices, transceivers are often designed with matching networks that compensate for mutual coupling. Some works have suggested that when optimal matching is applied to such a system, performance at the center frequency can be improved at the expense of an apparent reduction in the system bandwidth. This paper addresses the question of how coupling impacts bandwidth in the context of circular arrays. It will be shown that mutual coupling creates eigen-modes (virtual antennas) with diverse frequency responses, using the standard matching techniques. We shall also demonstrate how common communications techniques such as Diversity-OFDM would need to be optimized in order to compensate for these effects. <s> BIB008

In this section, we mainly discussed the performance of diverse TPCs and detectors for LS-MIMO systems. In the single-cell scenario, low-complexity linear precoders and detectors may perform similarly to other complex precoders and detectors in terms of their sum-rate BIB003 , BIB004 . The beneficial combination of TPCs and detectors lead to the achievement of a performance close to their theoretical performance under realistic channel conditions. Both the attainable BE and EE can be improved by employing more number of antennas at the eNB. When near-perfect CSI can be acquired through channel estimation with high accuracy, the radiated power can be made inversely proportional to the number of transmit antennas at the eNB BIB004 , but only inversely proportional to the square-root of the number of antennas in the presence of realistically estimated CSI. In the multi-cell scenario, upon employing linear channel estimation with moderate complexity, pilot contamination becomes a major obstacle to improve the BE, which is due to the limited number of available pilots. Hence, diverse counter-measures have been conceived for mitigating pilot contamination BIB005 - BIB001 , such as specially conceived pilot design BIB005 , power allocation and pilot-decontamination precoding BIB002 . Finally, the performance of practical LS-MIMO systems is affected by numerous non-ideal factors BIB002 - BIB006 , including imperfect CSI BIB002 , BIB007 and non-ideal hardware BIB008 - BIB006 . As one of major research issues, the design of non-coherent blind detection or semi-blind assisted LS-MIMO systems has to be explored, which is capable of dispensing much less channel estimation information.

Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> This paper proposes a decentralized adaptive soft frequency reuse scheme for the uplink of 4G long-term evolution (LTE) systems. While universal frequency reuse (UFR) is being targeted for next generation multi-cellular wireless networks, ongoing efforts supporting the LTE standard have proved that actual implementations of UFR in LTE lead to unacceptable interference levels experienced by user equipments near the cell edge area in a multi-cellular configuration. The herein proposed adaptive soft frequency reuse scheme is a step forward towards effective inter-cell interference coordination (ICIC) in next-generation wireless networks. Our solution to the uplink ICIC problem stands out for its two essential features that consist of physical resource block (PRB) reuse avoidance/minimization and cell-edge bandwidth breathing which can be implemented at the cost of a negligible information exchange over the X2 interface (backbone). The PRB reuse avoidance feature significantly decreases inter-cell interference levels while improving the achievable average throughput per user, especially for those identified as cell-edge ones. The cell-edge bandwidth breathing strategy allows to track and adapt to semi-static changes in traffic loading and user distributions within each cell which drastically reduces the blocking probability of incoming calls under cell-edge bandwidth constrained traffic. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> Cooperative base-station (BS) signaling using MU-MIMO (Network MIMO) has received a great deal of attention given its ability to reduce inter-cell interference (ICI) and improve the system spectral efficiency. Indeed for a given number of antennas per BS, cooperative systems can have significant benefits over conventional cellular architectures. However cooperative signaling requires and uses more channel state information (CSI). This increases CSI signaling overhead, and can have a non-negligible effect on the system throughput. In fact once systems are compared while taking into account CSI overhead, the question of what system architecture is best becomes interesting and non-trivial. We provide one such alternative look at cooperative architectures. We show, despite prevailing views, that cellular architectures, using coordinated colocated antennas, can be quite attractive compared to Network MIMO. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> This paper gives an overview of the objectives and current research activities on interference avoidance in the EC funded research project ARTIST4G. It is the main objective of the ARTIST4G Work Package 1 (WP1), to build forward on the 3GPP LTE Release 8 baseline, proposing a novel fair mobile broadband technological framework in which to design innovative, practical, scalable and cost-effective interference avoidance solutions. This paper provides an overview of the fundamental classes of innovations studied in ARTIST4G Work Package 1 (WP1) and presents some performance results obtained by means of numerical simulations and test-bed measurements run in the field. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> Fractional frequency reuse (FFR) is an interference management technique well-suited to OFDMA-based cellular networks wherein the bandwidth of the cells is partitioned into regions with different frequency reuse factors. To date, FFR techniques have been typically been evaluated through system-level simulations using a hexagonal grid for the base station locations. This paper instead focuses on analytically evaluating the two main types of FFR deployments - Strict FFR and Soft Frequency Reuse (SFR) - using a Poisson point process to model the base station locations. The results are compared with the standard grid model and an actual urban deployment. Under reasonable special cases for modern cellular networks, our results reduce to simple closed-form expressions, which provide insight into system design guidelines and the relative merits of Strict FFR, SFR, universal reuse, and fixed frequency reuse. Finally, a SINR-proportional resource allocation strategy is proposed based on the analytical expressions and we observe that FFR provides an increase in the sum-rate as well as the well-known benefit of improved coverage for cell-edge users. <s> BIB004 </s> Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> This paper gives an overview on the possibilities and potential of dynamic vertical beamsteering in a cellular mobile radio system with focus on the interference limited macro-cell scenario. Different realization options for dynamic terminal specific downtilt adaptation at the base station (eNB) are introduced and simulated performance figures are given. Beam coordination methods for interference avoidance without and with the requirement for inter-eNB communications are considered. The impact of the most important system parameters like downtilt angle variation and coordination algorithm parameter setting is investigated. Basic measurements in real environment for proof of concept are introduced and their relation to simulation results are discussed. Finally the major conclusions and an outlook for future investigations are given. <s> BIB005 </s> Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> This paper presents a framework for a three-dimensional (3D) beam pattern design with a load-balanced cell sectorization strategy. First, characteristics of the 3D beam pattern were observed, and convex optimization was used to provide a solution based on the criteria that describe the observations. In addition, a user equipment (UE) determination problem that arose because of cell sectorization was also addressed. By incorporating a beam pattern design with a proposed UE determination scheme, a framework of a backward compatible system was completed. The network performance regarding throughput for cell sectorization with a dedicated beam pattern design was evaluated and the simulation results show that the proposed system is superior compared to the conventional system and other proposed sectorization schemes. <s> BIB006 </s> Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> We consider a heterogeneous cellular network (HetNet) where a macrocell tier with a large antenna array base station (BS) is overlaid with a dense tier of small cells (SCs). We investigate the potential benefits of incorporating a massive MIMO BS in a TDD-based HetNet and we provide analytical expressions for the coverage probability and the area spectral efficiency using stochastic geometry. The duplexing mode in which SCs should operate during uplink macrocell transmissions is optimized. Furthermore, we consider a reverse TDD scheme, in which the massive MIMO BS can estimate the SC interference covariance matrix. Our results suggest that significant throughput improvement can be achieved by exploiting interference nulling and implicit coordination across the tiers due to flexible and asymmetric TDD operation. <s> BIB007 </s> Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> In this paper, we present a vision beyond the conventional Long Term Evolution Fourth Generation (LTE-4G) evolution path and suggest that time division duplexing (TDD) could be a key enabler for a new heterogeneous network architecture with the potential to provide ubiquitous coverage and unprecedented spectral area efficiencies. This architecture is based on a cochannel deployment of macro base stations (BSs) with very large antenna arrays and a secondary tier of small cells (SCs) with a few antennas each. Both tiers employ a TDD protocol in a synchronized fashion. The resulting channel reciprocity enables not only the estimation of large-dimensional channels at the BSs, but also an implicit coordination between both tiers without the need to exchange user data or channel state information (CSI) over the backhaul. In particular, during the uplink (UL), the BSs and SCs can locally estimate the dominant interference sub-space. This knowledge can be leveraged for downlink (DL) precoding to reduce intra- and inter-tier interference. In other words, the BSs and SCs “sacrifice” some of their degrees of freedom for interference rejection. Our simulation results demonstrate that the proposed architecture and precoding scheme can achieve a very attractive rate region compared to several baseline scenarios. For example, with 100 antennas at each BS and four antennas at each SC, we observe an aggregate area throughput of 7.63 Gb/s/km2 (DL) and 8.93 Gb/s/km2 (UL) on a 20 MHz band shared by about 100 mobile devices. <s> BIB008 </s> Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> We propose a time division duplex (TDD) based network architecture where a macrocell tier with a “massive” multiple-input multiple-output (MIMO) base station (BS) is overlaid with a dense tier of small cells (SCs). In this context, the TDD protocol and the resulting channel reciprocity have two compelling advantages. First, a large number of BS antennas can be deployed without incurring a prohibitive overhead for channel training. Second, the BS can estimate the interference covariance matrix from the SC tier which can be leveraged for downlink precoding. In particular, the BS designs its precoding vectors to transmit independent data streams to its users while being orthogonal to the subspace spanned by the strongest interference directions; thereby minimizing the sum interference imposed on the SCs. In other words, the BS “sacrifices” some of its antennas for interference cancellation while the TDD protocol allows for an implicit coordination across both tiers. Simulation results suggest that, given a sufficiently large number of BS antennas, the proposed scheme can significantly improve the sum-rate of the SC tier at the price of a small macro performance loss. <s> BIB009 </s> Survey of Large-Scale MIMO Systems <s> A. Inter-Cell Interference Coordination <s> This paper compares two distinct downlink multicell interference mitigation techniques for wireless cellular networks: large-scale (LS) multiple-input multiple-output (MIMO) and network MIMO. The considered cellular network operates in a time-division duplex (TDD) fashion and includes non-overlapping cooperating clusters, where each cluster comprises B base-stations (BSs), each equipped with multiple antennas, and schedules multiple single-antenna users. In the LS-MIMO system, each BS is equipped with BM antennas, serving its K scheduled users using zero-forcing (ZF) beamforming, while sacrificing its excess number of spatial degrees of freedom (DoF) using interference coordination to prevent causing interference to the other K (B - 1) users within the cooperating cluster. In the network MIMO system, although each BS is equipped with M antennas, the intra-cluster interference cancellation is enabled by data and channel state information sharing across the cooperating BSs and joint downlink transmission to BK users via ZF beamforming. Accounting for uplink-downlink channel reciprocity provided by TDD and invoking the orthogonality principle of ZF beamforming, respectively, the channel acquisition overhead in each cluster and the number of spatial DoF per user are identical in both systems. Therefore, it is not obvious whether one system is superior to the other from the performance point of view. Building upon the channel distribution functions in the two systems and adopting tools from stochastic orders, this paper shows that in fact an LS-MIMO system provides considerably better performance than a network MIMO system. Thus, given the likely lower cost of adding excess number of antennas, LS-MIMO could be a preferred multicell coordination approach for interference mitigation. <s> BIB010

Cellular communication systems suffer from ICI at the cell boundaries, especially when all the channels are fully reused in adjacent cells. As a result, interference mitigation and coordination techniques are needed for alleviating ICI so as to well support frequency reuse. Here, we focus only on static or semi-static ICIC approaches for LS-MIMO systems in different network deployments. Since dynamic ICIC can be regarded as some kind of multi-cell scheduling schemes, it is left to be discussed later. 1) Homogeneous Networks: ICIC techniques, such as Fractional Frequency Reuse (FFR) BIB004 and Soft Frequency Reuse (SFR) BIB001 , have been widely investigated in the context of efficient radio resource management in multi-cell environments in an attempt to coordinate co-channel interference, resulting in improved cell-edge coverage, cell edge data rates and area BE. A large-scale AA provides additional spatial DoF. Therefore, ICIC for LS-MIMO systems is able to exploit the spatial DoF for mitigating ICI by nulling certain spatial direction to the neighboring cell . In an LS-MIMO system, each eNB is equipped with a huge number of antennas, serving its scheduled UEs with beamforming, while trading off its excess DoF against coordinating the interference to other cells within a cluster. Compared with network MIMO, LS-MIMO is preferred to because of its low costs of deploying an excessive number of antennas at the cell site BIB002 . Under the assumption of the same number of DoF per UE and same amount of channel estimation overhead, LS-MIMO with spatial interference coordination outperforms network MIMO BIB010 . The 3D MIMO system, one of LS-MIMO systems, has the capability to dynamically adapt the shape of the vertical beamforming pattern to the UEs at different locations. In other words, the UEs at the cell center and cell edge are covered by different vertical beamforming patterns with specific downtilt such that the received signal power for each UE is maximized. Then, cell sectorization in the 3D MIMO system can be carried out not only along the horizontal but also the vertical axis, which results in increased system throughput BIB006 . However, the ICI problem becomes more complicated with much more sectors per cell. Therefore, it is not straightforward whether the overall BE performance as well as the cell edge UE can be improved. The preliminary study in BIB005 shows that dynamic vertical beam pattern adaptation can provide BE performance gains even with either simplified or suboptimum approaches. Meanwhile, there exists some work on coordinated vertical beamforming with well-known ICIC schemes applied in LTE such as FFR BIB003 . In the literature, there is a lack of comprehensive studies on this issue to date. 2) Heterogeneous Networks: The HetNet is an attractive means of increasing achievable network capacity and of enhancing the coverage area and/or QoE. In a HetNet, small cells as a tier are capable of providing hotspot capacity enhancements, whereas macro cells as another tier are responsible for large area coverage in support of high mobility UEs. However, the MeNBs and SeNBs may interfere with each other, if they use the same time-frequency resources without careful coordination. Fortunately, when the MeNBs, or even the SeNBs, are equipped with a large-scale AA, the AA can provide an additional spatial DoF for multiplexing the data of several UEs onto the same time-frequency resource. Furthermore, it can concentrate the radiated energy precisely on the intended UEs, thereby reducing both the intra-and inter-tier interference. LS-MIMO systems are also capable of supporting cooperation in an implicit way between the different tiers in the HetNet for the sake of improving the overall system performance. To satisfy ever increasing data rate demands, a two-tier TDDbased HetNet is introduced in BIB007 , where the macro-cell tier served by the MeNBs equipped with a large-scale AA is overlaid with the small cell tier of single-antenna SeNBs. Making use of explicit benefits of channel reciprocity under the TDD mode, the MeNBs estimate the UL interference covariance matrix characterizing the interference from the overlay small cells, which can be used for DL ZF based TPC to reduce the interference to the SUEs. The MeNBs with LS-MIMO can significantly improve the BE of small cells at the expense of a moderate loss of the macro cell performance. Additionally, the SeNBs can also be equipped with multiple antennas if needed BIB008 . Recently, the so-called Reversed TDD (RTDD) protocol has been proposed for the HetNet BIB009 . In the RTDD protocol as shown in Fig. 15 , the sequence of the UL and DL transmission periods in one of the tiers is reversed to the other. For example, in Slot 1, while the MeNB transmits the signals to MUEs in the DL, the SeNB received the signals from SUEs in the UL, and vice versa. In the traditional TDD protocol, the MeNBs and SUEs interfere with each other, and so do the SeNBs and MUEs. The channels between the eNBs and UEs potentially fluctuate rapidly thanks to UE mobility. Therefore, less interference samples are available for approximating the time-averaged interference covariance, and hence the resultant estimation errors may degrade the attainable system performance. However, the interference scenario in the HetNet is different if the RTDD protocol is applied, where the MeNB and SeNB interfere with each other, and so do the MUEs and SUEs. Since both the MeNB and SeNB are fixed in location, the interference between them are quasi-static. Hence, the estimated covariance of the channels between them is not sensitive to instantaneous channel variation. Moreover, a massive AA can be deployed at the MeNB and even at the SeNBs so that the interference between the eNBs can be nearly eliminated by narrow beamforming. Meanwhile, the interference between UEs is usually not very serious in most cases because of the low transmission power of UEs. As a result, the RTDD protocol is more suitable for the HetNet with LS-MIMO, since it readily lends itself to cooperative interference cancellation.

Survey of Large-Scale MIMO Systems <s> B. Scheduling <s> Wireless data traffic is expected to grow over the next few years and the technologies that will provide data services are still being debated. One possibility is to use multiple antennas at base stations and terminals to get very high spectral efficiencies in rich scattering environments. Such multiple-input/multiple-output (MIMO) channels can then be used in conjunction with scheduling and rate-feedback algorithms to further increase channel throughput. This paper provides an analysis of the expected gains due to scheduling and bits needed for rate feedback. Our analysis requires an accurate approximation of the distribution of the MIMO channel mutual information. Because the exact distribution of the mutual information in a Rayleigh-fading environment is difficult to analyze, we prove a central limit theorem for MIMO channels with a large number of antennas. While the growth in average mutual information (capacity) of a MIMO channel with the number of antennas is well understood, it turns out that the variance of the mutual information can grow very slowly or even shrink as the number of antennas grows. We discuss implications of this "channel-hardening" result for data and voice services, scheduling, and rate feedback. We also briefly discuss the implications when shadow fading effects are included. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> B. Scheduling <s> We consider schemes for joint scheduling training and downlink MU-MIMO in multi-cell deployments that can yield high cell and cell-edge throughputs with tens of antennas per base-station and single antenna terminals. TDD schemes are considered as they are able to support MU-MIMO over large antenna arrays despite the inherent training dimensionality bottleneck. The proposed schemes operate by dividing users into bins based on their relative location (pathloss) to different stations. MU-MIMO is then performed by splitting transmission resources among such bins, and optimizing the MU-MIMO design on a per-bin basis. We demonstrate the viability of these schemes in the context of a one-dimensional uniform cellular topology and show that high cell and cell-edge rates can be expected, provided the palette of per-bin signaling options is sufficiently large. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> B. Scheduling <s> Future generation cellular networks are expected to provide ubiquitous broadband access to a continuously growing number of mobile users. In this context, LTE systems represent an important milestone towards the so called 4G cellular networks. A key feature of LTE is the adoption of advanced Radio Resource Management procedures in order to increase the system performance up to the Shannon limit. Packet scheduling mechanisms, in particular, play a fundamental role, because they are responsible for choosing, with fine time and frequency resolutions, how to distribute radio resources among different stations, taking into account channel condition and QoS requirements. This goal should be accomplished by providing, at the same time, an optimal trade-off between spectral efficiency and fairness. In this context, this paper provides an overview on the key issues that arise in the design of a resource allocation algorithm for LTE networks. It is intended for a wide range of readers as it covers the topic from basics to advanced aspects. The downlink channel under frequency division duplex configuration is considered as object of our study, but most of the considerations are valid for other configurations as well. Moreover, a survey on the most recent techniques is reported, including a classification of the different approaches presented in literature. Performance comparisons of the most well-known schemes, with particular focus on QoS provisioning capabilities, are also provided for complementing the described concepts. Thus, this survey would be useful for readers interested in learning the basic concepts before going into the details of a particular scheduling strategy, as well as for researchers aiming at deepening more specific aspects. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> B. Scheduling <s> To improve the cellular energy efficiency, without sacrificing quality-of-service (QoS) at the users, the network topology must be densified to enable higher spatial reuse. We analyze a combination of two densification approaches, namely “massive” multiple-input multiple-output (MIMO) base stations and small-cell access points. If the latter are operator-deployed, a spatial soft-cell approach can be taken where the multiple transmitters serve the users by joint non-coherent multiflow beamforming. We minimize the total power consumption (both dynamic emitted power and static hardware power) while satisfying QoS constraints. This problem is proved to have a hidden convexity that enables efficient solution algorithms. Interestingly, the optimal solution promotes exclusive assignment of users to transmitters. Furthermore, we provide promising simulation results showing how the total power consumption can be greatly improved by combining massive MIMO and small cells; this is possible with both optimal and low-complexity beamforming. <s> BIB004

Based on the status of queue, channel quality, QoS requirements and so on, the eNB schedules limited radio resources across the time, frequency and spatial domains among the UEs. Different design objectives, e.g., the affordable complexity, overhead, BE, and fairness, are targeted by a variety of scheduling schemes BIB003 . Usually a good trade-off among all these goals is strived for practical wireless communication systems. Given a large number of UEs and a limited number of antennas, the problem of sum capacity scaling with UE selection has been widely investigated BIB004 , BIB002 . In particular, opportunistic beamforming yields significant gains by exploiting the independence of the UEs' channel fluctuation, which may be conducive to achieving multiple UE diversity. Moreover, in order to evaluate the gains of scheduling, the mutual information of the LS-MIMO system may be modeled as a normal distribution under the assumption of i.i.d. Rayleigh fading channels BIB001 . It turns out that the variance of the channel coefficients grows slowly or even decreases with the number of transmit antennas, but increases with the number of UEs. This conclusion implies that carefully designed robust scheduling schemes would reduce the CSI-feedback rate required. The more UE-specific and eNB-specific channel information is obtained by the scheduler, the more efficient the system can be. Nonetheless, with increased amount of required information, the overhead and computational complexity for channel information may become prohibitive. Due to the channel hardening phenomenon 1 of LS-MIMO systems, the scheduling mechanisms relying on full CSI, including large-scale and small-scale prorogation characteristics, are not cost-efficient, because no significant performance gains can be achieved compared to those requiring only partial CSI including the path loss and shadow fading. So, more attention has been paid to scheduling schemes with partial CSI instead of full CSI in LS-MIMO systems, both of which are presented for comparison in this section.

Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> We consider schemes for joint scheduling training and downlink MU-MIMO in multi-cell deployments that can yield high cell and cell-edge throughputs with tens of antennas per base-station and single antenna terminals. TDD schemes are considered as they are able to support MU-MIMO over large antenna arrays despite the inherent training dimensionality bottleneck. The proposed schemes operate by dividing users into bins based on their relative location (pathloss) to different stations. MU-MIMO is then performed by splitting transmission resources among such bins, and optimizing the MU-MIMO design on a per-bin basis. We demonstrate the viability of these schemes in the context of a one-dimensional uniform cellular topology and show that high cell and cell-edge rates can be expected, provided the palette of per-bin signaling options is sufficiently large. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> We consider the downlink of a multicell system with multiantenna base stations and single-antenna user terminals, arbitrary base station cooperation clusters, distance-dependent propagation pathloss, and general “fairness” requirements. Base stations in the same cooperation cluster employ joint transmission with linear zero-forcing beamforming, subject to sum or per-base station power constraints. Intercluster interference is treated as noise at the user terminals. Analytic expressions for the system spectral efficiency are found in the large-system limit where both the numbers of users and antennas per base station tend to infinity with a given ratio. In particular, for the per-base station power constraint, we find new results in random matrix theory, yielding the squared Frobenius norm of submatrices of the Moore-Penrose pseudo-inverse for the structured non-i.i.d. channel matrix resulting from the cooperation cluster, user distribution, and path-loss coefficients. The analysis is extended to the case of nonideal Channel State Information at the Transmitters obtained through explicit downlink channel training and uplink feedback. Specifically, our results illuminate the trade-off between the benefit of a larger number of cooperating antennas and the cost of estimating higher-dimensional channel vectors. Furthermore, our analysis leads to a new simplified downlink scheduling scheme that preselects the users according to probabilities obtained from the large-system results, depending on the desired fairness criterion. The proposed scheme performs close to the optimal (finite-dimensional) opportunistic user selection while requiring significantly less channel state feedback, since only a small fraction of preselected users must feed back their channel state information. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> This paper considers base station (BS) cooperation in the form of coordinated beamforming, focusing on min-max fairness in the power usage subject to target SINR constraints. We show that the optimal beamforming strategies have an interesting nested zero-forcing structure. We then consider the asymptotic regime where the number of antennas at each BS and the number of users in each cell both grow large with their ratio tending to a finite constant. The limiting optimal beamformer structure is characterized in terms of the solution of a convex optimization problem. The dimensionality of this optimization problem is constant1, as opposed to the problem of solving for the exactly optimal beamformer2, and only knowledge of statistics is required to solve it. These asymptotic results provide insights into the average performance, as well as simple but efficient strategies for the finite system case. In particular, the optimal beamforming strategy from the large systems analysis only requires the base stations to have local instantaneous channel state information; the remaining parameters of the beamformer can be calculated using channel statistics which can easily be shared amongst the base stations. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> In this paper, resource allocation for energy-efficient communication in an orthogonal frequency division multiple access (OFDMA) downlink network with a large number of transmit antennas is studied. The considered problem is modeled as a non-convex optimization problem which takes into account the circuit power consumption, imperfect channel state information at the transmitter (CSIT), and different quality of service (QoS) requirements including a minimum required data rate and a maximum tolerable channel outage probability. The power allocation, data rate adaptation, antenna allocation, and subcarrier allocation policies are optimized for maximization of the energy efficiency of data transmission (bit/Joule delivered to the users). By exploiting the properties of fractional programming, the resulting non-convex optimization problem in fractional form is transformed into an equivalent optimization problem in subtractive form, which leads to an efficient iterative resource allocation algorithm. In each iteration, the objective function is lower bounded by a concave function which can be maximized by using dual decomposition. Simulation results illustrate that the proposed iterative resource allocation algorithm converges in a small number of iterations and demonstrate the trade-off between energy efficiency and the number of transmit antennas. <s> BIB004 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> In this paper, we discuss the joint precoding with finite rate feedback in the so-called network multiple-input multiple-output (MIMO) where the TXs share the knowledge of the data symbols to be transmitted. We introduce a distributed channel state information (DCSI) model where each TX has its own local estimate of the overall multiuser MIMO channel and must make a precoding decision solely based on the available local CSI. We refer to this channel as the DCSI-MIMO channel and the precoding problem as distributed precoding. We extend to the DCSI setting the work from Jindal in 2006 for the conventional MIMO broadcast channel (BC) in which the number of degrees of freedom (DoFs) achieved by zero forcing (ZF) was derived as a function of the scaling in the logarithm of the signal-to-noise ratio of the number of quantizing bits. Particularly, we show the seemingly pessimistic result that the number of DoFs at each user is limited by the worst CSI across all users and across all TXs. This is in contrast to the conventional MIMO BC where the number of DoFs at one user is solely dependent on the quality of the estimation of his own feedback. Consequently, we provide precoding schemes improving on the achieved number of DoFs. For the two-user case, the derived novel precoder achieves a number of DoFs limited by the best CSI accuracy across the TXs instead of the worst with conventional ZF. We also advocate the use of hierarchical quantization of the CSI, for which we show that considerable gains are possible. Finally, we use the previous analysis to derive the DoFs optimal allocation of the feedback bits to the various TXs under a constraint on the size of the aggregate feedback in the network, in the case where conventional ZF is used. <s> BIB005 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> This paper studies a joint optimization of beamforming and power control to enforce egalitarian fairness for all users in a multicell setting. In a single cell downlink with instantaneous channel information, the well-known uplink-downlink duality via Lagrange interpretation helps in algorithm design. By leveraging nonlinear Perron-Frobenius theory, we establish the network duality for the max-min weighted SINR problem in a general multicell network with multiple serving users per cell, whose generic feature reduces to two special scenarios: the single cell multiuser scenario and the multiuser ad hoc network scenario. The network duality can be observed through Perron-Frobenius characterization, and holds for both forms of channel information: the instantaneous channel information and statistical channel information. This characterization of network duality further enables the design of distributed algorithms with geometrically fast convergence rate, which are readily applicable to practical implementation. <s> BIB006 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> To improve the cellular energy efficiency, without sacrificing quality-of-service (QoS) at the users, the network topology must be densified to enable higher spatial reuse. We analyze a combination of two densification approaches, namely “massive” multiple-input multiple-output (MIMO) base stations and small-cell access points. If the latter are operator-deployed, a spatial soft-cell approach can be taken where the multiple transmitters serve the users by joint non-coherent multiflow beamforming. We minimize the total power consumption (both dynamic emitted power and static hardware power) while satisfying QoS constraints. This problem is proved to have a hidden convexity that enables efficient solution algorithms. Interestingly, the optimal solution promotes exclusive assignment of users to transmitters. Furthermore, we provide promising simulation results showing how the total power consumption can be greatly improved by combining massive MIMO and small cells; this is possible with both optimal and low-complexity beamforming. <s> BIB007 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> We propose a hierarchical interference mitigation scheme for massive MIMO cellular networks. The MIMO precoder at each base station (BS) is partitioned into an inner precoder and an outer precoder. The inner precoder controls the intra-cell interference and is adaptive to local channel state information (CSI) at each BS (CSIT). The outer precoder controls the inter-cell interference and is adaptive to channel statistics. Such hierarchical precoding structure reduces the number of pilot symbols required for CSI estimation in massive MIMO downlink and is robust to the backhaul latency. We study joint optimization of the outer precoders, the user selection, and the power allocation to maximize a general concave utility which has no closed-form expression. We first apply random matrix theory to obtain an approximated problem with closed-form objective. We show that the solution of the approximated problem is asymptotically optimal with respect to the original problem as the number of antennas per BS grows large. Then using the hidden convexity of the problem, we propose an iterative algorithm to find the optimal solution for the approximated problem. We also obtain a low complexity algorithm with provable convergence. Simulations show that the proposed design has significant gain over various state-of-the-art baselines. <s> BIB008 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> This paper considers base station (BS) cooperation in the form of coordinated beamforming, focusing on min-max fairness in the power usage subject to target SINR constraints at each single-antenna user. We show that the optimal beamforming strategies have an interesting nested zero-forcing structure. In the asymptotic regime where the number of antennas at each BS and that of users in each cell both grow large with their ratio tending to a finite constant, the dimensionality of the optimization problem that needs to be solved is greatly reduced, and only knowledge of statistics is required to solve it. The optimal solution is characterized in general, and an algorithm is proposed that converges to the optimal transmit parameters, for feasible SINR targets. For the two cell case, a simple single parameter characterization is obtained. These asymptotic results provide insights into the average performance, as well as simple but efficient beamforming strategies for the finite system case. In particular, the asymptotically optimal beamformers only require the BSs to have local instantaneous channel state information; the remaining parameters of the beamformers can be calculated using channel statistics, thereby reducing the channel state information estimation and signaling overhead. <s> BIB009 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> This paper studies joint beamforming and power control in a coordinated multicell downlink system that serves multiple users per cell to maximize the minimum weighted signal-to-interference-plus-noise ratio. The optimal solution and distributed algorithm with geometrically fast convergence rate are derived by employing the nonlinear Perron-Frobenius theory and the multicell network duality. The iterative algorithm, though operating in a distributed manner, still requires instantaneous power update within the coordinated cluster through the backhaul. The backhaul information exchange and message passing may become prohibitive with increasing number of transmit antennas and increasing number of users. In order to derive asymptotically optimal solution, random matrix theory is leveraged to design a distributed algorithm that only requires statistical information. The advantage of our approach is that there is no instantaneous power update through backhaul. Moreover, by using nonlinear Perron-Frobenius theory and random matrix theory, an effective primal network and an effective dual network are proposed to characterize and interpret the asymptotic solution. <s> BIB010 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> Joint Spatial Division and Multiplexing (JSDM) is a downlink multiuser MIMO scheme recently proposed by the authors in order to enable “massive MIMO” gains and simplified system operations for Frequency Division Duplexing (FDD) systems. The key idea lies in partitioning the users into groups with approximately similar channel covariance eigenvectors and serving these groups by using two-stage downlink precoding scheme obtained as the concatenation of a pre-beamforming matrix, that depends only on the channel second-order statistics, with a multiuser MIMO linear precoding matrix, which is a function of the effective channels including pre-beamforming. The role of pre-beamforming is to reduce the dimensionality of the effective channel by exploiting the near-orthogonality of the eigenspaces of the channel covariances of the different user groups. This paper is an extension of our initial work on JSDM, and addresses some important practical issues. First, we focus on the regime of finite number of antennas and large number of users and show that JSDM with simple opportunistic user selection is able to achieve the same scaling law of the system capacity with full channel state information. Next, we consider the large-system regime (both antennas and users growing large) and propose a simple scheme for user grouping in a realistic setting where users have different angles of arrival and angular spreads. Finally, we propose a low-overhead probabilistic scheduling algorithm that selects the users at random with probabilities derived from large-system random matrix analysis. Since only the pre-selected users are required to feedback their channel state information, the proposed scheme realizes important savings in the CSIT feedback. <s> BIB011 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> The massive multiple-input multiple-output (MIMO) system has drawn increasing attention recently as it is expected to boost the system throughput and result in lower costs. Previous studies mainly focus on time division duplexing (TDD) systems, which are more amenable to practical implementations due to channel reciprocity. However, there are many frequency division duplexing (FDD) systems deployed worldwide. Consequently, it is of great importance to investigate the design and performance of FDD massive MIMO systems. To reduce the overhead of channel estimation in FDD systems, a two-stage precoding scheme was recently proposed to decompose the precoding procedure into intergroup precoding and intragroup precoding. The problem of user grouping and scheduling thus arises. In this paper, we first propose three novel similarity measures for user grouping based on weighted likelihood, subspace projection, and Fubini-Study, respectively, as well as two novel clustering methods, including hierarchical and K-medoids clustering. We then propose a dynamic user scheduling scheme to further enhance the system throughput once the user groups are formed. The load balancing problem is considered when few users are active and solved with an effective algorithm. The efficacy of the proposed schemes are validated with theoretical analysis and simulations. <s> BIB012 </s> Survey of Large-Scale MIMO Systems <s> 1) Full CSI-Based Scheduling: <s> Coordinated scheduling/beamforming (CS/CB) substantially mitigates the intercell interference (ICI), hence increasing the cell-edge throughput on the downlink (DL) of coordinated multipoint (CoMP) systems. To maximize the DL throughput, the cooperating base stations (BSs) jointly select the best set of users for DL scheduling and then jointly design a set of beamforming (BF) vectors to approach the throughput limit. However, finding the optimal BF vectors requires an exhaustive search and substantial channel state information (CSI) feedback, hence resulting in high algorithmic complexity and heavy uplink traffic load. Hence, we conceive a new cross-layer algorithm to achieve high performance at a lower feedback amount and at lower algorithmic complexity. Based on the fact that different BSs usually have different traffic loads, we divide the BSs into two different types, i.e., the master BSs (MBSs) and the slave BSs (SBSs), where MBSs have a higher transmission priority than SBSs. The scheduling relies on an interference threshold, whereas our robust BF scheme exploits both the channel direction information (CDI), which is quantized using the technique of limited feedback, and the channel quality information (CQI), which is assumed to be fed back accurately. Our numerical results show that the proposed algorithm does not lose much performance compared with that achieved by an exhaustive search, whereas the algorithmic complexity is as low as that of the solutions operating without CoMP. <s> BIB013

In order to achieve the optimal network performance, the scheduler has to acquire the full and accurate knowledge of all the channels. This knowledge can be exploited to minimize the total power consumption, while satisfying the QoS and power constraints at the eNBs in the HetNet with LS-MIMO BIB007 . Toward this end, a spatial softcell approach can be taken, where each UE is dynamically assigned to access the optimal nodes, i.e., LS-MIMO MeNB, conventional SeNB or both. If the system assigns a UE to an MeNB and an SeNB at the same time, multiple transmitters serve the UE through joint non-coherent multi-flow beamforming. Also, the total EE can be further improved by applying a low-complexity efficient algorithm, which exploits the hidden convexity in this problem. However, the scheduler has to know the full instantaneous CSI of all the UEs, which incurs both potentially overwhelming estimation overheads and excessive computational complexity due to the large number of AEs and UEs in a practical system. The costs of the scheduling scheme based on full CSI are likely to outweigh the gains. 2) Partial CSI-Based Scheduling: Fortunately, scheduling schemes with partial CSI in LS-MIMO systems are capable of reducing implementation complexity and overhead with an acceptable performance loss. • Single-cell scheduling: Under the single-cell scenarios, resource allocation for energy-efficient OFDMA systems with a large number of AEs was studied in BIB004 . Taking into account the associated circuit power dissipation, the imperfect CSI at the Transmitter (CSIT) and QoS requirements, the resources are assigned with the objective of maximizing EE. The considered parameters include the subcarrier allocation, power allocation, antenna allocation and available data rates. It is demonstrated that even though the use of a large number of transmit antennas reduces the multipath-induced fluctuation of each channel, the system performance can still benefit from the different path losses and shadow fading conditions of different UEs. Furthermore, the scheduling scheme can be updated periodically, because the path loss and shadow fading parameters vary slowly, depending on UE mobility. The reduced-complexity probabilistic scheduling algorithms have been proposed in BIB001 , BIB011 , BIB002 , where the number of AEs is quite large. Firstly, different UEs are clustered into groups based on their channel covariance matrices. Then, the UEs in each group are pre-selected randomly based on the group-specific probabilities derived from the whole system. Only the pre-selected UEs are required to transmit their training signals or CSI feedback, leading to markedly reduced overheads and complexity. Moreover, different measures have been conceived both for UE grouping and for scheduling in a FDD based LS-MIMO system based on two-stage precoding, namely on inter-group precoding and intra-group precoding, which are capable of reducing the channel estimation overhead while guaranteeing fairness to the UEs BIB012 . • Multi-cell scheduling: Unlike the single-cell scenario, both inter-cell and intra-cell interferences have to be taken into account under the multi-cell scenario. Although fully multi-cell scheduling schemes, like the coordinated joint processing in network MIMO, can achieve large performance gains, they are not practical due to prohibitive costs. That is, not only the full CSI but also data streams intended for different mobile users at different cells need to be shared among the eNBs BIB005 . Instead, another feasible solution is to improve the overall network performance by allowing beamforming vectors from different eNBs to be coordinated for the sake of implementation BIB013 . The main objective of applying coordinated beamforming in LS-MIMOs is to improve the overall system performance, while reducing the coordination overhead. There are two popular coordinated beamforming schemes, namely the hierarchical BIB008 and the nested structure BIB003 , BIB006 . Explicitly, the BF relying on the hierarchical structure BIB008 at each eNB consists of an inner precoder and an outer precoder. The inner precoder supports the transmission of data to the serving UEs by exploiting the knowledge of the time-variant CSI. Meanwhile, the outer precoder exploits the remaining spatial DoF for mitigating the ICI by relying only on the knowledge of the average CSI BIB008 . This structure requires only a modest amount of backhaul overhead and pilots for CSI estimation. Moreover, only the knowledge of the channel's spatial correlation matrices is needed for the BF-weight optimization, which is insensitive to the backhaul latency. By contrast, in the BF associated with the nested structure, the optimal strategy can be found recursively, where the BF weight optimization Objective Function (OF) may be based on the fairness in power usage subject to satisfying the target SINR constraints BIB003 , BIB009 . Another nested structure, whose OF focuses on Maximizing the Minimum (MAX-MIN) weighted SINR among UEs was proposed in BIB006 , BIB010 . Unlike the hierarchical structure of BIB008 , the optimal precoder of the nested structure BIB003 , BIB006 is found as the solution of a joint optimization problem, which aims for striking a trade-off between providing a high SINR for the intra-cell UEs and mitigating the ICI. For instance, when this BF is applied in an LS-MIMO system, all the eNBs are divided into two groups, 1) Selfish eNBs, whose UE SINRs are relative low; and 2) altruistic eNBs, whose UEs SINRs are relative high. The altruistic group may be empty; or it may use zeroforcing for eliminating the interference by imposing it on the selfish group. If the optimal ZF beamforming scheme is used in the altruistic group, each eNB in this group has to transmit less power than that in the selfish group. The precoder at each eNB uses the optimal BF parameters along with its own instantaneous CSI. Additionally, only the average CSI has to be exchanged among the eNBs. On the other hand, different metrics can be used to indicate different system performances. Thus, it is crucial to choose an objective with an appropriate metric function for the multi-cell scheduling schemes. There are several kinds of objectives in terms of either efficiency or fairness when multi-cell scheduling is applied in LS-MIMO systems, e.g., a) Minimizing the Maximum (MIN-MAX) the fairness in power consumption subject to certain SINR constraints BIB003 , BIB009 : It aims to maximize the overall EE as a high priority by adjusting its coordinated beamforming scheme. A efficient solution can be obtained through Lagrange duality and random matrix theory; b) MAX-MIN SINR subject to certain sum-rate constraints BIB006 , BIB010 : It enforces the overall system fairness by guaranteeing each UE's promised SINR. In the case of non-convex optimization problems, the optimal scheme may be derived by using nonlinear Perron-Frobenius theory; c) Maximizing the weighted sum rate subject to some eNBs power consumption constraints: This objective can be viewed as a combination of MAX-MIN fairness and maximum sum-rate. Moreover, efficient schemes can be obtained through hidden convexity and random matrix theory BIB008 .

Survey of Large-Scale MIMO Systems <s> C. Green LS-MIMO <s> A multiplicity of autonomous terminals simultaneously transmits data streams to a compact array of antennas. The array uses imperfect channel-state information derived from transmitted pilots to extract the individual data streams. The power radiated by the terminals can be made inversely proportional to the square-root of the number of base station antennas with no reduction in performance. In contrast if perfect channel-state information were available the power could be made inversely proportional to the number of antennas. Lower capacity bounds for maximum-ratio combining (MRC), zero-forcing (ZF) and minimum mean-square error (MMSE) detection are derived. An MRC receiver normally performs worse than ZF and MMSE. However as power levels are reduced, the cross-talk introduced by the inferior maximum-ratio receiver eventually falls below the noise level and this simple receiver becomes a viable option. The tradeoff between the energy efficiency (as measured in bits/J) and spectral efficiency (as measured in bits/channel use/terminal) is quantified for a channel model that includes small-scale fading but not large-scale fading. It is shown that the use of moderately large antenna arrays can improve the spectral and energy efficiency with orders of magnitude compared to a single-antenna system. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> C. Green LS-MIMO <s> Large-Scale Antenna Systems (LSAS) is a form of multi-user MIMO technology in which unprecedented numbers of antennas serve a significantly smaller number of autonomous terminals. We compare the two most prominent linear pre-coders, conjugate beamforming and zero-forcing, with respect to net spectral-efficiency and radiated energy-efficiency in a simplified single-cell scenario where propagation is governed by independent Rayleigh fading, and where channel-state information (CSI) acquisition and data transmission are both performed during a short coherence interval. An effective-noise analysis of the pre-coded forward channel yields explicit lower bounds on net capacity which account for CSI acquisition overhead and errors as well as the sub-optimality of the pre-coders. In turn the bounds generate trade-off curves between radiated energy-efficiency and net spectral-efficiency. For high spectral-efficiency and low energy-efficiency zero-forcing outperforms conjugate beamforming, while at low spectral-efficiency and high energy-efficiency the opposite holds. Surprisingly, in an optimized system, the total LSAS-critical computational burden of conjugate beamforming may be greater than that of zero-forcing. Conjugate beamforming may still be preferable to zero-forcing because of its greater robustness, and because conjugate beamforming lends itself to a de-centralized architecture and de-centralized signal processing. <s> BIB002

With increasing carbon emissions imposed by the excessive power consumption of wireless communications networks, EE has become a significant performance metric of evaluating future wireless networks. However, the evaluation of EE in LS-MIMO systems in the literature focuses primarily on the EE in the single-cell environment. Furthermore, only the radiated power of the PAs is considered, while ignoring the power dissipation of signal processing circuit and the efficiency of the PAs BIB001 , BIB002 . Therefore, the realistic EE performance of LS-MIMO systems needs to be analyzed by considering both the radiated power and circuit power in multi-cell scenarios.

Survey of Large-Scale MIMO Systems <s> F. Complexity Versus Performance <s> This paper considers multicell multiuser MIMO systems with very large antenna arrays at the base station. We propose an eigenvalue-decomposition-based approach to channel estimation, that estimates the channel blindly from the received data. The approach exploits the asymptotic orthogonality of the channel vectors in very large MIMO systems. We show that the channel to each user can be estimated from the covariance matrix of the received signals, up to a remaining scalar multiplicative ambiguity. A short training sequence is required to resolve this ambiguity. Furthermore, to improve the performance of our approach, we combine it with the iterative least-square with projection (ILSP) algorithm. Numerical results verify the effectiveness of our channel estimation approach. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> F. Complexity Versus Performance <s> A subspace projection to improve channel estimation in massive multi-antenna systems is proposed and analyzed. Together with power-controlled hand-off, it can mitigate the pilot contamination problem without the need for coordination among cells. The proposed method is blind in the sense that it does not require pilot data to find the appropriate subspace. It is based on the theory of large random matrices that predicts that the eigenvalue spectra of large sample covariance matrices can asymptotically decompose into disjoint bulks as the matrix size grows large. Random matrix and free probability theory are utilized to predict under which system parameters such a bulk decomposition takes place. Simulation results are provided to confirm that the proposed method outperforms conventional linear channel estimation if bulk separation occurs. <s> BIB002

An LS-MIMO system often entails considerably complex signal processing. Hence, substantial efforts have been devoted to streamline and optimize both signal processing algorithms and their implementations. However, low-complexity algorithms generally decrease the performance of LS-MIMO. For example, having more accurate CSI leads to improved performance at the expense of increased processing complexity. When low-complexity linear channel estimation is adopted, the attainable performance is constrained by the pilot contamination phenomenon, especially in multi-cell scenarios. However, pilot contamination can be overcome by using complex channel estimation algorithms BIB001 , BIB002 . There exists a tradeoff between the complexity of channel prediction as well as the TPC design and the experienced channel, when the UEs move around in a cell. Therefore, simple but efficient algorithms have to be conceived for channel estimation, channel prediction, TPC and detection.

Survey of Large-Scale MIMO Systems <s> G. Effects of Non-Ideal Hardware <s> Multiuser multiple-input–multiple-output (MIMO) cellular systems with an excess of base station (BS) antennas (Massive MIMO) offer unprecedented multiplexing gains and radiated energy efficiency. Oscillator phase noise is introduced in the transmitter and receiver radio frequency chains and severely degrades the performance of communication systems. We study the effect of oscillator phase noise in frequency-selective Massive MIMO systems with imperfect channel state information. In particular, we consider two distinct operation modes, namely, when the phase noise processes at the $M$ BS antennas are identical (synchronous operation) and when they are independent (nonsynchronous operation) . We analyze a linear and low-complexity time-reversal maximum-ratio combining reception strategy. For both operation modes, we derive a lower bound on the sum-capacity, and we compare their performance. Based on the derived achievable sum-rates, we show that with the proposed receive processing, an $O(\sqrt{M} ) $ array gain is achievable. Due to the phase noise drift, the estimated effective channel becomes progressively outdated. Therefore, phase noise effectively limits the length of the interval used for data transmission and the number of scheduled users. The derived achievable rates provide insights into the optimum choice of the data interval length and the number of scheduled users. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> G. Effects of Non-Ideal Hardware <s> The use of large-scale antenna arrays can bring substantial improvements in energy and/or spectral efficiency to wireless systems due to the greatly improved spatial resolution and array gain. Recent works in the field of massive multiple-input multiple-output (MIMO) show that the user channels decorrelate when the number of antennas at the base stations (BSs) increases, thus strong signal gains are achievable with little interuser interference. Since these results rely on asymptotics, it is important to investigate whether the conventional system models are reasonable in this asymptotic regime. This paper considers a new system model that incorporates general transceiver hardware impairments at both the BSs (equipped with large antenna arrays) and the single-antenna user equipments (UEs). As opposed to the conventional case of ideal hardware, we show that hardware impairments create finite ceilings on the channel estimation accuracy and on the downlink/uplink capacity of each UE. Surprisingly, the capacity is mainly limited by the hardware at the UE, while the impact of impairments in the large-scale arrays vanishes asymptotically and interuser interference (in particular, pilot contamination) becomes negligible. Furthermore, we prove that the huge degrees of freedom offered by massive MIMO can be used to reduce the transmit power and/or to tolerate larger hardware impairments, which allows for the use of inexpensive and energy-efficient antenna elements. <s> BIB002

LS-MIMO relies mainly on the law of large numbers to average out noise, channel fading characteristics, etc. However, in practice, low-cost imperfect components are employed to implement the LS-MIMO system. Therefore, the imperfections of non-ideal hardware, such as the non-linearities of the amplifier, I/Q imbalance, A/D and D/A nonlinearities, should be taken into account BIB001 , BIB002 in practical LS-MIMO systems. Some existing studies have been reported on evaluating the effects of non-ideal hardware. However, more efforts are required for designing efficient algorithms to mitigate these non-ideal factors.

Survey of Large-Scale MIMO Systems <s> H. LS-MIMO in the mm-Wave Band <s> This article presents empirically-based largescale propagation path loss models for fifthgeneration cellular network planning in the millimeter-wave spectrum, based on real-world measurements at 28 GHz and 38 GHz in New York City and Austin, Texas, respectively. We consider industry-standard path loss models used for today’s microwave bands, and modify them to fit the propagation data measured in these millimeter-wave bands for cellular planning. Network simulations with the proposed models using a commercial planning tool show that roughly three times more base stations are required to accommodate 5G networks (cell radii up to 200 m) compared to existing 3G and 4G systems (cell radii of 500 m to 1 km) when performing path loss simulations based on arbitrary pointing angles of directional antennas. However, when directional antennas are pointed in the single best directions at the base station and mobile, coverage range is substantially improved with little increase in interference, thereby reducing the required number of 5G base stations. Capacity gains for random pointing angles are shown to be 20 times greater than today’s fourth-generation Long Term Evolution networks, and can be further improved when using directional antennas pointed in the strongest transmit and receive directions with the help of beam combining techniques. <s> BIB001 </s> Survey of Large-Scale MIMO Systems <s> H. LS-MIMO in the mm-Wave Band <s> The millimeter-wave (mmWave) band offers the potential for high-bandwidth communication channels in cellular networks. It is not clear, however, whether both high data rates and coverage in terms of signal-to-noise-plus-interference ratio can be achieved in interference-limited mmWave cellular networks due to the differences in propagation conditions and antenna topologies. This article shows that dense mmWave networks can achieve both higher data rates and comparable coverage relative to conventional microwave networks. Sum rate gains can be achieved using more advanced beamforming techniques that allow multiuser transmission. The insights are derived using a new theoretical network model that incorporates key characteristics of mmWave networks. <s> BIB002 </s> Survey of Large-Scale MIMO Systems <s> H. LS-MIMO in the mm-Wave Band <s> This article discusses the challenges, benefits and approaches associated with realizing largescale antenna arrays at mmWave frequency bands for future 5G cellular devices. Key design considerations are investigated to deduce a novel and practical phased array antenna solution operating at 28 GHz with near spherical coverage. The approach is further evolved into a first-of- a-kind cellular phone prototype equipped with mmWave 5G antenna arrays consisting of a total of 32 low-profile antenna elements. Indoor measurements are carried out using the presented prototype to characterize the proposed mmWave antenna system using 16-QAM modulated signals with 27.925 GHz carrier frequency. The biological implications due to the absorbed electromagnetic waves when using mmWave cellular devices are studied and compared in detail with those of 3/4G cellular devices. <s> BIB003 </s> Survey of Large-Scale MIMO Systems <s> H. LS-MIMO in the mm-Wave Band <s> The exponential increase of mobile data traffic requires disrupting approaches for the realization of future 5G systems. In this article, we overview the technologies that will pave the way for a novel cellular architecture that integrates high-data-rate access and backhaul networks based on millimeter-wave frequencies (57-66, 71-76, and 81-86 GHz). We evaluate the feasibility of short- and medium-distance links at these frequencies and analyze the requirements from the transceiver architecture and technology, antennas, and modulation scheme points of view. Technical challenges are discussed, and design options highlighted; finally, a performance evaluation quantifies the benefits of millimeter- wave systems with respect to current cellular technologies. <s> BIB004

In the popular frequency bands below 2 GHz, the physical size of a large-scale AA is excessive to be feasibly installed in a UE, although it may be readily accommodated by the eNB. In order to avoid this problem, LS-MIMO systems are likely to operate in the millimeter (mm)-wave band . For example, in the 60 GHz band, the corresponding wavelength is only 0.5 cm, hence a large-scale AA having λ/2-spaced elements may accommodate as many as 40 AEs within a physical space of 10 cm. However, the channel measurements indicate propagation losses that are severe in the mm-wave band, hence the propagation distance becomes very short, resulting in small cells BIB001 . In order to attain both high data rates and a satisfactory coverage, the ultra-dense deployment of small cells operating in the mm-wave band is one of the potential solutions envisaged for next-generation wireless communications networks BIB002 . On the other hand, another challenge is to realize large-scale AAs in the mm-wave band, because it is hard to design practical phased AAs operating at 28 GHz. Nonetheless, a large-scale AA prototype with up to 32 miniaturized antennas was reported in BIB003 . Apart from the relevant physical layer techniques, the network architecture of LS-MIMOs operating in the mm-wave band has to be studied as well. A cellular architecture relying on mm-wave frequencies has been proposed and evaluated, where the MUEs communicating with the MeNB operate in the microwave band and the SUEs use the mm-wave band BIB004 . Although much attention has been dedicated to the design of LS-MIMO systems in the mm-wave band, there are still many uncertain issues surrounding the practical mm-wave communication systems, such as their hardware design, interference control and resource management.

A survey of of blockchain-based solutions for Energy Industry <s> Conclusion: <s> In this paper we introduce a new decentralized digital currency, called NRGcoin. Prosumers in the smart grid trade locally produced renewable energy using NRGcoins, the value of which is determined on an open currency exchange market. Similar to Bitcoins, this currency offers numerous advantages over fiat currency, but unlike Bitcoins it is generated by injecting energy into the grid, rather than spending energy on computational power. In addition, we propose a novel trading paradigm for buying and selling green energy in the smart grid. Our mechanism achieves demand response by providing incentives to prosumers to balance their production and consumption out of their own self-interest. We study the advantages of our proposed currency over traditional money and environmental instruments, and explore its benefits for all parties in the smart grid. <s> BIB001 </s> A survey of of blockchain-based solutions for Energy Industry <s> Conclusion: <s> The increasing amount of renewable energy sources in the energy system calls for new market approaches to price and distribute the volatile and decentralized generation. Local energy markets, on which consumers and prosumers can trade locally produced renewable generation directly within their community, balance generation and consumption locally in a decentralized approach. We present a comprehensive concept, market design and simulation of a local energy market between 100 residential households. Our approach is based on a distributed information and communication technology, i.e. a private blockchain, which underlines the decentralized nature of local energy markets. Thus, we provide energy prosumers and consumers with a decentralized market platform for trading local energy generation without the need of a central intermediary. Furthermore, we present a preliminary economic evaluation of the market mechanism and a research agenda for the technological evaluation of blockchain technology as the local energy market’s main information and communication technology. <s> BIB002 </s> A survey of of blockchain-based solutions for Energy Industry <s> Conclusion: <s> Electricity is the commonest commodity for most businesses in our world today. The use of electricity has been a breakthrough for the discovery of new technologies and has become the main driving force behind several innovations. With the introduction of smart grid systems, there have been improvements in how utility companies interact with their customers with regards to electricity use. However, since the readings are done via the Internet, there is the tendency for the data to be compromised when it gets into the hands of the wrong people. Moreover, customers mostly do not know why they pay huge amounts and which appliances use more electricity, since they are not privy to the readings. The sovereign blockchain technology, which provides transparency and provenance, is utilized in this paper to mitigate these above mentioned problems. A smart contract, which executes laid down procedures to provide a trust-based system between participants on the network is also implemented. Our system proves very efficient as the user can monitor how the electricity is used, and it also provides a platform where there is no manipulation from either party. <s> BIB003 </s> A survey of of blockchain-based solutions for Energy Industry <s> Conclusion: <s> In this paper, we investigate the use of decentralized blockchain mechanisms for delivering transparent, secure, reliable, and timely energy flexibility, under the form of adaptation of energy demand profiles of Distributed Energy Prosumers, to all the stakeholders involved in the flexibility markets (Distribution System Operators primarily, retailers, aggregators, etc.). In our approach, a blockchain based distributed ledger stores in a tamper proof manner the energy prosumption information collected from Internet of Things smart metering devices, while self-enforcing smart contracts programmatically define the expected energy flexibility at the level of each prosumer, the associated rewards or penalties, and the rules for balancing the energy demand with the energy production at grid level. Consensus based validation will be used for demand response programs validation and to activate the appropriate financial settlement for the flexibility providers. The approach was validated using a prototype implemented in an Ethereum platform using energy consumption and production traces of several buildings from literature data sets. The results show that our blockchain based distributed demand side management can be used for matching energy demand and production at smart grid level, the demand response signal being followed with high accuracy, while the amount of energy flexibility needed for convergence is reduced. <s> BIB004 </s> A survey of of blockchain-based solutions for Energy Industry <s> Conclusion: <s> Climate change represents a serious threat to the health of our planet and imposed a discussion upon energy waste and production. In this paper we propose a smart grid architecture relying on blockchain technology aimed at discouraging the production and distribution of non-renewable energy as the one derived from fossil fuel. Our model relies on a reverse application of a recently introduced attack to the blockchain based on chain forking. Our system involves both a central authority and a number of distributed peers representing the stakeholders of the energy grid. This system preserves those advantages derived from the blockchain and it also address some limitations such as energy waste for mining operations. In addition, the reverse attack we rely on allows to mitigate the behavior of a classic blockchain, which is intrinsecally self-regulated, and to trigger a sort of ethical action which penalizes non-renewable energy producers. Blacklisted stakeholders will be induced to provide their transaction with higher fees in order to preserve the selling rate. <s> BIB005 </s> A survey of of blockchain-based solutions for Energy Industry <s> Conclusion: <s> Implementing blockchain techniques has enabled secure smart trading in many realms, e.g. neighboring energy trading. However, trading information recorded on the blockchain also brings privacy concerns. Attackers can utilize data mining algorithms to obtain users’ privacy, specially, when the user group is located in nearby geographic positions. In this paper, we present a consortium blockchain-oriented approach to solve the problem of privacy leakage without restricting trading functions. The proposed approach mainly addresses energy trading users’ privacy in smart grid and screens the distribution of energy sale of sellers deriving from the fact that various energy trading volumes can be mined to detect its relationships with other information, such as physical location and energy usage. Experiment evaluations have demonstrated the effectiveness of the proposed approach. <s> BIB006

The proposal focuses on security requirements and ensures better performance than other popular methods. It is a proposal with a simulation. Quality Focus: Security Paper Title: RP-3-GridMonitoring: Secured Sovereign Blockchain Based Monitoring on Smart Grid BIB003 . Problem Addressed: SG readings are transmitted via the internet which generates the risk of data manipulation also with one way communication customer is unaware of consumption of appliances at their homes. Proposed Solution:The paper proposes a sovereign blockchain technology, which provides transparency and provenance(authenticity). Realized by installing smart meters and smart contracts. The smart meter enables the user to monitor consumption and smart contract ensures no data manipulation. Quality Focus: Transparency, Reliability Paper Title: RP-4-Privacy-preserving Energy Trading Using Consortium Blockchain in Smart Grid BIB006 . Problem Addressed: Blockchain records trading info and has the risk of attacks via data mining algorithm and misuse users data more threat is when the geographical location is nearby. Proposed Solution:Proposes consortium blockchain-oriented to solve privacy leakage without compromising trading features. Implemented using the account mapping technique which hides user data hence avoids direct data access. The account is created based on the users usage. Conclusion: Investigated privacy issues in blockchain-based neighboring energy trading system on the locality of 20 neighbors. Proposed Solution:The paper proposes blockchain-based decentralized access control system, the system is based on consortium blockchain, proposes consensus algorithm for the selection of private key generator (PKG) in smart power grids with incentive and penalty mechanism, the PKG here is reusable and could be used for many scenarios. The proposed scheme focuses on satisfying security requirements of the power grid, on solving the key escrow problem, making the user more involved in the daily management of smart grids and reusability of PKG reduces communication costs and suits multi-environments in smart grids. Quality Focus: Reliability Paper Title: RP-14-A blockchain-based smart grid towards sustainable local energy BIB002 . Problem Addressed: Centralized energy market used for NRE is not suitable for RES which is volatile, requires distributed generation and real-time mechanism to react in the wholesale market. Proposed Solution: Managing decentralized generation through digitalization. The proposed new market mechanism called Local Energy Market (LEM). It was proposed by the European commission in 2016. Implemented using private and permission blockchain which doesnt require any central authority. LEM is limited to the geographical or social community and balance demand and supply locally. The mathematical formula is developed for accurate calculation of demand and supply and is being forecasted. Conclusion: Implemented online mechanism matching supply and demand, reduce trading complexity by forecasting consumption and generation, the system does not involve any intermediaries. The system is built for small geographical communities using private and permission blockchain which ensure greater security. Quality Focus: Demand Supply balance, Security, Efficiency, Ease of trade Paper Title: RP-18-Blockchain Based Decentralized Management of Demand Response Programs in Smart Energy Grids. Problem Addressed: Issues of demand-supply management BIB004 . Proposed Solution:Proposes a blockchain-based distributed ledger the ledger stores the information of prosumers collected from the internet of things in a tamperproof manner and self-enforce smart contracts. prosumers get incentives or penalties for balancing demand-supply. Consensus-based validation is used for financial settlements. Conclusion: The prototype was implemented in Ethereum to validate and test the blockchain-based decentralized management using energy traces of UK building datasets. Results show that the grid is capable of balancing energy demand in near real-time. Quality Focus: Demand Supply balance, Ease of trade Paper Title: RP-22-Designing micro-grid energy markets A case study The Brooklyn Microgrid-new . Problem Addressed: Demand supply balance in distributed RES generation which by nature volatile. Proposed Solution:Proposes self-trading within the community and also keeping the profit within the community, blockchain-based decentralized energy market without central intermediaries. Quality Focus: Demand Supply balance, Ease of trade Paper Title: RP-7-Feather forking as a positive force: incentivizing green energy production in a blockchain-based smart grid. Problem Addressed: Increased use of non-renewable energy BIB005 . Proposed Solution:Proposes an SG architecture(distributed) incorporating blockchain which focuses on discouraging production and distribution of non-renewable energy. An ethical mechanism is being introduced inside blockchain to incentivize the green energy production the paper discusses the feather forking attack and formalize it for green energy. Conclusion: The proposed techniques seems promising for ethical smart grid system where all stakeholders collaborate for a greener environment. Quality Focus: Green Energy Paper Title: RP-15-NRGcoin virtual currency for trading of renewable energy in smartgrids BIB001 . Problem Addressed: Implement Green energy. Distributes energy market for RES generally rely on the day-ahead prediction of demand and supply for trading. This forecast needs to be near to accurate and also prosumer have to develop a trading strategy to maximize profit. Proposed Solution: D-CETI is designed generate carbon credits, trade and manage them. It has inherited the bitcoin mining process based on cryptographic proof of work, dataset is distributed and public and provides flexible way of participating in distributed network there is no central bank or authority to manage carbon emission credits. It is a high privacy system focused on reducing carbon emission. Quality Focus: Privacy, Carbon emission reduction, green energy.

A survey of of blockchain-based solutions for Energy Industry <s> Maturity <s> In this paper we introduce a new decentralized digital currency, called NRGcoin. Prosumers in the smart grid trade locally produced renewable energy using NRGcoins, the value of which is determined on an open currency exchange market. Similar to Bitcoins, this currency offers numerous advantages over fiat currency, but unlike Bitcoins it is generated by injecting energy into the grid, rather than spending energy on computational power. In addition, we propose a novel trading paradigm for buying and selling green energy in the smart grid. Our mechanism achieves demand response by providing incentives to prosumers to balance their production and consumption out of their own self-interest. We study the advantages of our proposed currency over traditional money and environmental instruments, and explore its benefits for all parties in the smart grid. <s> BIB001 </s> A survey of of blockchain-based solutions for Energy Industry <s> Maturity <s> Due to the recent advances in household-level renewable electricity generation technology, a new type of market based on peer-to-peer (P2P) electricity trading between households will emerge. This poster presents a technical platform which will enable such a market to function. <s> BIB002 </s> A survey of of blockchain-based solutions for Energy Industry <s> Maturity <s> As decentralized renewable energy has been installed into the present grid, millions of consumers and prosumers interacting each other over power grid. The variety of needs for exchanging energy would arise based on each customers' situation. The future electricity grid should be a bi-directional system with interconnected using distributed energy management software. Being able to provide a secure and decentralized control to these autonomous, peer-to-peer exchanges is one of the biggest challenges. We propose to use Blockchain-based electricity trading system with Digitalgrid router as an underlying platform because it could fit these requirements. Digitalgrid router, which consists back-to-back bi-directional digital inverters with software-based control, enables us to realize power exchange. <s> BIB003 </s> A survey of of blockchain-based solutions for Energy Industry <s> Maturity <s> Bloekehain — software technology used for peer-to-peer transaction platforms — is being increasingly applied to alternative opportunities for a variety of industries and use cases such as also in the energy sector. As of now all energy related to blockchain applications are still in a concept stage. This paper aims to present the preliminary ideas of a research project that we have launched in collaboration with a small Fintech company: the Crypto-Trading project. From a research perspective we have explored how a blockchain-based system integrating smart contract functionality can be used to share energy in order to promote smart grids for the management of electricity in the Sardinia Region. From a technical perspective the project will implement a modular blockchain-based software platform for extending the features of cryptocurrency exchanges to the renewable Energy Market, including a robo-advisor which will suggest prosumers the best selling strategy. Blockchain technology shows a lot of promise. In our opinion this might boost the growth of renewable energy production and consequently also have a positive effect on the regional economy. <s> BIB004 </s> A survey of of blockchain-based solutions for Energy Industry <s> Maturity <s> Blockchain may help solve several complex problems related to integrity and trustworthiness of rapid, distributed, complex energy transactions and data exchanges. In a move towards resilience, blockchain commoditizes trust and enables automated smart contracts to support auditable multiparty transactions based on predefined rules between distributed energy providers and customers. Blockchain based smart contracts also help remove the need to interact with third-parties, facilitating the adoption and monetization of distributed energy transactions and exchanges, both energy flows as well as financial transactions. This may help reduce transactive energy costs and increase the security and sustainability of distributed energy resource (DER) integration, helping to remove barriers to a more decentralized and resilient power grid. This paper explores the application of blockchain and smart contracts to improve smart grid cyber resiliency and secure transactive energy applications. <s> BIB005 </s> A survey of of blockchain-based solutions for Energy Industry <s> Maturity <s> Blockchain may help solve several complex problems related to securing the integrity and trustworthiness of rapid, distributed, complex energy transactions and data exchanges. In a move towards grid resilience, blockchain commoditizes trust and enables automated smart contracts to support auditable multiparty transactions based on predefined rules between distributed energy providers and customers. Blockchain based smart contracts also help remove the need to interact with third-parties, facilitating the adoption and monetization of distributed energy transactions and exchanges, both energy flows as well as financial transactions. This may help reduce transactive energy costs and increase the security and sustainability of distributed energy resource (DER) integration, helping to remove barriers to a more decentralized and resilient power grid. This paper explores the application of blockchain and smart contracts to improve smart grid cyber resiliency and secure transactive energy applications. <s> BIB006 </s> A survey of of blockchain-based solutions for Energy Industry <s> Maturity <s> The increasing amount of renewable energy sources in the energy system calls for new market approaches to price and distribute the volatile and decentralized generation. Local energy markets, on which consumers and prosumers can trade locally produced renewable generation directly within their community, balance generation and consumption locally in a decentralized approach. We present a comprehensive concept, market design and simulation of a local energy market between 100 residential households. Our approach is based on a distributed information and communication technology, i.e. a private blockchain, which underlines the decentralized nature of local energy markets. Thus, we provide energy prosumers and consumers with a decentralized market platform for trading local energy generation without the need of a central intermediary. Furthermore, we present a preliminary economic evaluation of the market mechanism and a research agenda for the technological evaluation of blockchain technology as the local energy market’s main information and communication technology. <s> BIB007 </s> A survey of of blockchain-based solutions for Energy Industry <s> Maturity <s> Electricity is the commonest commodity for most businesses in our world today. The use of electricity has been a breakthrough for the discovery of new technologies and has become the main driving force behind several innovations. With the introduction of smart grid systems, there have been improvements in how utility companies interact with their customers with regards to electricity use. However, since the readings are done via the Internet, there is the tendency for the data to be compromised when it gets into the hands of the wrong people. Moreover, customers mostly do not know why they pay huge amounts and which appliances use more electricity, since they are not privy to the readings. The sovereign blockchain technology, which provides transparency and provenance, is utilized in this paper to mitigate these above mentioned problems. A smart contract, which executes laid down procedures to provide a trust-based system between participants on the network is also implemented. Our system proves very efficient as the user can monitor how the electricity is used, and it also provides a platform where there is no manipulation from either party. <s> BIB008 </s> A survey of of blockchain-based solutions for Energy Industry <s> Maturity <s> The improvement of the Quality of Life (QoL) and the enhancement of the Quality of Services (QoS) represent the main goal of every city evolutionary process. It is possible making cities smarter promoting innovative solutions by use of Information and Communication Technology (ICT) for collecting and analysing large amounts of data generated by several sources, such as sensor networks, wearable devices, and IoT devices spread among the city. The integration of different technologies and different IT systems, needed to build smart city applications and services, remains the most challenge to overcome. In the Smart City context, this paper intends to investigate the Smart Environment pillar, and in particular the aspect related to the implementation of Smart Energy Grid for citizens in the urban context. The innovative characteristic of the proposed solution consists of using the Blockchain technology to join the Grid, exchanging information, and buy/sell energy between the involved nodes (energy providers and private citizens), using the Blockchain granting ledger. <s> BIB009

The different solutions we have surveyed can also be categorized according to an orthogonal dimension, meaning how solid they are considering a range from idea to profitable product. We will use here the following scale: • Idea: nothing has been done apart from a proposal • Concept: the idea has been structured and conceptualized • Prototype: there is a minimal working system • Product: there is a fully functional system that could be commercialized • Profit: the system is able to generate profit Most of the survey paper included here could be categorized into the top 4 categories namely idea, concept, prototype and product. RP-11 BIB003 , RP-23 BIB004 , RP-24 BIB002 is just an idea of blockchain-based electricity trading system. RP-15 BIB001 is a just idea/concept of new currency to trade renewable energy. RP-5 BIB005 and Rp-6 [28], RP-21 BIB006 are just exploring the use of blockchain to improve cybersecurity and demand-side management respectively in smart grids and do not propose any solution (Exploring several ideas BIB008 has proposed design algorithms and has also implemented a sovereign blockchain-based system on the smart grid and have shown real results for the improved quality attributes such as transparency. RP-14 BIB007 presents a comprehensive concept, market design, and simulation of a local energy market between 100 residential households. RP-17 BIB009 proposes a complex architecture for the implementation of SGs using blockchain and implemented a mobile application for information exchange among prosumers hence fall in the category of product.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> This paper investigates the maximal channel coding rate achievable at a given blocklength and error probability. For general classes of channels new achievability and converse bounds are given, which are tighter than existing bounds for wide ranges of parameters of interest, and lead to tight approximations of the maximal achievable rate for blocklengths n as short as 100. It is also shown analytically that the maximal rate achievable with error probability ? isclosely approximated by C - ?(V/n) Q-1(?) where C is the capacity, V is a characteristic of the channel referred to as channel dispersion , and Q is the complementary Gaussian cumulative distribution function. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> In the past few decades, wireless communications have been evolved from GSM system (2G) to LTE-A networks (4G) with the major interest focusing on the throughput related criteria. 5G communication networks, however, extend to a 3-D performance metric cube based on throughput, number of links and delay simultaneously. Moreover, 5G networks confronts a wider range of future applications, including person-to-person, person-to-machine, or even machine-to-machine types. To deal with all the above cases and challenges, we start from the investigation of most popular 5G scenarios and identify the requirements on the energy efficiency, reliability and delay. Detailed technologies are surveyed and discussed thereafter, to achieve critical requirements and facilitate the fundamental tradeoffs among 3-D performance criteria. Through this study, we hope to shed some lights on the novel 5G communication system design, and further pave the way towards energy efficient, low latency and high reliable communication networks. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> Applications enabled by Cooperative Intelligent Transport Systems (C-ITS) represent a major step towards making the road transport system safer and more efficient (green), and thus suited for a sustainable future. Wireless communication between vehicles and road infrastructure is an enabler for high-performance C-ITS applications. State-of-the-art communication systems for supporting low-latency C-ITS applications are based on IEEE 802.11 medium access control (MAC) and physical (PHY) layers. In this paper, we argue that a well-designed 5G system can complement or even replace these systems. We will review the C-ITS application requirements and explain how these are well aligned with the foreseen generic 5G service of ultra-reliable machine-type communication (uMTC). Key technology components suitable for constructing the uMTC service are identified: reliable service composition (RSC) and device-to-device (D2D) links for all-to-all broadcast communication, operational at high mobility and with varying degree of network assistance. Important problems for future studies, including radio-resource management, medium access control, and physical layer challenges, are discussed. <s> BIB003 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> The fifth generation (5G) of cellular networks is starting to be defined to meet the wireless connectivity demands for 2020 and beyond. One area that is considered increasingly important is the capability to provide ultra-reliable and low-latency communication, to enable e.g., new mission-critical machine-type communication use cases. One such example with extremely demanding requirements is the industrial automation with a need for ultra-low latency with a high degree of determinism. In this paper, we discuss the feasibility, requirements and design challenges of an OFDM based 5G radio interface that is suitable for mission-critical MTC. The discussion is further accompanied with system-level performance evaluations that are carried out for a factory hall-wide automation scenario with two different floor layouts. <s> BIB004 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> Fifth generation wireless networks are currently being developed to handle a wide range of new use cases. One important emerging area is ultra-reliable communication with guaranteed low latencies well beyond what current wireless technologies can provide. In this paper, we explore the viability of using wireless communication for low-latency, high-reliability communication in an example scenario of factory automation, and outline important design choices for such a system. We show that it is possible to achieve very low error rates and latencies over a radio channel, also when considering fast fading signal and interference, channel estimation errors, and antenna correlation. The most important tool to ensure high reliability is diversity, and low latency is achieved by using short transmission intervals without retransmissions, which, however, introduces a natural restriction on coverage area. <s> BIB005 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> 5G is envisioned to support unprecedented diverse applications and services with extremely heterogeneous performance requirements, i.e., mission critical IoT communication, massive machine-type communication and Gigabit mobile connectivity. This imposes enormous challenges to fulfil the key performance requirements, in particular, mission critical IoT communication, which calls for a dramatic paradigm change in 5G. This paper presents vision and challenges of mission critical IoT scenarios and the enabling technologies in 5G. Several research opportunities are given as example for inspiration purpose. <s> BIB006 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> Author(s): Weiner, Matthew | Advisor(s): Nikolic, Borivoje | Abstract: In the near future, the number of wireless devices will outnumber humans by an order of magnitude, and most of these devices will communicate with each other instead of people. They will not only sense the environment, as most do today, but they will also manipulate it. This closed loop operation will not require the high data rates that today’s people-centric networks provide; instead it will require low-latency and high-reliability communication at moderate overall data rates. Currently, high-performance industrial control is one of the few applications that has similar requirements to the IoT of the future. However, control systems exclusively use wired networks because existing wireless systems and standards cannot achieve the latency and reliability required since they are designed for either high- throughput or low-power communication between a small number of terminals.The first part of this work focuses on the design and evaluation of a wireless system architecture appropriate for high-performance control systems with a large number of sensors and actuators. Based on a model and representative set of specifications, an initial wireless system architecture is developed that is aimed at addressing the issues current WLAN and cellular systems have with supporting low-latency and high-reliability operation for a large number of users. The initial architecture’s achievable latency is a strong function of the available diversity, so it requires a scheme to generate the diversity in a low-latency manner without relying heavily on the channels provided by nature. One option is to use multiple, cooperating access points that are distributed around the system, as is done with coordinated multipoint in cellular systems. This has limited usefulness in many IoT and control systems since additional infrastructure is not possible or desirable, especially in the likely case that the access points are connected with wires. As an alternative option, a cooperative relaying system is developed that is based on decentralized relaying with semi-scheduled transmissions where relays can transmit simultaneously using a distributed space-time code, such as cyclic delay diversity. The proposed cooperative relaying system and other baseline schemes are analyzed using a simplified link level analysis, and the proposed system significantly outperforms the other schemes.The second part of this work looks at aspects of implementing the physical layer of the proposed cooperative relaying system architecture, including the analog front end, modulation, baseband processing, and multiple access protocol. An emphasis is put on reusing as many blocks from current systems as possible.Low-latency systems require efficient hardware, and error control decoders are an essential part of high-reliability wireless systems.The third part of this work looks at the implementation of a low-latency, low-power LDPC decoder for the IEEE 802.11ad standard, whose LDPC codes have many features applicable to wireless control. The decoder’s highly parallel, deeply pipelined, flooding architecture balances latency and power. Row-merging, simultaneous processing of multiple codewords, reduced marginalization memory precision, and using back-biasing to optimally trade off active and leakage power further reduce latency and power. The decoder is implemented in a 28nm FDSOI technology, has sub-microsecond level latency, and consumes only 6.2mW at a throughput of 1.5Gb/s and 38.1mW at 6Gb/s. <s> BIB007 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> The development of every new generation of wireless communication systems starts with bold, high-level requirements and predictions of its capabilities. The 5G system will not only have to surpass previous generations with respect to rate and capacity, but also address new usage scenarios with very diverse requirements, including various kinds of machine-type communication. Following this, the METIS project has developed a 5G system concept consisting of three generic 5G services: extreme mobile broadband, massive machine-type communication, and ultra-reliable MTC, supported by four main enablers: a lean system control plane, a dynamic radio access network, localized contents and traffic flows, and a spectrum toolbox. This article describes the most important system-level 5G features, enabled by the concept, necessary to meet the very diverse 5G requirements. System-level evaluation results of the METIS 5G system concept are presented, and we conclude that the 5G requirements can be met with the proposed system concept. <s> BIB008 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> mcMTC is starting to play a central role in the industrial Internet of Things ecosystem and have the potential to create high-revenue businesses, including intelligent transportation systems, energy/ smart grid control, public safety services, and high-end wearable applications. Consequently, in the 5G of wireless networks, mcMTC have imposed a wide range of requirements on the enabling technology, such as low power, high reliability, and low latency connectivity. Recognizing these challenges, the recent and ongoing releases of LTE systems incorporate support for lowcost and enhanced coverage, reduced latency, and high reliability for devices at varying levels of mobility. In this article, we examine the effects of heterogeneous user and device mobility -- produced by a mixture of various mobility patterns -- on the performance of mcMTC across three representative scenarios within a multi-connectivity 5G network. We establish that the availability of alternative connectivity options, such as D2D links and drone-assisted access, helps meet the requirements of mcMTC applications in a wide range of scenarios, including industrial automation, vehicular connectivity, and urban communications. In particular, we confirm improvements of up to 40 percent in link availability and reliability with the use of proximate connections on top of the cellular-only baseline. <s> BIB009 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> The fifth generation (5G) wireless network technology is to be standardized by 2020, where main goals are to improve capacity, reliability, and energy efficiency, while reducing latency and massively increasing connection density. An integral part of 5G is the capability to transmit touch perception type real-time communication empowered by applicable robotics and haptics equipment at the network edge. In this regard, we need drastic changes in network architecture including core and radio access network (RAN) for achieving end-to-end latency on the order of 1 ms. In this paper, we present a detailed survey on the emerging technologies to achieve low latency communications considering three different solution domains: 1) RAN; 2) core network; and 3) caching. We also present a general overview of major 5G cellular network elements such as software defined network, network function virtualization, caching, and mobile edge computing capable of meeting latency and other 5G requirements. <s> BIB010 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> I. INTRODUCTION <s> URLLC is a new service category in 5G to accommodate emerging services and applications having stringent latency and reliability requirements. In order to support URLLC, there should be both evolutionary and revolutionary changes in the air interface named 5G NR. In this article, we provide an up-to-date overview of URLLC with an emphasis on the physical layer challenges and solutions in 5G NR downlink. We highlight key requirements of URLLC and then elaborate the physical layer issues and enabling technologies including packet and frame structure, scheduling schemes, and reliability improvement techniques, which have been discussed in the 3GPP Release 15 standardization. <s> BIB011

The focus of cellular communication has been historically engineered on human-centric communication, which typically provides some form of information to people from network-hosted servers, such as mobile Internet services or video streaming. Hence, the 5G is expected to play a central role in Machine-Type Communications (MTC) and Human-to-Human (H2H) communications. MTC has been formed by 3rd Generation Partnership Project (3GPP) and it is characterized by entirely automatic data generation, actuation, processing, and exchange between intelligent machines, with or without low intervention of humans . MTC, often termed Machine to Machine (M2M), starts to play an increasing role in cellular networks and the efforts have been put in the latest Long-Term Evolution (LTE) releases to address MTC requirements BIB003 i.e., wireless system to a larger extent, will be used in the context of MTC . As a result of the rapid development of embedded devices, MTC is becoming the dominant communication paradigm for various emerging smart services including industrial automation, public safety, health-care, utilities, transportation, smart metering, remote manufacturing, and numerous other applications BIB009 , BIB010 , while mobile devices communication among humans still exists. Unlike H2H telecommunication, in which messages, voices, and videos are transmitted, most MTC devices transmit only a small amount of data. A great deal of efforts has been dedicated in Telecommunications industry to realize high-capacity and high-throughput infrastructure. However, supporting small data communications for MTC subscribers has been neglected. The MTC in context of 5G promises huge market growth with an expectation to support connectivity of at least 100 billion devices with the connection speed of 10 Gigabits per second (Gb/s). Furthermore, to provide a zero-distance connectivity between people and connected machines with extremely low latency and response times . In addition, enhanced mobile Internet services will improve consumers experience and guarantee profits for operators. MTC will also drive economic and societal growth to entirely new diversities. Thus, the emerging Industrial Internet promises a surpassing transformation of global industry, with a targeted economic boost that is more than $19 trillion within the next decade BIB009 . Deployment of these networks is expected to be emerged between 2020 and 2030. The continuing growth in demand for cellular-based MTC from 2014 to 2019 motivates initiatives in the industry, mobile networks operators, academic communities, and standardization bodies to investigate evolutionary and revolutionary radio access technologies to make MTC traffic accommodate 5G. For instance, the European Union has formed an official project called ''Mobile and wireless communications Enablers for the Twenty-twenty Information Society (METIS)'' for joining 5G research and standard promotion BIB008 . The Chinese government established an official promotion group called ''International Mobile Telecommunications Twenty-Twenty'' (IMT2020), whose main role is to establish the foundation of 5G mobile and wireless communications system. In addition, manufacturing companies have invested potential research projects to participate to the foundation of 5G communications and to enable mission-critical applications. Some of these industrial partners are: Ericsson BIB004 , , BIB005 Huawei , 5GNOW BIB006 , 3GPP , NOKIA , FuTURE Forum of China, 5GPPP of Europe Union, ADWICS of Japan, 5G forum of Korea, 5G Novel Radio Multiservice adaptive network Architecture (5G NORMA) , T-Mobile and others BIB002 . URLLC2017 conference gathered more than ten commercial and industrial companies . Figure 2 lists most of the industrial companies involved in industrial-leading demos, simulations, testing, and trials on the path to commercialization of 5G. Although, MTC has extensive potentials to accompany improvements for both lifestyle and business, MTC is confronting several unsolved technical challenges that differ from those in H2H communications, which are related to massive connectivity, latency, reliability, availability, energy efficiency, heterogeneous devices, limited spectrum, hardware miniaturization, security and privacy BIB006 . In essence, Mission-critical applications require very low latency, ultra-high reliability, and high availability communications i.e. mission-critical MTC (mcMTC) . Unfortunately, all current mission-critical applications use wired networks because no current wireless networks can meet their mcMTC requirement BIB007 . To the best of our knowledge, no work has been done yet to combine all these requirements into a theoretical framework, although the foundation has been provided in Polyanskiy work BIB001 . Moreover, it is not possible to satisfy these mcMTC relentless requirements with a small modification of current radio access technology BIB011 . Hence, This study investigates how next generation of cellular networks must be designed in order to cater to this type of traffic. Also, to study the flexibility options of mcMTC to support its various use cases which have different requirements for connectivity in terms of reliability, latency, and availability. Furthermore, it is important to explore the 5G multiplexing of mcMTC services without contradicting with other operating modes which are enhanced Mobile Broadband (eMBB) and massive Machine-Type Communications (mMTC). Other challenges aside from these three mcMTC requirements such as low power consumption and expected security level of the network are also interesting, but are often irrelevant for mcMTC and are not covered in this review. This paper presents a critical review of mcMTC requirements and challenges in the context of 5G wireless technology. It first presents the 5G operating regions and the existing MTC access methods; then it explains in detail the key requirements of mcMTC applications/services. Finally, it provides an extensive review on enabling technologies towards providing mcMTC for critical-mission applications. It shows the vision of different companies, organizations, and researchers who are pioneers in this field of study focusing on mcMTC and the proposed solutions for its requirements towards 5G. Eventually, this paper aims to outline and analyze the involved technical issues, to review recent advances, and to bring the state-of-the-art of mcMTC solutions along with future research directions. The rest of this review paper is organized as follows. Section II presents the three 5G service regions. MTC access methods are presented in section III. Section IV discusses the key requirements of mcMTC. Enabling technologies of mcMTC which includes air interface and multiplexing schemes, network architecture, packet and frame structure, control channel design, communication diversity, proactive packet drop, and coding and modulation are discussed in section V. Finally, conclusion and future work are drawn in section VI. The organization of the paper is illustrated in Figure1, while the acronyms used in this paper are listed in Table1.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> II. 5G WIRELESS OPERATING REGIONS <s> Machine-to-machine communication has attracted a lot of interest in the mobile communication industry and is under standardization process in 3GPP. Of particular interest is LTE-Advanced support for various M2M service requirements and efficient management and handling of a huge number of machines as mobile subscribers. In addition to the higher throughput, one of the main advantages of LTE/LTE-A in comparison with the previous cellular networks is the reduced transmission latency, which makes this type of networks very attractive for real-time mobile M2M communication scenarios. This paper presents a M2M system architecture based on LTE/LTE-A and highlights the delays associated with each part of the system. Three real-time M2M applications are analyzed and the main latency bottlenecks are identified. Proposals on how the latency can be further reduced are described. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> II. 5G WIRELESS OPERATING REGIONS <s> The grand objective of 5G wireless technology is to support three generic services with vastly heterogeneous requirements: enhanced mobile broadband (eMBB), massive machine-type communications (mMTC), and ultra-reliable low-latency communications (URLLC). Service heterogeneity can be accommodated by network slicing, through which each service is allocated resources to provide performance guarantees and isolation from the other services. Slicing of the Radio Access Network (RAN) is typically done by means of orthogonal resource allocation among the services. This work studies the potential advantages of allowing for non-orthogonal sharing of RAN resources in uplink communications from a set of eMBB, mMTC and URLLC devices to a common base station. The approach is referred to as Heterogeneous Non-Orthogonal Multiple Access (H-NOMA), in contrast to the conventional NOMA techniques that involve users with homogeneous requirements and hence can be investigated through a standard multiple access channel. The study devises a communication-theoretic model that accounts for the heterogeneous requirements and characteristics of the three services. The concept of reliability diversity is introduced as a design principle that leverages the different reliability requirements across the services in order to ensure performance guarantees with non-orthogonal RAN slicing. This study reveals that H-NOMA can lead, in some regimes, to significant gains in terms of performance trade-offs among the three generic services as compared to orthogonal slicing. <s> BIB002

The evolution of cellular wireless systems in the last three decades from 2G to the current LTE-Advanced (LTE-A) has been dedicated to offer the users connectivity with a higher demand of data rates. 5G features at least two new operating modes mMTC and mcMTC and increasing the speed of existing operating mode enhanced Mobile Broadband (eMBB). These two new operating modes and the existing mode have been described in and as the three generic 5G services (i.e. eMBB, mMTC, and mcMTC). Among these three generic 5G services, mcMTC is expected to be a technology driver of 5G , BIB002 . In contrast to the other services, mcMTC refers to the services that provide a very high reliability and a very low latency as well as ubiquitous communication to support mcMTC applications, on which service failures have severe consequences. Table 3 summarizes the key features and the main requirements of these three 5G operating regions. A. EMBB eMBB initially becomes an extension to existing 4G services and therefore it is the first 5G service which evolved in the 3GPP Release 15 standards in June 2018. The standard was approved in December 2017 and already standardized in 3GPP release 15 which could be commercially available as early as 2019. Ericsson estimates that there will be one billion 5G subscriptions for eMBB globally by 2023 BIB001 , with Asia and North America likely to be the first adopters. However, 5G Phase 2 will go beyond eMBB services to more transformational mMTC and mcMTC applications and will be included in Release 16, which is due to be completed at the end of 2019.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> B. MMTC <s> The fifth generation (5G) of cellular networks is starting to be defined to meet the wireless connectivity demands for 2020 and beyond. One area that is considered increasingly important is the capability to provide ultra-reliable and low-latency communication, to enable e.g., new mission-critical machine-type communication use cases. One such example with extremely demanding requirements is the industrial automation with a need for ultra-low latency with a high degree of determinism. In this paper, we discuss the feasibility, requirements and design challenges of an OFDM based 5G radio interface that is suitable for mission-critical MTC. The discussion is further accompanied with system-level performance evaluations that are carried out for a factory hall-wide automation scenario with two different floor layouts. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> B. MMTC <s> Effective emergency and natural disaster management depend on the efficient mission-critical voice and data communication between first responders and victims. Land Mobile Radio System (LMRS) is a legacy narrowband technology used for critical voice communications with limited use for data applications. Recently Long Term Evolution (LTE) emerged as a broadband communication technology that has a potential to transform the capabilities of public safety technologies by providing broadband, ubiquitous, and mission-critical voice and data support. For example, in the United States, FirstNet is building a nationwide coast-to-coast public safety network based of LTE broadband technology. This paper presents a comparative survey of legacy and the LTE-based public safety networks, and discusses the LMRS-LTE convergence as well as mission-critical push-to-talk over LTE. A simulation study of LMRS and LTE band class 14 technologies is provided using the NS-3 open source tool. An experimental study of APCO-25 and LTE band class 14 is also conducted using software-defined radio, to enhance the understanding of the public safety systems. Finally, emerging technologies that may have strong potential for use in public safety networks are reviewed. <s> BIB002

mMTC refers to services, in which there is an enormous amount of sensors and actuators to control an environment with simple, scalable and energy efficient communication supporting massive number of devices in some areas. Moreover, mMTC devices required to remain very simple and can operate on batteries for many years. mMTC data transferred per device are typically infrequent and are accompanied by relaxed delay requirements. mMTC already emerges as an extension of the 4G LTE systems and becomes in the scope of current standardization of 3GPP and IEEE. C. MCMTC mcMTC is having the most crucial challenges towards making it accommodates 5G. mcMTC is also called Ultra Reliable Low Latency Communications (URLLC), but we prefer in this review to use mcMTC which is relevant to real-time control and automation of dynamic processes in automated cyber-physical systems, such as industrial process control. mcMTC is still in the early-standardization phase of 5G New Radio (NR) and is expected to be standardized on the late of 2019, poses a lot of research problems that remain unsolved. This is due to the fact that mcMTC require communications with a very high reliability, high availability, and small latency going down to millisecond level , BIB001 , BIB002 .

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> III. MTC ACCESS METHODS <s> Factory automation is one of the challenging use cases that the fifth generation, 5G, networks are expected to support. It involves mission-critical machine-type communications, MTC, with requirements of extreme low-latency and ultra-reliable communication to enable real-time control of automation processes in manufacturing facilities. In this paper, we discuss the deployment strategies for the 5G mission-critical MTC solution designed to meet the needs of factory automation applications. The paper analyzes the coverage and capacity aspects based on a series of system-level evaluations considering both noise-limited and interference- limited operations. It further analyzes the related trade-offs to provide insights on the network deployment strategies for a realistic factory scenario. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> III. MTC ACCESS METHODS <s> Author(s): Weiner, Matthew | Advisor(s): Nikolic, Borivoje | Abstract: In the near future, the number of wireless devices will outnumber humans by an order of magnitude, and most of these devices will communicate with each other instead of people. They will not only sense the environment, as most do today, but they will also manipulate it. This closed loop operation will not require the high data rates that today’s people-centric networks provide; instead it will require low-latency and high-reliability communication at moderate overall data rates. Currently, high-performance industrial control is one of the few applications that has similar requirements to the IoT of the future. However, control systems exclusively use wired networks because existing wireless systems and standards cannot achieve the latency and reliability required since they are designed for either high- throughput or low-power communication between a small number of terminals.The first part of this work focuses on the design and evaluation of a wireless system architecture appropriate for high-performance control systems with a large number of sensors and actuators. Based on a model and representative set of specifications, an initial wireless system architecture is developed that is aimed at addressing the issues current WLAN and cellular systems have with supporting low-latency and high-reliability operation for a large number of users. The initial architecture’s achievable latency is a strong function of the available diversity, so it requires a scheme to generate the diversity in a low-latency manner without relying heavily on the channels provided by nature. One option is to use multiple, cooperating access points that are distributed around the system, as is done with coordinated multipoint in cellular systems. This has limited usefulness in many IoT and control systems since additional infrastructure is not possible or desirable, especially in the likely case that the access points are connected with wires. As an alternative option, a cooperative relaying system is developed that is based on decentralized relaying with semi-scheduled transmissions where relays can transmit simultaneously using a distributed space-time code, such as cyclic delay diversity. The proposed cooperative relaying system and other baseline schemes are analyzed using a simplified link level analysis, and the proposed system significantly outperforms the other schemes.The second part of this work looks at aspects of implementing the physical layer of the proposed cooperative relaying system architecture, including the analog front end, modulation, baseband processing, and multiple access protocol. An emphasis is put on reusing as many blocks from current systems as possible.Low-latency systems require efficient hardware, and error control decoders are an essential part of high-reliability wireless systems.The third part of this work looks at the implementation of a low-latency, low-power LDPC decoder for the IEEE 802.11ad standard, whose LDPC codes have many features applicable to wireless control. The decoder’s highly parallel, deeply pipelined, flooding architecture balances latency and power. Row-merging, simultaneous processing of multiple codewords, reduced marginalization memory precision, and using back-biasing to optimally trade off active and leakage power further reduce latency and power. The decoder is implemented in a 28nm FDSOI technology, has sub-microsecond level latency, and consumes only 6.2mW at a throughput of 1.5Gb/s and 38.1mW at 6Gb/s. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> III. MTC ACCESS METHODS <s> Real-time applications require latencies on the order of a millisecond with very high reliabilities, paralleling the requirements for high-performance industrial control. Current wireless technologies like WiFi, Bluetooth, LTE, etc. are unable to meet these stringent latency and reliability requirements, forcing the use of wired systems. This paper introduces a wireless communication protocol based on network coding that in conjunction with cooperative communication techniques builds the necessary diversity to achieve the target reliability. The proposed protocol is analyzed using a communication theoretic delay-limited-capacity framework and compared to proposed protocols without network coding. The results show that for larger network sizes or payloads employing network coding lowers the minimum SNR required to achieve the target reliability. For a scenario inspired by an industrial printing application with $30$ nodes in the control loop, aggregate throughput of $4.8$ Mb/s, $20$MHz of bandwidth and cycle time under $2$ ms, the protocol can robustly achieve a system probability of error better than $10^{-9}$ with a nominal SNR less than $2$ dB under ideal channel conditions. <s> BIB003

The connection of MTC devices to the network infrastructure could be either wireless or wired (i.e., cable, xDSL, and optical). Wired transmission technologies, such as Fieldbus and Ethernet-based systems, have dominated factories automation because of their reliability, fast response time and availability. However, the interest in employing wireless communications has grown recently to use the MTC connectivity for real-time applications, such as factory automation and process control in industrial networks BIB001 , . The third Generation of the Serial Real-time Communication System (SERCOS-III) interface is the international standard on the third generation open digital interface, which provides an openness, compatibility, high speed, and reliable real-time data communication among the controller and actuators BIB002 . SERCOS-III has been designed to provide reliable Media Access Control (MAC) sublayer schemes and to guarantee a certain level of performance (i.e. Latency 4ms, Reliability:10 −8 − 10 −10 , and Availability 100% BIB002 , BIB003 ). For the sake of supporting critical mission services, Sercos-III uses the speed of fast Ethernet (i.e. IEEE 802.3 100 Mbit/s baseband) and supports full duplex Physical (PHY) layer entities. In addition, it uses Cyclic Redundancy Check (CRC) codes to detect errors in the packet and Automatic Repeat reQuest (ARQ) policy to correct errors which improve the reliability of the service. Moreover, SERCOS-III supports for either line or ring topologies which improves fault tolerance (i.e. availability) by supporting redundant data transfer. In case of a break at any point in the ring, the SERCOS-III protocol automatically switches over to a dual-line structure. However, latency is not much more affected by recovery, which occurs in 25 µs. Although the wired solution can provide reliability, high rates and a very high response time, it is not suitable for mcMTC applications due to its expensive cost, lack of scalability and mobility. Alternatively, wireless access methods can be either cellular (i.e. LTE-A, GPRS, 3G, and WiMAX) or capillary (i.e., WLAN, IEEE 802.15.4x, and ZigBee). The wireless capillary solution, commonly used for shared short-range links, is generally cost-effective, low power, and scalable. It also allows more detailed monitoring than in wired communications, which would be economically more viable. However, it is not applicable to mcMTC applications due to the lack of coverage, low rate, weak security, and interference problem. On the other hand, the wireless cellular solution is promising to be as an infrastructure-based solution for the mcMTC applications due to its ability to provide ubiquitous coverage, mobility, roaming, and security solutions.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> IV. THE KEY REQUIREMENTS OF MCMTC <s> This paper investigates the maximal channel coding rate achievable at a given blocklength and error probability. For general classes of channels new achievability and converse bounds are given, which are tighter than existing bounds for wide ranges of parameters of interest, and lead to tight approximations of the maximal achievable rate for blocklengths n as short as 100. It is also shown analytically that the maximal rate achievable with error probability ? isclosely approximated by C - ?(V/n) Q-1(?) where C is the capacity, V is a characteristic of the channel referred to as channel dispersion , and Q is the complementary Gaussian cumulative distribution function. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> IV. THE KEY REQUIREMENTS OF MCMTC <s> Fifth generation wireless networks are currently being developed to handle a wide range of new use cases. One important emerging area is ultra-reliable communication with guaranteed low latencies well beyond what current wireless technologies can provide. In this paper, we explore the viability of using wireless communication for low-latency, high-reliability communication in an example scenario of factory automation, and outline important design choices for such a system. We show that it is possible to achieve very low error rates and latencies over a radio channel, also when considering fast fading signal and interference, channel estimation errors, and antenna correlation. The most important tool to ensure high reliability is diversity, and low latency is achieved by using short transmission intervals without retransmissions, which, however, introduces a natural restriction on coverage area. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> IV. THE KEY REQUIREMENTS OF MCMTC <s> The fifth generation (5G) of cellular networks is starting to be defined to meet the wireless connectivity demands for 2020 and beyond. One area that is considered increasingly important is the capability to provide ultra-reliable and low-latency communication, to enable e.g., new mission-critical machine-type communication use cases. One such example with extremely demanding requirements is the industrial automation with a need for ultra-low latency with a high degree of determinism. In this paper, we discuss the feasibility, requirements and design challenges of an OFDM based 5G radio interface that is suitable for mission-critical MTC. The discussion is further accompanied with system-level performance evaluations that are carried out for a factory hall-wide automation scenario with two different floor layouts. <s> BIB003 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> IV. THE KEY REQUIREMENTS OF MCMTC <s> Motivated by the current interest in ultra-reliable, low-latency, machine-type communication systems, we investigate the tradeoff between reliability, throughput, and latency in the transmission of information over multiple-antenna Rayleigh block-fading channels. Specifically, we obtain finite-blocklength, finite-SNR upper and lower bounds on the maximum coding rate achievable over such channels for a given constraint on the packet error probability. Numerical evidence suggests that our bounds delimit tightly the maximum coding rate already for short blocklengths (packets of about 100 symbols). Furthermore, our bounds reveal the existence of a tradeoff between the rate gain obtainable by spreading each codeword over all available time-frequency-spatial degrees of freedom, and the rate loss caused by the need of estimating the fading coefficients over these degrees of freedom. In particular, our bounds allow us to determine the optimal number of transmit antennas and the optimal number of time-frequency diversity branches that maximize the rate. Finally, we show that infinite-blocklength performance metrics such as the ergodic capacity and the outage capacity yield inaccurate throughput estimates. <s> BIB004

Historically, wireless networks have not been designed to satisfy the mcMTC applications/services BIB002 , while most of the efforts have been dedicated to offer the users connectivity with a higher demand for data rates. For example, LTE networks can perform a reliability rate of nearly 10 −2 , while the end-to-end (E2E) latency is approximately few milliseconds . For flexible requirements of latency, LTE can perform satisfied reliability via re-transmissions at different protocol layers. However, accomplishing a high reliability while satisfying a very-low latency presents a new challenge for the NR deployment. The key requirements of mcMTC can be considered as the ultra-high reliability, ultra-low latency and availability , , BIB004 . Table 2 presents a comparison of the mcMTC requirements in wired, LTE-A, and 5G networks. In mcMTC, the required levels of reliability, latency, and availability vary according to the selected use case and application. Therefore, the NR air interface and system design should be scalable enough to efficiently use the available network resources. Figure 4 depicts the diversity of latency and reliability requirements for different mcMTC use cases and applications. For mcMTC services, the levels of availability are required to be on the order of 10 −8 to 10 −12 and the packet loss probability has to be on the scale of 10 −5 to 10 −9 , while the latency has to go down to millisecond level BIB003 , . The International Telecommunication Union (ITU) has set a target to achieve a 1ms Round-Trip Time (RTT) over-the-air communication for a single transmission . This includes transmission of the payload until the corresponding acknowledgment is received. To the best of our knowledge, no work has been done yet to combine reliability, latency, and availability into a theoretical framework, although the foundation has been provided in Polyanskiy work BIB001 . Also, no wireless communication systems have been proposed for systems considering mcMTC requirements of reaching 1ms of latency with tens to hundreds of nodes and reliability constraint less than 10 −6 .