Context-aware Self-Optimization in Small-Cell Networks

Universidad de Málaga Escuela Técnica Superior de Ingeniería de Telecomunicación Programa de Doctorado en Ingeniería de Telecomunicación TESIS DOCTORAL Context-aware Self-Optimization in Small-Cell Networks Autor: ALEJANDRO AGUILAR GARCÍA Directora: RAQUEL BARCO MORENO Julio 2016 AUTOR: Alejandro Aguilar García http://orcid.org/0000-0002-2270-2260 EDITA: Publicaciones y Divulgación Científica. Universidad de Málaga Esta obra está bajo una licencia de Creative Commons Reconocimiento-NoComercialSinObraDerivada 4.0 Internacional: http://creativecommons.org/licenses/by-nc-nd/4.0/legalcode Cualquier parte de esta obra se puede reproducir sin autorización pero con el reconocimiento y atribución de los autores. No se puede hacer uso comercial de la obra y no se puede alterar, transformar o hacer obras derivadas. Esta Tesis Doctoral está depositada en el Repositorio Institucional de la Universidad de Málaga (RIUMA): riuma.uma.es A mi familia. Acknowledgements Firstly, I would like to express my sincere gratitude to my supervisor, Raquel Barco, for the continuous support of my PhD study and related research, for her motivation, patience, and knowledge. Her guidance helped me in all the time of research and writing of papers, reviews and this thesis. A very special thanks goes out to Sergio Fortes, for his motivation, encouragement and technical support, and for the sleepless nights we were working together before deadlines. I would also like to thank the rest of my laboratory mates for the stimulating discussions, the technical assistance and all the fun we have had in the last four years. I wish to thank the people I had the great opportunity to meet in MONOLOC project and ESIGETEL engineering school. Thanks Elizabeth Colin for her hospitality, her positive attitude and her technical assistance in the field of RFID-based indoor positioning systems. I must also acknowledge the financial support given by the Junta de Andalucía, the Spanish Ministry of Economy and Competitiveness and the European Development Fund, together with the research group TIC-102 Ingeniería de Comunicaciones and the University of Málaga, which have been equally important to make possible this work and allowed me to present results and exchange knowledge and skills in conferences, journals, workshops and short stays abroad. Finally, I would like to strongly thank my family for supporting me through my entire life, and in particular, my mother, Rosa, for all of the sacrifices she has made on my behalf, and my wife and best friend, María, without whose love, encouragement and assistance, I would not have finished this thesis. vii Index Abstract .................................................................................................. xi Resumen ............................................................................................... xiii Acronyms ............................................................................................... xv 1 Introduction ........................................................................................ 1 1.1 Motivation ................................................................................................. 1 1.2 Preliminaries .............................................................................................. 3 1.3 Research objectives .................................................................................... 4 1.4 Document structure ................................................................................... 6 2 Technical background ......................................................................... 9 2.1 Overview of the LTE standard .................................................................. 9 2.2 Self-Organizing Networks ......................................................................... 13 2.2.1 Self-optimization ................................................................................ 14 2.3 Femtocells ................................................................................................ 16 2.3.1 Characteristics ................................................................................... 16 2.3.2 LTE architecture ............................................................................... 18 2.3.3 SON for femtocells ............................................................................. 20 2.4 Context information ................................................................................. 21 2.4.1 Indoor positioning systems ................................................................ 22 2.4.2 Context-aware SON ........................................................................... 22 2.5 Conclusions .............................................................................................. 23 3 Context-aware SON framework ........................................................ 25 3.1 Introduction ............................................................................................. 25 3.2 Related work ............................................................................................ 26 3.3 Problem description ................................................................................. 27 3.4 Framework for context-aware SON .......................................................... 28 3.4.1 Framework characteristics ................................................................. 28 viii Index 3.4.2 OAM architecture .............................................................................. 32 3.4.3 SON use case ..................................................................................... 39 3.5 Evaluation ................................................................................................ 43 3.5.1 Simulation set-up ............................................................................... 43 3.5.2 Simulation results .............................................................................. 44 3.6 Conclusions .............................................................................................. 46 4 Indoor positioning strategies ............................................................. 47 4.1 Introduction ............................................................................................. 47 4.2 Related work ............................................................................................ 49 4.3 Problem description .................................................................................. 51 4.4 Methods for indoor positioning ................................................................. 52 4.4.1 Fingerprinting-based scheme .............................................................. 52 4.4.2 RFID-based positioning system ......................................................... 53 4.4.3 Cellular technology into RFID-based positioning system ................... 62 4.5 Evaluation ................................................................................................ 64 4.5.1 Trial set-up ........................................................................................ 65 4.5.2 Trial Results ...................................................................................... 70 4.5.3 Considerations for real deployments .................................................. 79 4.6 Conclusions .............................................................................................. 81 5 Indoor mobility load balancing techniques ........................................ 83 5.1 Introduction ............................................................................................. 83 5.2 Related work ............................................................................................ 85 5.3 Problem description .................................................................................. 87 5.3.1 Operators’ policy................................................................................ 87 5.3.2 Network configuration parameters ..................................................... 88 5.3.3 Key performance indicators ............................................................... 90 5.4 Methods for MLB in femtocell networks .................................................. 92 5.4.1 Fuzzy-based MLB mechanisms .......................................................... 92 5.5 Methods for context-aware MLB in femtocell networks ......................... 100 5.5.1 System set-up .................................................................................. 101 5.5.2 Users-distribution-based method ...................................................... 105 5.5.3 Virtual-Maps-based method ............................................................. 108 5.6 Simulation evaluation ............................................................................. 122 5.6.1 Simulation set-up ............................................................................. 123 5.6.2 Simulation results ............................................................................ 126 5.7 Field trial evaluation .............................................................................. 145 5.7.1 Trial set-up ...................................................................................... 145 5.7.2 Trial results ..................................................................................... 151 Index ix 5.7.3 Additional performance metrics ....................................................... 156 5.8 Conclusions ............................................................................................ 158 6 Conclusions ..................................................................................... 159 6.1 Contributions ......................................................................................... 159 6.2 Future work ........................................................................................... 161 6.3 List of publications................................................................................. 162 6.3.1 Journals ........................................................................................... 162 6.3.2 International conferences ................................................................. 163 6.3.3 National conferences ........................................................................ 163 6.3.4 Other publications ........................................................................... 164 6.3.5 Projects ........................................................................................... 165 A Technology characteristics .............................................................. 167 A.1 RFID technology .................................................................................... 167 A.2 Cellular technology ................................................................................ 168 B Signal assessment ............................................................................ 171 B.1 Introduction ........................................................................................... 171 B.2 RFID signal ............................................................................................ 173 B.2.1 Analysis of the measurements .......................................................... 173 B.2.2 Analysis per position ....................................................................... 174 B.3 GSM and UMTS signals ........................................................................ 175 B.3.1 Analysis of the measurements .......................................................... 175 B.3.2 Analysis per position ....................................................................... 177 B.4 Applicability of the results ..................................................................... 179 C Fuzzy logic controllers .................................................................... 181 C.1 System parameters and functions ........................................................... 182 C.2 System processes .................................................................................... 184 D Summary (Spanish) ......................................................................... 185 D.1 Introducción ........................................................................................... 185 D.1.1 Objetivos ......................................................................................... 188 D.2 Estado del arte ....................................................................................... 188 D.3 Información de contexto y métodos SON ............................................... 189 D.4 Técnicas de posicionamiento en interiores .............................................. 191 D.5 Técnicas de balance de carga en interiores ............................................. 192 D.6 Conclusiones ........................................................................................... 193 D.7 Lista de publicaciones ............................................................................ 195 Bibliography ........................................................................................ 199 xvi Acronyms DR Direct Retry E-UTRAN Evolved UMTS Terrestrial Radio Access Network EIA Extended Indoor A eNB eNodeB EM Element Manager EPC Evolved Packet Core EPS Evolved Packet System FDD Frequency-Division Duplex FLC Fuzzy Logic Controller FM Figure of Merit GPS Global Positioning System GSM Global System for Mobile Communications HCN Heterogeneous Cellular Networks HeNB Home eNB HetNets Heterogeneous Networks HF High Frequency HPLM Historical Path Loss Maps HSS Home Subscriber Server IM Integration Module JCR Journal Citation Reports JSON JavaScript Object Notation KPI Key Performance Indicator LAN Local Area Network LBS Location-based Service LF Low Frequency LIPA Local IP Access LNMA Local Network Manager Agent LoS Line of Sight MCQI Mean CQI MDT Minimization of Drive Test MEU MEasurement Unit MILES Mobile Indoor Localization Engine for SON MIMO Multiple Input Multiple Output MISO Multiple Input Single Output MLB Mobility Load Balancing MME Mobility Management Entity MSE Mean Square Error NCM Neighbor Cell Maps NNSF NAS Node Selection Function NE Network Element Acronyms xvii NFC Near Field Communication NM Network Management OAM Operation, administration and management OCAS OAM Context-Aware System OCU OAM Coordination Unit OFDM Orthogonal Frequency Division Multiplexing OFDMA Orthogonal Frequency Division Multiple Access OPEX Operational Expenditures OR Outage Ratio P-GW Packet Data Network Gateway PCI Physical Cell Identity PCRF Policy and Charging Rules Function PCS Proposed Cell Status PDF Probability Density Function PLS Power Load Sharing PLUS Power Load and User Sharing PRB Physical Resource Block PRX Power Received PTS Power Traffic Sharing PTX Power Transmitted PSC Primary Synchronization Code PUS Power User Sharing QAM Quadrature Amplitude Modulation QoE Quality of Experience QoS Quality of Service QPSK Quadrature Phase Shift Keying S-GW Serving Gateway SAU SON Algorithmic Unit SIPTO Selected IP Traffic Offload SPM Simple Point Matching SPS Semipersistent Scheduling SSID Service Set IDentifier RAN Radio Access Network RAT Radio Access Technologies RD Reference Distance REM Radio Environmental Maps RF Radio Frequency RFID Radio Frequency IDentification RRM Radio Resource Management RSCP Received Signal Code Power xviii Acronyms RSRP Reference Signal Received Power RSSI Received Signal Strength Indicator RTD Round Trip Delay RxLev Received Signal Level SC-FDMA Single-Carrier Frequency Division Multiple Access SCTP Stream Control Transmission Protocol SIMO Single Input Multiple Output SINR Signal-to-Interference-plus-Noise Ratio SISO Single Input Single Output SON Self-Organizing Networks SPM Simple Point Matching TCP Transmission Control Protocol TDD Time-Division Duplex ToF Time of Flight UD Users Distribution UDP User Datagram Protocol UDR User Dissatisfaction Rate UDS United States Dollars UE User Equipment UHF Ultra High Frequency UMTS Universal Mobile Telecommunications System UR Utility Function UWB Ultra Wide Band VM Virtual Maps WLAN Wireless Local Area Network XML EXtensible Markup Language 1 Chapter 1 1 Introduction This opening chapter aims to introduce the reader to the motivation and the context of this PhD thesis, the research objectives and the structure of the document. 1.1 Motivation Mobile networks have become a standard infrastructure in human beings life for several services, e.g., voice calls, text messages, video streaming, etc. Furthermore, the requirements posed by the massive expansion of smart-devices and the demand of mobile network services and applications increase the global mobile data traffic (74% of growth in 2015 [1]). Consequently, current mobile network infrastructures are driving towards their limits. In order to overcome these limits, mobile communications have dramatically changed and rapidly evolved. Operators are deploying new mobile technologies, e.g., LTE (Long Term Evolution). The LTE technology has been integrated over the already existing mobile infrastructures; GSM (Global System for Mobile Communications), UMTS (Universal Mobile Telecommunications System), etc., leading to different Radio Access Technologies (RAT). Additionally, whilst classical base stations (macrocells) covered areas of tens of kilometers, operators trend to deploy new short-range base stations: small cells (e.g., microcells, picocells, femtocells, etc.). These solutions create complex Heterogeneous Cellular Networks (HCNs), which partially support the growing demand of clients and applications services. Mobile communication providers and network engineers have to analyze a huge amount of data in HCNs to assess network performance and to propose the best strategy to solve performance issues. The analysis of these large amounts of information is not a simple task and requires long time and vast resources. That translates into increased capital (CAPEX) and operational expenditures (OPEX) for 2 Introduction the operator. Therefore, the complexity in the operation, administration and management (OAM) of HCNs demands the development of new network selfmanagement techniques. In this sense, key features to intelligently and autonomously automate network management procedures are required. Self-Organizing Networks (SON) [2] [3] paradigm is proposed to automatically and cleverly manage mobile cellular network procedures, helping to reduce both, CAPEX and OPEX. SON mechanisms help engineers to reduce time and effort to plan, optimize and troubleshoot cellular networks. Some early studies in this field are [4] [5] [6] [7] [8] [9]. Another challenge for operators is the high number of mobile connections originated at indoor environments, i.e., home, work, shopping malls, etc., especially in commercial and corporate scenarios. Recent market surveys [10] demonstrate that around 80% of all mobile broadband traffic is consumed by users located indoors. These places normally present coverage and capacity issues due to the poor signal quality, which degrade the end-user Quality of Experience (QoE). To solve or reduce the impact of these indoor issues, operators are proposing small cells. In particular, they are deploying low-cost radio base stations called femtocells [11]. These devices are small versions of standard macrocells which have special and specific characteristics (e.g., low-cost, short-range, open/close control access, etc.) compared to other base stations. Indoor environments present hard and difficult conditions to self-manage these femtocell networks due to the unplanned deployments, cell overlapping, lack of coverage, interference, etc. This fact has stimulated research activity in the field of parameters self-tuning [12] [13] [14] [15] [16] [17]. Additionally, data traffic and local user densities present temporal and spatial concentrations. In these conditions, indoor environments might suffer from serious network problems because most traffic could be located in the same femtocell(s) during short periods. Hence, some femtocells could be overloaded while others are low-loaded. An example of these temporal and spatial variations is easily found in shopping malls, where a temporal spectacle or event could gather many people interested in taking pictures to share them in social networks for a while. A simple solution to support these situations could be to plan the network to offer the maximum expected resources all the time. Nevertheless, this solution would largely increase CAPEX. As a consequence, new SON mechanisms focused on offloading those temporal overloaded femtocells to avoid or solve these situations and to guarantee the end-user QoE are required. In parallel, software applications and electronic devices evolve rapidly. Smartdevices such as mobile phones are able to provide terminal-centric information thanks to the large number of integrated sensors (accelerometers, barometers, GPS, etc.) and applications. Others like surveillance cameras provide an overview of a place (e.g., Introduction 3 number of people in a room or restaurant). All of them are able to collect context information in real time and share it with the mobile network with low delay thanks to the high-speed communication infrastructures (optical-fiber, xDSL, etc.). In this sense, this context information from sources out of the mobile networks could provide additional valuable information to SON mechanisms beyond traditional network indicators (i.e., alarms, counters, etc.). Some early studies focused on context-aware SON algorithms are [18] [19] [20] [21]. 1.2 Preliminaries This PhD thesis has been developed within the Ingeniería de Comunicaciones research group (GIC, TIC-102), in the framework of four research lines. Firstly, the main project supporting this PhD thesis was the MONOLOC project [22]. MONOLOC has focused on the development of an advanced platform for the management of mobile and next-generation heterogeneous networks with indoor user positioning. The project’s consortium was composed by leading members of the mobile communication industry: Alcatel-Lucent and Grupo Innovati; and academia: University of Málaga (UMA), Universidad Carlos III de Madrid (UC3M) y Universidad Politécnica de Madrid (UPM). The project provided an innovative indoor solution based on the combination of user positioning calculation, self-management of small cells and location-based applications. Its aim was to identify and develop tracking techniques especially for new mobile technologies and use them to the dynamic self-management of the network, being this able to auto-configure, optimize and heal itself. In addition, location-based applications were studied and deployed in live scenarios. In this context, strong and reliable network architectures were developed in order to assure the support to these specifications. Within MONOLOC project, UMA was responsible for defining the system architecture and for devising location-based SON mechanisms. Secondly, the GIC has an steady collaboration with a french partner: École Supérieure d'Ingénieurs en Informatique et Génie des Télécommunications (ESIGETEL). In this framework, a professor from ESIGETEL has done several stays at UMA and this PhD candidate has also visited ESIGETEL. The result is a joint research on indoor RFID-based positioning systems. Thirdly, the research in this PhD thesis was also part of a project called “Técnicas adaptativas de gestión de recurso radio en redes B3G”, funded by Junta de Andalucía 4 Introduction (Proyectos de Investigación de Excelencia), which is related to the development of novel self-optimization techniques, especially focused on the application of reinforcement learning techniques to Mobility Load Balancing and Mobility Robustness Optimization. Finally, this PhD thesis has also been related to a project called “Gestión integral avanzada de funciones SON (Self-Organizing Networks) para redes móviles futuras”, funded by Junta de Andalucía (Proyectos de Investigación de Excelencia), which aims to design advanced coordinated SON mechanisms. 1.3 Research objectives The aim of this PhD thesis is the design and development of novel SON mechanisms for open access femtocell mobile networks in commercial and corporate indoor scenarios, focusing on the mobility load balancing (MLB) use case. This means, the implementation of algorithms to offload temporary overloaded femtocells to lowloaded femtocells due to the high concentration of users in temporal and spatial domain, avoiding slow adaptive processes. Additionally, some of these methods would be supported by context information in order to improve the network performance. Figure 1.1: Research lines. To accomplish these goals, three main research lines are studied: 1) a framework and architecture for commercial and corporate environments to integrate context data into the mobile network, 2) a context source focused on the provisioning of indoor positioning and 3) MLB techniques to solve temporary and focused network congestion issues in open access femtocell networks (visual description in Figure 1.1). Simulation Field-Trial Context Awareness Self-Organizing Networks Indoor Positioning Systems Context-aware SON Framework Mobility Load Balancing Introduction 5 Initially, a framework for context-aware SON systems is proposed for commercial and corporate small cell networks. This framework is the base for the implementation of an OAM architecture which integrates and manages context data. It also supports the management of the SON mechanisms. Here, the main objectives are: • To propose a context-aware SON framework to integrate context data into SON systems. • To design an OAM architecture to support context-aware SON systems at commercial and corporate small cell networks. Secondly, an RFID-based (Radio Frequency IDentification) indoor positioning system is designed as a context source. The position of the terminals is an example of context information that could be supplied to the SON system to improve its performance. Note that the performance of the SON system would be affected by the accuracy of the indoor positioning system. Here, the main objectives are: 1. To study RFID-based techniques for fusion methods when having multiple antennas in the receiver. 2. To assess the trade-off between the number of active tags and the number of antennas in the receiver. 3. To study techniques for integrating cellular technology information into the RFID-based indoor positioning system. 4. To carry out a radio frequency measurement campaign analyzing both RFID and cellular signals. Finally, context-aware MLB mechanisms are designed to move terminals from overloaded cells to low-loaded cells. These methods study the singularities of femtocells, the characteristics of indoor environments and the mobility pattern of indoor terminals to achieve their goals. For that purpose, the femtocell transmission power is adjusted in order to force the handover of a terminal from its serving femtocell to a neighboring femtocell. Here, the main objectives are: 1. To design novel SON mechanisms for offloading overloaded open access femtocells at commercial and corporate indoor environments. 2. To mitigate temporal and focused overloaded situations at commercial and corporate indoor scenarios due to the high concentration of users in temporal and spatial domain. 6 Introduction 3. To integrate context-aware data into SON algorithms and to assess the benefits of processing that information. 4. To evaluate the impact of indoor positioning system accuracy on location-based SON algorithms. The evaluation of these systems and methods are demonstrated and assessed in a simulator and in a real testbed. 1.4 Document structure The organization of this PhD thesis is depicted in Figure 1.2. The first chapter corresponds to this introduction. Chapter 2 provides a brief description of the required technical background to follow the rest of the chapters. This part comprises an introduction to the LTE architecture, the SON paradigm, femtocells and context information. The relationship between each topic is also covered. Figure 1.2: Organization of chapters. Chapter 3 proposes to apply the context-awareness concept to SON mechanisms. Firstly, the context-awareness in SON is introduced. Then, the related work and problem description of this topic are presented. Afterwards, a framework for contextaware SON in indoor environments is proposed. This framework is supported by the designed OAM architecture and the evaluation of a specific use case. Finally, the conclusions and perspectives of this chapter are detailed. Introduction 7 Chapter 4 presents an indoor positioning system as a possible context source for context-aware SON systems. Firstly, this chapter introduces a review of the state-ofthe-art regarding indoor positioning systems. Then, the problem description is formulated. Afterwards, the proposed indoor positioning techniques based on multiantenna RFID system and mobile cellular signal are described. Subsequently, the accuracy of these techniques is evaluated in both simulated and real scenarios. Finally, this chapter includes the conclusions of the study. Additionally, this chapter refers to Appendix A and Appendix B where the characterization and the signal assessment of RFID and cellular technologies are carried out, respectively. Chapter 5 develops MLB mechanisms for commercial and corporate LTE femtocells with open access but they could be also adapted to other cellular technologies (UMTS or GSM). Firstly, it introduces MLB use case and the state-of-the-art. Then, it depicts the problem description. Afterwards, MLB methods are designed focusing on femtocell characteristics and context information, in particular the position of the terminals. Subsequently, these systems are evaluated in both a dynamic LTE system-level simulator and in a field-trial. Finally, the conclusions of these methods are discussed. Additionally, this chapter refers to Appendix C where a description of a fuzzy logic controller (FLC) is detailed. Chapter 6 summarizes the main conclusions of this research, proposes future lines of action and lists the publications supporting this PhD thesis. To conclude, Appendix D includes a summary of this PhD thesis in Spanish. 14 Technical background These SON functions could be located at different OAM levels in the 3GPP OAM architecture as Figure 2.3 shows. They could be either centralized or de-centralized. Figure 2.3: Location of SON functions in the 3GPP OAM architecture [8] [26]. Self-optimization is one of the most important functionalities in SON because it ensures the network operates to its best level of efficiency once the base stations have been deployed. Since this PhD thesis is focused on self-optimization, the following section will further describe this function. 2.2.1 Self-optimization Self-optimization is necessary due to the changes that the environment around the base station might suffer once it is installed and well-configured. Some of these changes are related to: - Changes in deployments: Modifications in a base station could affect the others. New base stations could be integrated in the network, a base station could be optimized, etc. - Changes in traffic patterns: The concentrations of users evolve over time. Some examples could be beaches on summer season, sporting events, etc. - Changes in propagation characteristics: This could arise or have a significant effect when new buildings are erected or demolished, leaves from trees fall in autumn, etc. The main self-optimization use cases are the following [2] [30]: Technical background 15 • Coverage and capacity optimization: parameters such as transmitter power levels and antenna tilts are adapted to maximize coverage and minimizing interference levels. • Energy saving: operators aim to reduce the power consumption as well as carbon dioxide emissions. Hence, there are a number of option to accomplish this goal such as decreasing active carries for off-peak times, switching cells to sleep mode, etc. • Mobility robustness optimization: operators are interested in robust mobility and handovers within the mobile network to avoid the disruption of the service. It would minimize the number of unnecessary handovers, dropped calls, etc. • Mobility load balancing: congested cells should transfer calls to other cells with spare resources. Antenna tilts, handover margins and transmitter power levels are the main adapted parameters. This PhD thesis is focused on the Mobility Load Balancing (MLB) use case. A graphical explanation of this functionality is depicted in Figure 2.4. Left image shows an overloaded situation. Here, left cell suffers congestion which decreases network performance and increases the number of unsatisfied users. Conversely, adjacent cells are less loaded in the network. Under this situation, as the right image illustrates, MLB forces users from its serving cell (left cell) to handover to target cells (right top cell) by changing configuration management parameters (e.g., transmitter power level of serving cell or handover margins from serving cell to target cell). Figure 2.4: Mobility Load Balancing use case. 16 Technical background A simple solution to support these congested scenarios could be to plan the network to offer the maximum expected resources all the time but it would largely increase CAPEX. Nevertheless, by applying MLB in the network, this temporal congestion would be overcome without further investments. 2.3 Femtocells Nowadays, most cellular traffic is generated indoors (e.g., home, work or shopping malls), where there is often a lack of coverage or insufficient QoS [10]. In these cases, network operators are offering small cells to overcome the indoor issues, being femtocells the main deployed base stations. The deployment of femtocell networks provides significant benefits to network operators and their clients. The most common benefits result in: - reduction of operational costs. - potential macrocell offloading gain. - improvement of the end-user QoE. - terminals save battery and increase lifetime. However, the smooth integration and maintenance of femtocells into classical macrocell networks is an important challenge for operators. 2.3.1 Characteristics Femtocells are simple and small versions of standard macrocells, manufactured to be deployed at indoor environments. These low-cost and low-power devices cover areas of several meters, work in the licensed frequency band and are under operator management. Additionally, femtocells are connected to the operator’s network by a broadband connection, i.e., through client’s broadband backhaul (e.g., cable or xDSL) as Figure 2.5 shows. Thanks to this singularity, they are plug and play devices and operators offer unplanned deployments, which means, the client is free to locate the femtocell anywhere. For this reason, default femtocell transmission power is the maximum transmission power. Another important characteristic of femtocells is the control access (see Figure 2.5). The accessibility could be close, i.e., access is restricted to subscribers. The Closed Subscriber Group (CSG) is in charge of registering those specific users. Home femtocells are an example of close access, as they are deployed for private usage. In an Technical background 17 open access scenario (e.g., public buildings), any terminal could be connected to the femtocell without any restriction, like a macrocell. Enterprise femtocells are usually open access. Hybrid access is also supported. Figure 2.5: Femtocell scenario. Since femtocells operate in residential or enterprise spaces and users may move between indoor and outdoor scenarios, a handover process is required between macrocells and femtocells to keep the service alive. Femtocell manufacturers have mainly developed two types of femtocells. Focused on home environments, the datasheet in [31] presents a femtocell with a baseband capacity of maximum 4 simultaneous connected users (voice call or data session). Thinking in crowed indoor scenarios such as offices or shopping centers, the datasheet in [32] provides a solution up to 64 (i.e., 8, 16, 32 or 64 users) active users in the femtocell. Based on the maximum capacity of the femtocell, the Admission Control (AC) algorithm is in charge of managing the access of new users. The AC algorithms are not standardized, i.e., different vendors would run different AC schemes. Then, the AC could reject those attempt connections once the maximum number of connected users is reached or drop specific connected users (e.g., per service, time connected, etc.) even if there are free radio resources. Note that this limitation is independent of the Scheduler, the availability of radio resources or the circuit/packet-switched channel. Additionally, it is also beyond the cellular technology (GSM, UMTS or LTE). For example, the 3G femtocell in [31] supports four devices at the same time. 18 Technical background Figure 2.6: Persistent allocation for VoLTE. Focusing on the LTE technology (IP-based technology), the resources could be scheduled and allocated in the next frame if the current frame is full (depending on their priority). However, VoLTE service is special. Semipersistent Scheduling (SPS) [33] would be in charge of managing and allocating VoLTE traffic. In this sense, the mechanism could assign predefined chunk of radio resources for VoLTE users with interval of 20ms (see Figure 2.6). It would decrease control channel overhead, reduce jitter and increase the QoS. Nevertheless, before scheduling a new connection, it must be accepted by the AC scheme and, as previously explained, that new user is blocked if the maximum femtocell capacity is reached (normally from 2 to 64 users). Therefore, even if there are available radio resources in the current frame or in the following one, that new user would be rejected by the AC. Note that other AC schemes could accept this user but an active user is dropped or tried to be handover to another cell. As an example, in [34] the maximum number of active users is restricted to 8 users: either VoLTE users or other data bearer per QCI (Quality Channel Indicator). 2.3.2 LTE architecture The 3GPP defines a base station as eNB while femtocells are defined as Home eNB (HeNB). HeNB is integrated in the LTE architecture as a new NE. The functions supported by the HeNB shall be the same as those supported by an eNB (with the possible exception of NAS Node Selection Function - NNSF) and the procedures run between a HeNB and the EPC shall be the same as those between an eNB and the EPC. The E-UTRAN architecture may deploy a Home eNB Gateway (HeNB GW) to allow the S1 interface between the HeNB and the EPC to scale to support a large Resource Units 20 ms Time Next 20 ms 1 ms Persistent Allocations for VoLTE repeat every 20 ms Technical background 19 number of HeNBs. The HeNB GW serves as a concentrator for the control plane, specifically the S1-MME interface. The S1-U interface from the HeNB may be terminated at the HeNB GW, or a direct logical user plane connection between HeNB and S-GW may be used. This logical architecture is shown in Figure 2.7. Figure 2.7: E-UTRAN HeNB logical architecture [24]. The E-UTRAN architecture with deployed HeNB and HeNB GW is depicted in Figure 2.8. The HeNB GW appears to the MME as an eNB. The HeNB GW appears to the HeNB as an MME. The S1 interface between the HeNB and the EPC is the same whether the HeNB is connected to the EPC via a HeNB GW or not. Release 9 does not support X2 connectivity of HeNBs. However, X2 connectivity of HeNBs is supported in LTE-Advance (Release 10 [24]). Figure 2.8: Overall E-UTRAN architecture with deployed HeNB GW [24]. 20 Technical background 2.3.3 SON for femtocells SON techniques play an important role at commercial and corporate indoor scenarios. Despite some classical SON mechanisms could be implemented indoors, their performance could be degraded compared to their application on macrocell scenarios. The main reasons are the particularities of femtocells, the characteristics of indoor scenarios and the influence of users’ mobility pattern: 1) Particularities of femtocells: a) The number of simultaneous users connected to a femtocell is very restricted in comparison to macrocell stations. In this way, femtocell equipment reduces the number of simultaneous users (from 2 to 64 users). It must be taken into account for the application of SON mechanisms. b) The deployment of femtocells is normally unplanned, contrary to the deployment of macrocells. As a consequence, some femtocells can be highly loaded while others are low loaded. In this sense, SON mechanisms would ensure a good cellular network performance. c) As low-power base stations, the coverage area of femtocells is short-range. That means, depending on the users’ mobility pattern, users could easily handover several times from one femtocell to another. That increases the network signaling overhead and femtocell congestions may occur. Therefore, the time to trigger SON methods should be as fast as possible to detect and solve these situations. d) Femtocells are normally placed at fixed positions. However, the customer could freely (i.e., without operator’s approval) and easily move a femtocell to another place as a plug and play device. By moving the femtocell, the customer might intend to solve specific issues, such as a lack of coverage or low throughput in an area, but other issues could arise with this change. In these cases, SON mechanisms are required to ensure a good network performance by automatically adjusting femtocells configuration parameters. 2) Characteristics of indoor environments: a) Indoor radio channel conditions suffer continuous changes due to multi-path reflections, wall obstacles, number of people, etc. Those situations are a challenge for SON mechanisms. b) Occasional events or unpredictable occurrences provoke unexpected temporal overload situations in surrounded femtocells, which could negatively affect Technical background 21 network performance. For example, restaurant zones would increase the traffic at launch/dinner time while sales period would create occasional hotspots in shops. In all these situations, the femtocell in charge of serving that area would be overloaded. c) Due to the geometry of indoor environments, a tridimensional deployment of femtocells on the different stages of a buildings, airports, shopping malls, etc., is often required. Based on this, classical SON mechanisms at indoor environments should take into account these characteristics before their implementation in femtocell networks. However, the design of new SON mechanisms for indoors would provide optimal solutions for these networks. For example, from the point of view of MLB, the number of active users at each femtocell should be analyzed in addition to the availability of radio resources to allocate new connections. 2.4 Context information A new concept is emerging thanks to the ubiquity and quality of electronic devices with stable Internet connections: context-awareness. Location, activity, time and identity are the primary context types for characterizing a situation of a particular entity. “Context is any information that can be used to characterize the situation of an entity. An entity is a person, place or object that is considered relevant to the interaction between a user and an application, including the user and applications themselves [35]”. Nowadays, smart-devices like smartphones integrate many radio frequency (RF) technologies as WiFi, Bluetooth, NFC and even UWB, in addition to the classical cellular technologies: GSM, UMTS or LTE. Furthermore, they are supported by dozens of integrated sensors like compass, barometer, gyroscope, GPS, etc. Besides, other systems like surveillance cameras are able to characterize an element or scenario. Valuable information about the people distribution, their position, the weather, etc. could be obtained through a digital image processing. Finally, the imminent development of apps (applications) in smartphones and other smart-devices achieve specific information of users such as the calendar events, the battery level of the 22 Technical background terminal, the age, etc. or particular information about the environment such as the temperature, etc. 2.4.1 Indoor positioning systems One of the key challenges for the mobile market is to find killer applications for the new terminals and data plans in order to increase service providers, manufacturers and application developers’ revenues. In this field, Location-based Services (LBS) will support new applications on several fields: health-care, advertising, emergencyresponse, etc. Most suitable scenarios for LBS are indoor areas (e.g., airports, hospitals, malls, etc.) given the large concentration of users in these environments. Based on this, the challenging deployment of indoor positioning techniques is a hot topic. In this sense, smartphones could be well-located (position error below one meter) thanks to the analysis of wireless technologies like Infrared Laser, UWB, etc. The drawback is that these kinds of systems are very expensive due to their low position error and high accuracy. Nevertheless, systems based on the analysis of the received signal strength indicator (RSSI) of wireless technologies such as RFID or WiFi [36] or cellular technologies [37] could reduce these expenses although the location accuracy would be degraded (position error of few meters). Additionally, the synergy of those technologies and the wide variety of integrated sensors into the smartphones could enhance the performance of these positioning systems in terms of location accuracy [38]. 2.4.2 Context-aware SON SON mechanisms are based on the analysis of the mobile network alarms, counters and metrics (e.g., handover failure rate). Indoor environments present specific characteristics (e.g., high level of coverage overlapping, rapid performance changes, user distribution variability, etc.) which make these SON methods solely based on the analysis of those network indicators, prone to take a long time to get the best configuration parameters. Context-awareness provides profitable information about the actual status of the environment such as smartphones position, mobility patterns, etc., which properly managed could help SON mechanism to speed up its convergence time, to get the optimal configuration parameters, to improve the network performance, etc. Technical background 23 2.5 Conclusions LTE provides reduced latency, increased peak data rates and scalable bandwidth while backwards compatibility with legacy mobile cellular technologies. Additionally, SON paradigm provides these networks with intelligent and autonomous procedures to automatically overcome network degradations and failures. Indoor environments with femtocells deployments are one of the most challenging scenarios to be managed. Here, SON mechanisms play a key role to guarantee coverage and capacity while ensuring the end-user QoE. These SON mechanisms are only based on the analysis of the network performance indicators. However, the expansion of smart-devices, systems and applications could provide valuable additional information from external sources to the network management layers, being important and useful for SON mechanisms. 30 Context-aware SON framework picture with him/her or ask for an autograph, provoking a huge concentration of users in a small area where most of them use their smartphones to share that moment in social networks. • Image/Video: Photographs or video images are collected from online photosharing or video-sharing as Instagram, Youtube, etc., or from surveillance cameras installed on a specific place, providing information about its context. For example, images could provide valuable green context information if SON algorithms turn off a small cell when no one is close to it, saving useless energy. These systems usually take some time to process all these data. • Operator: An operator manually introduces data into the CAM, about what he or she is monitoring. • Other sources: New smart and intelligent devices, such as drones, or even current context information that is not taken into account in current state-ofthe-art SON algorithms, like weather, could be also very helpful in the future in this kind of systems. Notice that several sources can provide the same type of final information, e.g., number of people in an area. This redundancy is very important in case of CAM connection loss or access restrictions to any of these sources. As mentioned before, some of these sources provide automatic and instantaneous information to the CAM while some others need some time to collect and send it to the CAM. Depending on the time required for data collection and transfer, the context-aware SON algorithms will run at set intervals as follows: • 1 hour: Context data is supplied to the CAM every hour, e.g., by an operator. • 30 minutes: Image/video processing systems need time to get the pictures. • 15 minutes: Sophisticated image/video processing systems are faster. • 5 minutes: Personal devices could have delays in the information delivery. • 1 minute: More accurate and online updating technologies could reduce this time to less than a minute. For example, positioning systems, social networks, personal devices, etc. • Predictions: The information is proposed to the system once an event is programmed, for example in social networks or into a personal device calendar. The proposed framework could present some challenges in the communication with external sources: Context-aware SON framework 31 • Signaling overhead: It depends on the communication protocols, the transmitted message and the context information. The transmitted message consists of attribute-value pairs (e.g., XML, JSON, etc.) to easily manage the information. The well-known Transmission Control Protocol (TCP) is in charge of transporting reliable data at reduced overhead. Also, more advanced (SCTP, DCCP, etc.) or simple (e.g., UDP) protocols could reduce even more the signaling cost. <index> <context> user_location <id> 3589870105 <time> 1413124586 <x> 14.2547 <y> 85.6214 <end> Figure 3.2: Example of context message. In order to provide an assessment of the cost, a preliminary message formatting is described to estimate and quantify the volume of data per message (see the example in Figure 3.2). This is a simple context message to provide the user location to the CAM. Each character is computed by 1 byte (there are 84 characters on the message). Consequently, the transmitted message is composed by this context data (84 bytes) plus the communication headers (TCP+IP=40 bytes). Therefore, for this kind of message, 124 bytes are transmitted from the external sources to the CAM. Mention that, any other message or communication protocol is also supported by the system (e.g., JSON). • Delay: The lack or delay in the information transmitted from the external sources would degrade the system performance as SON mechanisms will not properly work. However, the high speed of communication infrastructures (e.g., xDSL), the low signaling overhead and the redundancy of context information supplied to the system from multiple sources, could overcome this limitation. • Security: Most of the information supported to the CAM is public data, hence, no strategies or techniques of information security are mandatory. Nevertheless, the information from private sources or private data would be encrypted. Furthermore, it is crucial to confirm the veracity of the information against possible issues (e.g., malicious modification, delayed data, etc.), while the system shall be able to reject false or incomplete samples. In case the CAM is not able to do it, the context-aware SON methods might degrade the network performance. In that case, the 32 Context-aware SON framework context-aware SON algorithms will be disabled to come back to conventional SON mechanisms (algorithms only use information from the mobile network). The location of the CAM could be in middle layers of the operator’s OAM architecture (e.g., DM) to reduce the signaling overhead. However, other solutions like NEs or indoor local mobile network management systems are possible at indoors. 3.4.2 OAM architecture1 Following the idea of indoor local mobile network management system, the design of a novel OAM architecture that interacts with the classical OAM layers is required. The proposed OAM architecture is defined by several interrelated entities, where different approaches can be adopted in order to establish its functional scheme: centralized, (where a unique entity is in charge of managing the rest of the elements); distributed (peer-to-peer); and hybrid. The proposed solution follows a hybrid scheme, a combination of the characteristics of centralized and distributed schemes. Here, some mechanisms are completely local while others require coordination among distributed entities. It allows easy reuse of classical centralized OAM architecture, while the implementation of distributed mechanisms is also supported. The 3GPP OAM architecture [26] is maintained, adding new capabilities, functions, entities and interfaces to it. The placement of SON functions in the standard OAM elements (see Figure 3.3 on the left: NM, DM and NE) follows a similar scheme to that presented in [8]. The functions involving a specific subnetwork can be implemented at the DM layer. For functions involving more than one subnetwork, they will reside on higher layers of the OAM hierarchy. Conversely, distributed SON functions would be placed at NEs (e.g., small cells). These levels at the OAM architecture chain are directly related to the time span for monitoring/configuration and also the level of abstraction over the network layers (see Table 3.1). However, even for the lowest layer standard centralized entity (DM), time spans (in the range of hours) are still large. Also, DM usually operates nonoverlapped subnetworks covering wide areas. Hence, a novel additional OAM functional block, the OAM Context-Aware System (OCAS), is proposed to support innovative context-based SON mechanisms. This new proposed centralized entity is implemented at the lowest levels of the OAM hierarchy, being in charge of managing the set of small cells of one specific indoor area. 1 This work has been performed in collaboration with Sergio Fortes Rodriguez. Context-aware SON framework 33 Table 3.1: Characteristics of 3GPP OAM layers [8]. Task Parameter Abstraction Time Span NM Planning Vendor independent weeks/month EM/DM Network Operation Vendor independent/specific hours/days NE Element Configuration All parameters secs/mins 3.4.2.1 Functional architecture The proposed OAM architecture is shown in Figure 3.3. The standard 3GPP OAM architecture is represented (left square) containing the standard NM, EM/DM and NE elements. These are connected to the OCAS by newly defined interfaces (see Section 3.4.2.2 “Interface protocols”), which implements the following roles: • To register available context services (CSs) and obtains context information from them. • To implement context-aware SON functions. • To act as coordinator between the OAM elements of the mobile network, context-aware SON algorithms and CSs. It propagates the results of the context-aware SON algorithms to the OAM standard elements for their authorization to apply the decided commands in the network. Then, these commands may be applied through standard OAM elements or directly to the devices by the OCAS itself depending on the operator policies. Additionally, monitoring and reporting functions (M/R) can be incorporated into the UEs, so they can directly report to the OCAS information of the network status or their location. This M/R capability can be part of the context-based applications present in the terminals (e.g., navigation apps) or being implemented by means of directly invoking functionalities in the terminal API (Application Program Interface). The described OCAS roles are distributed in different functional elements, which allow a better insight into the defined functionality: • SON Algorithmic Unit (SAU) implements the local SON algorithms in the system. It can contain multiple interdependent SON functions for selfconfiguration, self-optimization and self-healing. If multiple SON use cases are implemented, it would be also responsible for the proper coordination and trade-off between the different SON use cases and mechanisms by the SON Coordination Layer, being its particularities dependent of the specific use cases implemented. One benefit of the integration of multiple SON mechanism in the 34 Context-aware SON framework SAU is that it supports the use of the same context sources (as well as network indicators/measures) for the multiple use cases implemented, reducing the possibility of collisions generated by using different information sources, as well as allowing a straightforward coordination between techniques. • Context Service Registering (CSR) is in charge of the incorporation and authentication of different sources of context information into the Registered Context Services Database. For a CS to be included, the main parameters for the information exchange with the OCAS have to be defined: IP address and format characteristics for the communication with the CS. These parameters have to be compiled in a set of profiles to be used by the Context Data Collector Unit (previously referred as CAM) in order to communicate with the different CSs available. • ACtuator Unit (ACU) configures the network elements with the new parameters calculated by the context-aware SON algorithms, directly or by the standard OAM pile through the OAM Coordination Unit. • Context Data Collector Unit (CDCU) gathers the information coming from the CSs or the terminals registered in the system. • MEasurement Unit (MEU) obtains information from the network elements, by direct network element connection or through standard OAM elements using the OAM Coordination Unit. It is also in charge of the possible acquisition of direct network measurements from the UEs (e.g., received power levels, etc.). • OAM Coordination Unit (OCU) serves as the interaction element between the OCAS and the OAM standard architecture. It translates the configuration orders coming from the ACU into commands for the operator’s OAM tools and it turns the OAM monitoring into a format usable by the MEU. Furthermore, it also supports the configuration of any of the OCAS functionalities by commands coming from the standard OAM architecture elements as well as by Local Network Manager Agents. • Local Network Manager Agent (LNMA) represents the specific operator or administrator that may be required to manage the OCAS. The LNMA will have two main capabilities: - It may register, via the CSR, new CSs to be used by the OCAS. - It may alter the policies and/or functionalities of the OCAS via the OCU. This capability should be restricted through the permissions defined in the Access Identities and Privileges Database to avoid erroneous/malicious access. Context-aware SON framework 35 Figure 3.3: Proposed OAM architecture. 36 Context-aware SON framework 3.4.2.2 Interface protocols According to the proposed architecture, the new main OCAS block introduces selfmanagement at the local mobile network. Consequently, new interfaces, protocols and applications should be implemented in order to coordinate this system with the rest of the OAM architecture as well as to measure and modify network devices: • NM-OCAS and DM/EM-OCAS are used for the coordination between OCAS and the elements of the operator’s OAM core. • NE-OCAS exchanges monitoring and configuration messages between the small cells and the OCAS through three different interfaces: - NE-OCAS/MEU focuses on monitoring and providing information about counters, alarms, KPIs, etc. to the MEU. - NE-OCAS/ACU carries direct configuration commands or files to the NEs. - NE-OCAS/OCU transports both monitoring and configuration messages when these cannot be directly sent/received to/from the NE by the OCAS blocks. • CS-OCAS interfaces communicate information from the CSs to the OCAS. This could be context-awareness messages in order to support the context-based SON functions (through the CS-OCAS/CDCU Itf) or the procedure to register a new context service in the CSR (through the CS-OCAS/CSR Itf). • LNMA-OCAS/OCU interface allows (subject to operator permission) the configuration of the OCAS system by the LNMA. In turn, LNMA-OCAS/CSR serves for the manual registration of a CS by the LNMA. • UE-OCAS logical connections send direct UE monitoring information to the OCAS (through the UE-OCAS/MEU interface) and UE provided context information (by the UE-OCAS/CDCU) that may be required for the SAU. For non-cellular external context services, this interface would be UE-dependent. Therefore, it may be only available for specific UE models such as smartphones. All the interfaces connecting the OCAS with the OAM architecture should follow the same standards as defined for 3GPP interfaces, being mainly based on TR-069 and XML [43]. CS-OCAS and UE-OCAS interfaces, however, are defined with elements that are independent of the mobile communications OAM network, as they are encapsulated on the user plane, so any communication protocol (over IP) can be freely defined for these data flows. Context-aware SON framework 37 3.4.2.3 3GPP LTE/LTE-A femtocell network The physical implementation of the proposed architecture for LTE/LTE-A systems, is centered on the case where the deployed small cell are open access femtocells (HeNBs). These are chosen because their limited capabilities, high vulnerabilities and wide usage make this case specially challenging and comprehensive from an OAM perspective. OAM-SON functionalities follow the standardized 3GPP architecture [26], with the novel addition of the OCAS, which implements local context-based SON functionalities. OCAS implementation can be local (if performed by hardware connected to the same LAN – local area network) or remote (by an external hardware connected to the system via the Internet). Remote solutions have high versatility in terms of using existing or leased equipment. However, the need of exchanging a high amount of information through the often limited network backhaul, highly encourages the adoption of local implementations as the one it is adopted here. Challenges for this approach include the need of OCAS additional on-site hardware, its installation and maintenance, although the related cost is expected to be minimal over the total deployment expenses. Figure 3.4: Proposed OAM architecture for LTE/LTE-A femtocells. 38 Context-aware SON framework Figure 3.4 presents the local physical implementation, for LTE/LTE-A commercial or corporate femtocell scenarios. The DM role is implemented by the HeMS. HeNBs user and control planes connect to the operator’s core through the S1 interface, while the X2 interconnects the femtocells for distributed cooperation (in LTE-A). The defined logical links are implemented by physical connections as follows: • UE-OCAS and NE-OCAS: The information transmitted from the UEs to the OCAS is sent through the Uu interface to the femtocell. This data (as well as the particular commands/information from the stations transmitted by the NEOCAS interface) is then retransmitted through the LAN to the OCAS. • CS-OCAS: The interface used to transmit the context information (in case it is not directly obtained from the terminals) is implemented by the CSs through the LAN or the Internet connection to the OCAS. • NM-OCAS and DM-OCAS: The information between the OCAS and the operator’s core (i.e., the highest elements of the standard OAM layers: DM, NM) is sent by the router through the backhaul to the operator’s core. The use of LAN for exchanging data between the OCAS and the UEs greatly minimizes the traffic in the backhaul and the operator’s core at reduced delay. This traffic local breakout has been envisaged by standards like Local IP Access and Selected IP Traffic Offload (LIPA-SIPTO) [44] or projects [42]. 3.4.2.4 Domain responsibility and security This characteristic refers to the commercial/legal entity in charge of managing the proposed OAM architecture. The responsible entities include the mobile network operator, the user/administrator of the local system or a third party. Even if the considered scenarios are essentially local, the small cells are currently part of the mobile network operator infrastructure and make use of its radio spectrum. Therefore, as SON will alter the cells configuration, OCAS should remain on the operator’s domain but they could transfer the responsibilities of the local-centralized system to a local user/administrator. The connections between the OCAS and the CSs shall avoid the disclosure of network status information that may be sensitive. Thus, the extent, authenticity, accountability and correctness of the information exchanged will be critical. Standardization on the CS-OCAS and LNMS-OCAS and related processes may be necessary to limit this issue. Context-aware SON framework 39 3.4.3 SON use case In this subsection, some baseline MLB algorithms are presented to verify the context-aware SON framework. For that purpose, a novel design is proposed to integrate context-awareness into these MLB algorithms. It is shown that context-aware SON compared to baseline algorithms achieves better network performance and reduces the optimization time to reach the permanent regime (steady state). 3.4.3.1 Mobility load balancing use case MLB is a use case defined by the 3GPP [45]. It has the ability to manage and direct voice and data traffic from loaded cells to the most suitable cell, independently of the end layer (macrocell or small cells) or RAT (LTE, UMTS or GSM). Normally, most non-context-aware algorithms that follow an adaptive process to optimize the cellular network work properly (i.e., a few iterations and short time to reach the optimal network parameters) when users are homogenously widespread around the coverage area of the loaded cell, due to the fact that neighboring cells equally catch some of those users [16] [17]. However, if the users’ distribution is concentrated in certain part of the cell coverage area (top Figure 3.5), most noncontext-aware mechanisms follow a similar process: all neighboring cells try to catch those users (e.g., by tuning cells transmission power as illustrated in Figure 3.5(a). The consequence of this act could be either the degradation of the network performance or the algorithm oscillation to reach the steady state. That conclusion will be seen in the evaluation subsection. Figure 3.5b illustrates an optimal solution to face that problematic situation. This scenario could be identified thanks to users and objects context-awareness. In this case, only the closest neighboring cell to the loaded cell should catch users. For example, by increasing its transmission power and, at the same time, the transmission power of the loaded cell is reduced. Those variations of transmission power are restricted to the cell transmission power range (e.g., from 0.016 to 250 mW). Additionally, some cellular methods to avoid high interference could limit that maximum value whereas other methods that analyze the QoS, limit the minimum value of transmission power to avoid coverage holes. Several MLB algorithms in open access femtocell scenarios have been developed in the literature. The Power Traffic Sharing (PTS) algorithm [16] has been chosen as a baseline method. Additionally, a modified version of Power Load Balance algorithm [15] is developed: Power Load Sharing (PLS). The objective of these procedures is to intelligently optimize an indicator or Figure of Merit (FM) while maintaining the network performance. 46 Context-aware SON framework bad QoE during the convergence time. Therefore, those users could leave their current cellular operators to look for better service (and, consequently, the operators’ revenues would be reduced). Figure 3.11: Convergence time (𝒕𝒐𝒐𝒕). 3.6 Conclusions A framework for context-aware SON has been presented, proposing different context sources and time intervals to get that information. Additionally, a novel OAM architecture extension for 3GPP-based SON is proposed for commercial and corporate indoor small cell scenarios to support the context-aware SON framework. An MLB mechanism has been taken as a baseline to be upgraded with extra data provided by external network sources. The results show that the proposed approach leads to a significant improvement in the response time by using position awareness. In consequence, the convergence time has been reduced about 50%. Subsequently, the optimal UR values have been enhanced, decreasing the number of dissatisfied users in the cellular network. These results and the qualitative analysis of the OAM impact in terms of signaling overhead and delay, demonstrate that placing most of the SON functionalities at the lowest levels of the OAM hierarchy is the right decision. 5 2.5 1 0.4 0.05 8 4 1.75 0.8 0.1 0 1 2 3 4 5 6 7 8 1 hour 30 min 15 min 5 min 1 min hours Time to trigger the algorithm CA-PTS PTS 4 3 1.75 0.3 0.05 7 4 1.75 0.7 0.1 0 1 2 3 4 5 6 7 8 1 hour 30 min 15 min 5 min 1 min hours Time to trigger the algorithm CA-PLS PLS 47 Chapter 4 4 Indoor positioning strategies This chapter presents a multi-antenna RFID-based indoor positioning system using active tags. In addition, it is studied the improvement of the positioning system accuracy by the integration and analysis of RF measurements from cellular technologies. For that purpose, an experimental analysis of simultaneous RF measurements from RFID and macrocellular networks is carried out. The structure of this chapter is as follows: Section 4.1 introduces current indoor positioning systems. Section 4.2 details the state of the art, focusing on wireless heterogeneous techniques. Section 4.3 presents the problem description. Section 4.4 describes the proposed indoor positioning strategies. Section 4.5 evaluates the accuracy of these indoor positioning systems. Finally, Section 4.6 includes the conclusions and perspectives of this chapter. 4.1 Introduction Nowadays the requirements of Location-Based Services (LBS) in terms of locationaccuracy at indoor environments go from few centimeters to several meters or even the detection of a room/office. Cost reduction of wireless devices makes RF (radio frequency) solutions good candidates for low cost indoor positioning systems. Several RF technologies have been proposed for providing indoor positioning such as UWB, which presents position errors of few centimeters, or RFID, which is in the range of few meters. However, accurate indoor positioning systems are still expensive and present high costs in terms of hardware price, deployment expenditures, computational cost and complexity, especially for very accurate solutions, such as UWB. 48 Indoor positioning strategies The identification of being inside a room or office could be easily performed with many RF-based indoor positioning systems. Nevertheless, other indoor scenarios like halls and corridors are commonly a challenge as they would require more accuracy. The use of multi-antenna devices could help to improve location accuracy in such a challenge scenarios and reduce the infrastructure costs. In parallel, a wide variety of wireless communication networks coexist simultaneously with human beings. These networks are the so-called HCNs where several wireless technologies surround daily life to create massive communication layers at different frequencies. These infrastructures provide a cloud of parallel information that could be, not only used for communication purposes, but also for enhancing accuracy and robustness of several, a priori, unrelated applications, such as indoor positioning. In this way, mechanisms for opportunistic radio positioning based on multiple technologies (e.g., WLAN together with GSM, etc.) have been devised as promising solutions. The integration of those technologies in current indoor positioning systems could lead to the required levels of accuracy for new applications at reduced costs. Based on these claims, the use of multi-antenna readers for RF-based indoor positioning systems to minimize infrastructure costs and achieving high accuracy is studied. Herein, RFID multi-antenna readers with active tags are selected as a low cost RF technology. In parallel, the combination of the RFID-based indoor positioning system with other low cost RF solutions is analyzed and developed in order to reduce the position error. Macrocellular networks, one of the most widely extended RF technologies, are selected. The main contributions of this chapter are the study and assessment of: • Techniques for data fusion when having multiple sets of RFID measurements, each one referred and collected by an antenna, at indoor challenge scenarios like halls and corridors. • Trade-off between the number of RFID tags and antennas in the reader. • Integration of cellular technology information into RFID-based indoor positioning systems. • Measurement campaign for an experimental analysis of both RFID and cellular technologies. Finally, real field measurements are used to evaluate the capabilities of the proposed indoor positioning systems. Indoor positioning strategies 49 4.2 Related work In indoor positioning and navigation fields, the large diversity of application environments leads to a large diversity of solutions. A summary of the main positioning technologies, techniques and their performances for typical applications are detailed in [48]. For these applications, several performance criteria can be defined, such as robustness, responsiveness, consumption, size, scalability, compatibility with human beings, accuracy or precision. Literature presented several RF-based indoor positioning systems. The authors in [49] and [50] evaluated Channel State Information (CSI-based) positioning versus RSSI-based positioning in three typical environments such as a research laboratory, a lecture theatre and a corridor. With the RSSI-based system the median position error is less than a meter for the laboratory and 2 m in the lecture theatre, while it is 2.5 m in the corridor. The cumulative distribution function (CDF) of the position error with 90% of probability is 1.5 m and 3 m for the two first environments, while it is almost 3.5 m for the corridor. Conversely, CSI-based method is the most accurate technique but the corridor environment is still harsher than the others; the 90th percentile of position error is more than 1.75 m while it is 1.5 m for the laboratory. The work in [51] proposed LuPI (Locating using Prior Information). It considered that human motion can be distinguished and recorded by radio information (RSSI deviation between different positions) and a pedometer (based on accelerometer embedded in a smartphone). LuPI utilized the RSSI and the sensor-based pedometer to build a RSSI variation space as prior information. In order to assess this solution, extensive measurements were made on the third floor of a middle-size building with different types of room/areas. In the analyzed corridor, the average error of the proposed system is 5.9 m. In a big room, the average error is 1.4 m, while it is 1.9 m in a small one. Focusing on RFID technology [52] [53] [54] and multi-antenna indoor positioning systems, several works can be found in literature [55] [56]. These works used the antenna diversity to assess the probability of a position when a tag was detected or not, or to provide Direction-of-Arrival (DoA) estimation by phase arrays in small covered areas, typically less than 10 m2. Other works studied the impact of the number of antennas and its configuration on location accuracy [57] or compared position estimation methods [58] [59]. The study described in [57] considered four different multiple antenna configurations: SISO, SIMO-MISO and MIMO. The best results were performed with MIMO and the average position error is 1.5 m with 3 dB shadowing standard deviation. When increasing the number of antennas from 2 to 14, the position error decreased from 2.3 m to 0.9 m. Conversely, [58] proposed three estimation methods for an increasing number of transmit antennas (2 to 4), the position error varies from 2 m to 1.5 m in 7x7 m2 room. Finally, [59] presented the 50 Indoor positioning strategies accuracy and stability results for RADAR, Area Based Probability (ABP), Simple Point Matching (SPM) and Bayesian Networks (BNs) algorithms. For each one, the CDF of the position error was calculated when averaging and not averaging the antenna data, and when applying a Gaussian distribution to the measured RSSI. The experimentation area was about half of a floor measuring 66.75x51.5 m2 and the best accuracy was 1.5 m. Experimentations at a desk level were small one-foot movements around a main center placement with accuracy of 0.9 m. In outdoor scenarios, the signal coming from macrocell base station can be processed in order to obtain just a rough estimation of a mobile terminal position (hundreds of meters). In indoor scenarios, some studies characterized GSM/UMTS signals from macrocells to check the reliability of building indoor radio maps for fingerprinting purposes [60]. Other works [37] [61] presented a fingerprinting technique based in the mapping and use of radio signals from femtocells for positioning terminals in indoor environments. The accuracy of those systems was few meters. However, such indoor cellular deployments are still no widely implemented and costly. The proposed option of combining RFID with macrocell technology measurements could successfully provide the level of accuracy required in RFID-based indoor positioning systems for pedestrian navigation in buildings. In this respect, the first works integrating diverse RF technologies for indoor positioning were presented a few years ago [62] [63] [64] where RFID, WLAN and GSM technologies were analyzed. The main purpose of combining different technologies is to increase the accuracy [65], or the density of the devices to be localized [66], or to overcome the continuity indoor/outdoor challenge [67], where these systems need to handle heterogeneous devices [68] and vertical handovers [63]. European Project WHERE2 [69] presented results of real-life experiments based on ZigBee and Orthogonal Frequency Division Multiplexing (OFDM) devices (emulating a multi-standard terminal moving in typical indoor environments), with measurements using RSSI and Round Trip Delay (RTD). A comparison between non-cooperative and cooperative positioning was done and several positioning algorithms were tested in both cases. The 90th percentile of position error is 4 m. In reference [63], algorithms with realistic heterogeneous wireless networks, including GSM, DVB, FM and WLAN, were evaluated with measurements of RSSI. That paper proposed two positioning algorithms: Direct Multi-Radio Fusion (DMRF) reorganized the information in a transformed space and Cooperative Eigen-Radio Positioning (CERP) used the spatial discrimination property. The mean position error of this approach is 1.5 m. Based on these works, the accuracy of these indoor positioning systems could be further improved to accomplish the requirements of current LBS applications (e.g., position error below 1m for pedestrian indoor navigation). Indoor positioning strategies 51 4.3 Problem description RF-based techniques are widely used for indoor positioning, but they usually do not provide enough accuracy for indoor navigation. In such techniques, metrics as RSSI, Angle of Arrival (AoA) or Time of Flight (ToF) may not be directly related to distance or relative position between transmitter and receiver. This is due to multipath and fading phenomena that can be dominating in indoor environments. Therefore, position estimation suffers from this initial uncertainty and sophisticated positioning algorithms and filtering are needed to compensate these drawbacks. Three main types of environments could be found for indoor positioning in a building: halls, corridors and offices. For offices and rooms, simple detection of presence in the room could be enough for many applications. However, accurate indoor positioning systems are specially required in corridors and halls with multiple doors. Such areas (e.g., hospital corridors surrounded by patient rooms) act as distributors to different locations making essential a precise position estimation, e.g. in order to indicate a specific door. Hence, the proposed indoor positioning system would be focused on a corridor environment which is a more critical scenario than halls in terms of multi-path propagation. Fingerprinting techniques are good candidates to overcome multi-path propagation uncertainty. Fingerprinting techniques compare several features of the fingerprinting pattern (large data) with current information. That identifies the original data with the analyzed information [70]. Focusing on RFID systems, they could be active or passive (in terms of tags activity). Passive systems could work in a reliable way for reading distances up to 3 m in indoor environments or 15 m in free space conditions [71]. Furthermore, passive tags embed a shunt resistance in order not to damage circuitry when the tag is near to the reader, avoiding the loading effect. Therefore, when tag-to-reader distance is less than 10 𝜆, the system does not ensure the received power is linked to the distance by any propagation model [72] [73]. In consequence, passive systems are not convenient for indoor challenge scenarios like corridors where tags and readers would be very close to each other. Conversely, active technology copes with these limitations regarding the tag-to-reader distance and reading distances are up to 15 m in indoor scenarios. Even so, passive tags cost ten times less than the active ones but the price of the readers is also very important. The use of multi-antenna readers would improve the system performance. It could be done by increasing the number of antennas of a reader or placing two (or more) readers together. The later would require a higher investment with readers for passive tags compared to the readers for active tags. Taking it into account, each device (e.g., smarthphone) would need a reader, thus, the price of the reader would be more important than the price of the tags. 52 Indoor positioning strategies In this respect, RFID systems with active tags: are not expensive compared to passive tags, have an operating range of several tens of meters and are easy to deploy, which comply with the application requirements for pedestrian navigation. However, the accuracy of this RFID-based indoor positioning system should be improved to reduce the position error (few meters). In addition, none of the previous work analyzed the promising approach of combining RFID and macrocell signal processing as a low cost solution. In fact, up to the author knowledge, there has not been any deep study in the field of combining RFID with macrocellular signals. Hence, this is a suitable option for enhancing precision for pedestrian navigation in buildings with a highly reduced implementation costs in comparison with previous solutions. Further details about RFID and cellular technologies are summarized in Appendix A. It describes their general characteristics and theoretical models in indoor scenarios. Additionally, Appendix B presents a signal study of these technologies based on their theoretical models and the measurement campaign. It would assess their applicability for indoor positioning systems. 4.4 Methods for indoor positioning To determine the possible improvement achievable by the use of different techniques and multi-antenna RFID reader or the combination of RFID and cellular technologies, different baseline single-technology solutions are defined. Firstly, the description of the selected scheme for the implementation of the proposed indoor positioning system is detailed. Then, several techniques and methods for data fusion are analyzed for the RFID-based indoor positioning system. Finally, the integration of the cellular technology into RFID-based positioning system is presented. 4.4.1 Fingerprinting-based scheme Fingerprinting is selected to provide indoor positioning, being a widely applied scheme for positioning in these environments [74]. It is defined by two main phases as Figure 4.1 depicts; calibration phase (offline) and localization phase (online). These stages are explained focused on the proposed approach: • Calibration phase: This stage is performed prior to the provision of the positioning service. RSSI (𝜇𝛼,𝜏 𝑃𝑃𝑃� 𝛾𝑓�) is collected by the antenna(s) of the reader (where 𝛼 represents the particular antenna of the set of antennas 𝐴 of the reader, 𝛼 ∈𝐴) from different transmitters (where τ represents the particular Indoor positioning strategies 53 transmitter of the set of transmitters 𝑇, τ ∈𝑇). That information is gathered from known positions (also called fingerprinting positions) 𝛾𝑓=�𝑥𝑓,𝑦𝑓,𝑧𝑓� of the scenario and it is referred as fingerprinting measurements or calibration data. • Localization phase: It is in charge of the online positioning service. The set of measured RSSI (𝜇𝛼,𝜏 𝑃𝑃𝑃( 𝛾𝑏𝑐𝑃𝑃)) collected in real time by the device to be localized, are used to estimate its current position γcurr= (𝑥,𝑦,𝑧) by means of comparing it to the stored fingerprinting data. Different techniques can be implemented. 4.4.2 RFID-based positioning system An RFID-based system with active tags is proposed. Different positioning techniques and methods for data fusion are analyzed for multi-antenna readers. This system collects information (𝜇𝛼,𝜏 𝑃𝑃𝑃) from each fingerprinting position in the calibration phase. Then, the localization phase processes the current measurements. This phase is divided in three main steps as Figure 4.2 shows: • Step 1 - Candidates ranking: Each of the fingerprinting positions is assigned with a certain weight 𝑤�𝛾𝑓,𝛾𝑏𝑐𝑃𝑃� based on different possible criteria (techniques): the similarity of the 𝜇𝛼,𝜏 𝑃𝑃𝑃 values received from each transmitter in respect to those stored in the fingerprinting database, the posterior probability of being in the fingerprinting positions given the current 𝜇𝛼,𝜏 𝑃𝑃𝑃 values, etc. • Step 2 - Candidates filtering: Based on the ranking generated in the previous step (weighted fingerprinting positions), one or several of them, candidates, are selected as possible estimated positions. The rest are discarded. • Step 3 - Position estimation: From the set of the selected candidate positions, the current position of the device is estimated. It could be observed in Figure 4.2 that the localization phase could be executed in parallel for each set of collected data from each antenna. This means, several singleantenna RFID-based positioning systems with multiple position solutions (one per system). Conversely, the system information from each antenna could be merged into a single flow at the beginning of each step (any of the 3 steps) or at the end of step 3, as the flowchart illustrates through the green dashed blocks (fusion methods). Four fusion methods are proposed to integrate the diversity of multi-antenna in the general positioning scheme. Note that, only one fusion methods could be triggered. An analysis to assess the optimal step to data fusion would be performed. Additionally, Figure 4.2 illustrates the different techniques developed at each step. 54 Indoor positioning strategies Figure 4.1: Fingerprinting-based scheme. Indoor positioning strategies 55 4.4.2.1 Step 1: Candidates ranking This step processes the information supplied to the indoor positioning system (current measurements and fingerprinting measurements) to get ranked candidates (ranked fingerprinting positions). Firstly, the input data is gathered. Afterwards, a technique is selected for candidates ranking. The outputs of this step are the ranked fingerprinting positions, 𝑤𝑃𝑃𝑃�𝛾𝑓, 𝛾𝑏𝑐𝑃𝑃�. The input data and the ranking techniques are detailed below: 1) Inputs Inputs could be either information from each antenna that would be processed in parallel and individually, or, information from all the antennas that would be processed together in a single positioning system: • Individually processed measurements Each set of information collected per antenna (𝜇𝛼,𝜏 𝑃𝑃𝑃) would be the input data of its indoor positioning system. There will be as many positioning systems as number of antennas in the receiver. • Fusion method 1: Measurements fusion The aim of this method is to collect all the fingerprinting and current measurements gathered from each antenna to be lately processed in a single indoor positioning system. This is depicted on the vertical flow of fusion method 1 and top horizontal scheme in Figure 4.2. Now, instead of processing the dataset from each antenna 𝛼 in an individual indoor positioning system, all the data is processed by one of them (e.g., indoor positioning system - 1). 2) Candidates ranking techniques Different techniques proposed in the literature can be applied to calculate the weights and perform the ranking of all the samples, particularly Number of RSSI matches, Bayes Classifier or Euclidean Distance. Focusing on the multi-antenna case (all the information collected from each single-antenna receiver is processed by a single indoor positioning system), these techniques are described below. • Technique 1: Number of RSSI matches The indoor positioning system performs the ranked candidates by means of calculating the number of matches between the current measurements and the fingerprinting ones: 62 Indoor positioning strategies • Fusion method 4: Estimated positions fusion Fusion method 4 is performed after the step 3 by calculating the geometric center or median (as step 3) of the set of the different estimated positions proposed by each indoor positioning system: 𝛾�𝑏𝑐𝑃𝑃 𝐴= {𝛾�𝑏𝑐𝑃𝑃 𝛼1,𝛾�𝑏𝑐𝑃𝑃 𝛼2,𝛾�𝑏𝑐𝑃𝑃 𝛼3, … }. 4.4.3 Cellular technology into RFID-based positioning system A new scheme for enhancing the proposed RFID-based indoor positioning system with cellular technologies is proposed in Figure 4.4. The aim of this approach is to analyze and study the advantages of having both cellular and RFID signals at low cost. The proposed cellular-based positioning system follows the same scheme and steps as the one proposed for the RFID-based positioning system (see Figure 4.1 and Figure 4.2). According to the study carried out in Appendix B, the cell identifier (𝐼𝐼) has shown the most promising qualities for the proposed system instead of the PRX from each cellular antenna, so it would be the key input for this system. In this study, a single-antenna indoor positioning system is analyzed. The step 1 of the cellular-based indoor positioning system performs the ranked candidates by means of calculating the number of matches between the current measurements and the fingerprinting ones: 𝑚𝑚𝑡𝑚ℎ𝐼𝐼𝛼�𝛾𝑖,𝛾𝑚𝑐𝛥𝛥�=�1𝑖𝑖 𝐼𝐼𝛼( 𝛾𝑏𝑐𝑃𝑃)∩𝐼𝐼𝛼�𝛾𝑓�≠0 0𝑜𝑡ℎ𝑒𝛥𝑤𝑖𝑒𝑒 , (4.11) where 𝐼𝐼𝛼�𝛾𝑓� is the set of received identifiers by antenna 𝛼 at the fingerprinting position, 𝛾𝑓, and 𝐼𝐼𝛼( 𝛾𝑏𝑐𝑃𝑃) is the set of received identifiers at the current position, 𝛾𝑏𝑐𝑃𝑃, during the acquisition time. The expressions mean that in case both sets shared common identifiers, the match is considered affirmative. Afterwards, the weight assigned to each fingerprinting position is the same as the matches per fingerprinting position 𝛾𝑓: 𝑤𝐼𝐼�𝛾𝑓, 𝛾𝑏𝑐𝑃𝑃�=�𝑚𝑚𝑡𝑚ℎ𝐼𝐼𝛼�𝛾𝑓,𝛾𝑏𝑐𝑃𝑃� ∀α∈𝑨 . (4.12) Indoor positioning strategies 63 Figure 4.4: Integration of cellular technology into RFID-based positioning system. 64 Indoor positioning strategies The ranked positions are supplied to step 2 which obtains the most probable candidate positions. Subsequently, these candidate positions from the cellular-based positioning system together with the candidate positions from the RFID-based positioning system are processed by the Integration system. Even if other approaches are possible, this work integrates results from both RFID and cellular systems. These solutions allow a clear comparison of the gain of the integrated system as well as facilitate its integration in already existing services. With this objective, the proposed Integration system defines two main steps. 1) Cellular discrimination The RFID filtered candidates per position 𝛤𝑓,𝑅𝑅𝐼𝐼 𝑓𝑓𝑏𝑎 are discriminated by the cellular filtered candidates 𝛤𝑓,𝐶𝐶𝐶𝐶𝐶𝐶𝐴𝑅 𝑓𝑓𝑏𝑎 as follows: an area of one squared meter (or maximum distance between two adjacent fingerprinting positions in the cellular calibration phase) is created for each cellular candidate position in order to evaluate if any of the RFID candidates are located inside this area. In this case, that RFID candidate position is selected, whereas any other candidate outside those areas is discarded. 2) Opportunistic localization Once the new set of candidates has been selected, the estimated position is calculated in the step 3 by centroid or median metrics. The cellular-based indoor positioning system is inaccurate itself. Nevertheless, the cellular candidates are used to discriminate aberrant candidates proposed by the RFID-based positioning system rather than adding new possible candidates. In this sense, this discrimination could enhance the solution provided by the RFID system through a reduction of some aberrant of its candidates. 4.5 Evaluation This section describes the configuration of the scenario, the equipment used and the measurement campaign. Then, the assessment of the proposed indoor positioning systems: RFID-based positioning system, cellular-based positioning system and their integration into a single system, is performed. Indoor positioning strategies 65 4.5.1 Trial set-up The corridor scenario, the equipment and the measurement campaign are described in the next subsections. 4.5.1.1 Scenario The selected scenario is a corridor located on the 4th floor of an indoor office building (526 m2). It is 22.5 m long x 2 m width x 2.5 m height. Standard objects or furniture were not removed from the corridor to assess the proposed systems under real environments conditions and topology. Further visual details about this scenario from both sides of the corridor are depicted in Figure 4.5. Figure 4.5: Scenario – Corridor. 4.5.1.2 Equipment The global indoor positioning system scheme is shown in Figure 4.6. Here, a cloudcomputing scheme is assumed, where most part of the positioning algorithm computation is performed externally by a remote entity. This is a widely extended approach in positioning systems in order to reduce the computational and storage costs for the device to be located (e.g., smartphone), at the cost of increasing the requirement of maintaining a continuous communication with the external entity. In this scheme, the information transmitted to the external entity can consist just in the RSSI, transmitter/receiver identifiers and timestamp. 66 Indoor positioning strategies 4.5.1.2.1 RFID equipment The UHF RFID technology system works at 433MHz (ISO 18000-7). It is composed by a two-antenna reader and 30 active tags, both from Ela-Innovation [77]. The reader model is “UTPDiff2” with 2 vertical dipole Rx/Tx antennas (see Figure 4.7(b)). Active tags are the “Thinline” model (see Figure 4.7(a)), which could be detected from as far as 20 m in an indoor environment. Figure 4.6: RFID-Cellular indoor positioning system scheme. Measurements collected by the RFID reader are quantified samples. The range goes from level 118 to level 215 in steps of 1 level. This is equivalent to -44 dBm for the lowest level and -106 dBm to the highest level with a resolution of 0.6 dB. a) Active tag b) Two-antenna reader Figure 4.7: RFID equipment. Indoor positioning strategies 67 Active tags were located on the walls of the corridor following a pattern. Some of them are placed at 1.4 m height which corresponds to a doorknob while some others are placed at 2.1 m height, corresponding to a standard door height. The tags are alternatively placed from one height to the other with a distance of 1.5 m between two tags. On both sides of the corridor, a tag is placed at 1.4 m. Figure 4.8 illustrates a scheme of the position of the tags and the layout of the deployment. Figure 4.8: RFID active tags position. Notice that the ceiling is free of tags due to the specific radiation pattern of the two-antennas of the reader. The axis of the dipole has a zero, which means, zero reception on the ceiling-floor line. 4.5.1.2.2 Mobile communications equipment For cellular network assessment two smartphones were used in the experimental evaluation: Samsung Galaxy S3 and Sony Ericsson Xperia X10 Mini E10i (as it will be later described, the former was used to measure UMTS technology and the later to measure GSM technology). They are widely extended commercial models running Android 4.2 and Android 2.1 respectively. Additionally, to measure and record the received power in those devices, a popular free Android application (app) for cellular monitoring, G-MoN [78], was used. This app is a powerful tool for monitoring cellular and other wireless technologies as a drive test tool. It provides cellular network information such as PRX, cell identifier, Local Area Code (LAC), etc. Note that, PRX values are averaged at Layer 1 and 3 [79] of the terminal protocol stack minimizing the impact of fast-fading in its values. Therefore, the app reported PRX reflects the path loss and shadowing characteristics of the signal. (0,0,0) z x y 1.4m 2.1m 1.5m 2 meters 22.5 meters 68 Indoor positioning strategies 4.5.1.3 Mobile platform The RFID reader and smartphones were placed on top of a trolley as Figure 4.9 depicts. This trolley was the item to be located, referred as the mobile platform. It was moved along the corridor while receivers were reading and collecting information. Both antennas of the RFID-based system work at 433 MHz and collect samples from each static active tag. Regarding the cellular technology, Samsung smartphone was forced to be camped on UMTS network whose frequency band was 2100 MHz. Likewise, the other smartphone was pushed to be connected to the GSM network at 900 MHz frequency band. LTE networks were only deployed in main cities on the moment of this study. Thus, the analysis of this technology would be a future work. Figure 4.9: Mobile platform. This technology diversity (433, 900 and 2100 MHz) allows a complete study in indoor conditions in order to enhance positioning systems (see Appendix B). 4.5.1.4 Measurement campaign On the one hand, with the presented RFID equipment, the main parameters that can be measured by the reader in a time slot are: • Tag ID: Numerical identifier of the tags. • RSSI: Quantified received power from a tag. The range is from -44 dBm (level: 118) to -106 dBm (level: 215) with a resolution of 0.6 dB (1 level). Indoor positioning strategies 69 On the other hand, for cellular signals, the main parameters that can be measured by common terminal applications are: • Cell ID: Numerical identifier of the serving cell and neighboring cells. • PRX: Received power from a macrocell. The resolution is 1 dB in the range [-115, -30] dBm. According to this, a signal assessment of RFID and cellular (GSM and UMTS) systems is performed in Appendix B in order to analyze the applicability of these technologies for indoor positioning systems. It is based on the theoretical models of each technology (see Appendix A) and a measurement campaign. Next subsections detail the measurement campaign carried out to collect samples and build the fingerprinting databases, as well as the localization phase of this study. 4.5.1.4.1 RFID-based positioning system This system is composed by two phases of measurements. The sampling campaign of each phase is explained below. 1) Calibration phase This phase collects and stores information about RSSI and Tag ID (𝜇𝛼,𝜏 𝑃𝑃𝑃) to build the RFID fingerprint model. Firstly, the corridor is divided in a mesh of 75x5 positions (375 positions: 75 positions per line and 5 lines) being the distance of adjacent positions of the same line equal to 30 cms (blue dots in Figure 4.10). Then, the trolley (i.e., two-antenna reader) is placed at each position, sampling and quantifying RSSI measurements. The equipment recorded samples during five minutes per position, which means, the fingerprinting database stores 675000 samples (375 positions, 60 samples per tag/position and 30 tags). These measurements were gathered statically as the period of characterization was large and it is the proper approach for the characterization of each point at calibration phase. 2) Localization phase Once the fingerprinting database is fully built, the system is able to automatically report the position of the trolley. For this study, the two-antenna reader monitors real time measurements and provides the estimated position of the trolley. Twenty one positions were dynamically estimated along the middle row of the corridor (red dotted line in Figure 4.10), emulating a commercial application. In this context, the 70 Indoor positioning strategies acquisition time at each point is estimated in 5 s while the system takes a mean time less than 0.1 s to calculate the estimated position. Figure 4.10: Positions. 4.5.1.4.2 Cellular-based positioning system This system is composed by two phases of measurements as well. The sampling campaign of each phase is explained below. 1) Calibration phase The measurements are recorded along the corridor every meter following 3 lines on the floor with a distance of 50cm from each other (63 positions) as Figure 4.10 shows (green dotted circles). In this case, the selected parameter is the Cell ID, i.e., the macrocell identifier (𝐼𝐼𝛼). Therefore, Cell ID information was recorded during five minutes at each point, which implies that around 37800 samples in the whole scenario from the cellular networks (18900 samples per technology GSM or UMTS) were stored. 2) Localization phase The localization phase was carried out at the same time as RFID localization phase, therefore, it followed the same period as RFID system to collect measurements and similar time to calculate the estimated position (less than 0.1s). 4.5.2 Trial results The proposed indoor positioning systems (RFID-based and cellular-based), the use of multi-antennas for RFID readers, the influence of the number of active tags and the 22.5 meters 2 meters 1.5 1.2 0.9 0.6 0.3 0.0 0.0 0.3 0.6 0.9 1.2 1.5 . . . . 20.1 20.4 20.7 21.0 RFID Calibration Phase Cellular Calibration Phase Localization Phase meters meters Indoor positioning strategies 71 achieved synergies due to the integration of cellular technologies in RFID-based positioning systems, are analyzed in the field of heterogeneous localization. 4.5.2.1 RFID-based positioning system In order to assess the RFID-based indoor positioning system, the techniques and fusion methods described in subsection 4.4.2 “RFID-based positioning system” are evaluated. Firstly, the study is focused on single-antenna system. Afterwards, the same analysis is performed with two-antenna and different fusion methods. Finally, a tradeoff between the number of antennas and the number of tags is detailed. 4.5.2.1.1 Single-antenna The evaluation of this system is carried out in the middle line of the corridor by moving the mobile platform along it (see Figure 4.10). Only one antenna (A1) is used for this study. The error between the real position and the computed positions are calculated for each technique at each step in the localization phase: three techniques in step 1 (with two different tolerance margins for technique 1 – ±0.3 dB and ±0.9 dB), one technique in step 2 (with two different k-intersected candidates – k=1 referred as mic and k=5 referred as kic) and two techniques in step 3. As only information from one antenna is processed, no data fusion (fusion methods) is performed. Figure 4.11 shows the CDF of the position error for the presented techniques. Figure 4.11(a) selects the candidates by the “Number of RSSI matches” technique (technique 1). The tolerance margin is set to ±0.3 dB, i.e., equal values in the quantified RSSI levels of the fingerprinting and the measured positions. The four combinations present similar characteristics, 95th percentile is around 4-5 m. Figure 4.11(b) also illustrates the results of the “Number of RSSI matches” technique. Nevertheless, the tolerance margin is set to ±0.9 dB. In this situation, the 95th percentile of position error is reduced to 3 m when the computed positions are calculated based on kic (blue and green lines). In case the mic approach is selected, i.e., the maximum intersected candidates are selected (k=1), the position error is increased. As observed, there is no significant difference between the centroid and the median metrics. Similar behavior is observed in Figure 4.11(c) for the “Bayes Classifier” technique (technique 2). In case k=1, the 95th percentile of position error is around 3 m and 78 Indoor positioning strategies the RFID-based positioning system. This enhancement provides mean position error values from 1.1 to 1.8 m. In the same context, the 95th percentile is reduced, reaching values from 2.2 m to 3.7 m. The “Bayes Classifier” and “Euclidean Distance” techniques are slightly improved when more than one candidate is selected (kic, k=5). In other case (k=1, mic), there is only one candidate and this opportunistic approach is not practical. For the best technique (Euclidean Distance, kic), results show 1.4 m for the 80th percentile and 1.8 m for the 95th percentile. Table 4.2: Summary of indoor positioning systems performance. Technique 1 (∆𝛥𝛥𝑥= 0.3) Technique 1 (∆𝛥𝛥𝑥= 0.9) Technique 2 Technique 3 mic kic mic kic mic kic mic kic RFID System Mean absolute error (m) 1.9 1.8 2.1 1.3 1.6 1.3 1.1 0.9 RFID System MSE 6.7 5.9 7.4 4.5 5.6 3.9 3.6 3.1 RFID System 95th percentile 5.5 4.1 5.3 3.5 3.4 2.6 2.0 1.9 Oportunistic System Mean absolute error (m) 1.5 1.4 1.8 1.1 1.6 1.2 1.1 0.9 Oportunistic System MSE 5.5 4.0 5.8 3.7 5.6 3.7 3.6 3.0 Oportunistic System 95th percentile 3.2 3 3.7 2.2 3.4 2.5 2.0 1.8 Compared to other positioning systems with more than one technology in the stateof-the-art, the work presented in [69] had a position error of 4 m at 90th percentile. Other works like in [63] had a mean position error of 1.5 m. Showing these results, the proposed system outperforms both of them, having a mean position error of 0.9 m and 1.8 m for the 95th percentile. Focusing on corridors, this approach outperforms the system proposed in [51] where the mean position error was 5.9 m compared to 0.9 m. 4.5.2.3.1 Sensitivity study of number of tags A sensitivity study has been performed to evaluate this approach in the same scenario with a low-dense number of tags. The number of tags has been reduced according to 1/2, 1/3 and 1/5 from the original amount (30 tags). Table 4.3 and Table 4.4 present the MSE of the RFID system and the proposed system respectively. It could be observed in both tables how the less number of tags are placed, the worse Indoor positioning strategies 79 values are obtained. Furthermore, with high density of tags, cellular technology improves the location accuracy over the RFID system. However, when the number of tags is extremely reduced, the influence of cellular technology does not enhance the location error as much as expected. The reason is related with the low number of candidates the RFID system proposes. Therefore, the discrimination of cellular technology is much more limited to those candidates. Table 4.3: Sensitivity study of RFID system based on the number of tags (MSE). RFID System MSE Technique 1 (∆𝛥𝛥𝑥= 0.3) Technique 1 (∆𝛥𝛥𝑥= 0.9) Technique 2 Technique 3 mic kic mic kic mic kic mic kic 30 tags (ratio 1) 6.7 5.9 7.4 4.5 5.6 3.9 3.6 3.1 15 tags (ratio 1/2) 15.6 9.9 9.1 6.5 9.2 6.0 5.6 4.8 10 tags (ratio 1/3) 24.4 20.8 14.3 11.0 13.5 10.9 9.8 8.7 6 tags (ratio 1/5) 43.8 38.9 31.7 21.2 26.7 19.6 16.9 14.6 Table 4.4: Sensitivity study based on the number of tags (MSE). Integrated System MSE Technique 1 ( ∆𝛥𝛥𝑥= 0.3 ) Technique 1 ( ∆𝛥𝛥𝑥= 0.9 ) Technique 2 Technique 3 mic kic mic kic mic kic mic kic 30 tags (ratio 1) 5.5 4.0 5.8 3.7 5.6 3.7 3.6 3.0 15 tags (ratio 1/2) 15.0 9.1 9.1 5.9 9.2 5.8 5.6 5.1 10 tags (ratio 1/3) 23.1 11.8 12.2 8.1 13.5 10.7 9.8 9.6 6 tags (ratio 1/5) 40.0 38.8 27.9 19.5 26.7 18.5 16.9 15.8 4.5.3 Considerations for real deployments By the time of this work, cost of each active tag is in the range of tens of USDs (United States dollars) while a RFID reader rounds few hundred USDs costs. Therefore, the defined RFID system is based on the distribution of RFID tags in the infrastructure, while the localized platform (e.g., pedestrian equipment) to be positioned is provided with a RFID tag reader, which is the optimal solution if a reasonable amount of mobile devices are to be localized in a wide and complex area. In this way, the infrastructure costs are minimized, while placing the cost in the positioned object. This greatly reduces installation costs and allows the implementation of the system in large areas at minimum expense. Additionally, the commercial penetration of portable RFID readers is growing, being also included in mobile phones as peripherals or embedded systems [80]. In any case, scenarios where 80 Indoor positioning strategies the located system is equipped with RFID active tags instead of a transceiver (being the transceivers part of the fixed infrastructure of the scenario) are also possible [81] [82], keeping the same general scheme and conditions in terms of performance and architectural needs. In order to cover an entire floor of a public building of some hundreds square meters with passive tags, a few hundreds readers must be deployed, as their read range is 1-3 m. Using active tags allows to cover a whole floor, without increasing the total number of tags. Further studies on the optimal number of tags and their placement are being carried out for the corridor [83] and the entire floor. Therefore, active RFID solution still cost effective when covering large areas even if we locate a hundred of mobile readers. In the proposed approach, the main elements of classical architectures for RFIDbased positioning are maintained: a remote localization server performs the calculation of the position based on the signals received from the RFID tags and the previously stored fingerprinting data. The only additions to this classic structure imply the inclusion of cellular signal fingerprint information as part of the localization server databases and the existence of a cellular receiver as part of the localized platform. The cellular receiver may be already part of the localized platform (e.g., for communications reasons). Otherwise, adding the cellular receiver to the previously existing pure RFID platforms can be done at a very low cost due to the wide popularity of the cellular technology: a low-budget smartphone may be enough. Additionally, the increasing availability of RFID commercial portable readers for active RFID tags could make this approach even more accessible for pedestrian applications in the close future. The acquisition of cellular technologies fingerprinting data should also not suppose any significant cost in terms of the calibration phase, as the cellular scenario characterization can be performed simultaneously to the RFID one. A shortcoming of the proposed system is that if the cellular infrastructure changes due to modifications done by operators (e.g., change of configuration parameters) or failure, the accuracy of the cellular based discrimination can be jeopardized and a new calibration phase may be required. This is the same case as for changes in the RFID infrastructure, but the latter are usually more accessible and under the control of local administrators. In order to overcome this challenge, coordination between the mobile operator and the positioning system should be defined, fitting in the initiated process by cellular standardization aiming to integrate localization (including the one coming from third party solutions) as part of the standard architecture interfaces [84]. Indoor positioning strategies 81 Another issue arises when the indoor area is covered by only one (or few) base stations, which makes the cellular information useless or not be as helpful as expected. However, the experience has shown that most time a certain level of signal is received from several cells simultaneously (it does not mean the quality of the signal is appropriate to attempt a call) or cellular technologies. Furthermore, the increasing deployments of small cells will help to further avoid this restriction. To conclude, the use of common application layer apps for the mobile terminal is assumed as the source of signal information. However, one of the main limitations of some terminals (due to their manufacturers) is their inability to report information about neighbor cells (e.g., PRX, Cell ID, etc.). That issue would affect the fingerprinting procedure as it could suffer from lack of information. However, handover/cell-reselection process would be useful in order to overcome this lack of information from several cells. Regarding this issue, this study has taken into account that restriction, focusing on this kind of terminals (e.g., Samsung) as a high percentage of smartphones has this limitation. 4.6 Conclusions An active UHF RFID-based positioning solution in a corridor has been analyzed. Several techniques and data fusion methods for multi-antenna reader and the fundamentals for innovative and opportunistic use of GSM and UMTS technologies with these systems have been presented. The approach is evaluated in a real corridor field-trial, comparing the position error of single-antenna and two-antenna readers, for different positioning techniques (number of RSSI matches, Bayesian Classifier and Euclidean Distance) and applying the proposed method for multi-antenna data fusion (measurements, ranked candidates, filtered candidates and estimated positions). The best results with the multi-antenna approach are obtained with a Bayes Classifier and measurements fusion. In this case, the achieved mean position error is 0.75 m and the 90th percentile is 1.1 m. The proposed solution achieves better accuracy than previous RFID solutions experimented in corridors, even with other technologies and indoors scenarios. Additionally, a trade-off between the number of antennas and the number of tags has been presented. This study shows that the two-antenna approach can provide equivalent accuracy as single-antenna approach but half number of tags in the infrastructure. 82 Indoor positioning strategies Finally, the analysis assesses the possibilities of using signals coming from already existent non-positioning oriented cellular networks to provide support to RFID-based indoor positioning mechanisms, which could highly benefit of such pre-existent communications infrastructure. Furthermore, the use of a first-approach, fingerprinting mechanism for positioning shows promising results in terms of accuracy enhancement. 83 Chapter 5 5 Indoor mobility load balancing techniques This chapter is focused on the development of novel MLB mechanisms to mitigate temporary traffic fluctuations and focused network congestion issues in open access LTE femtocell networks in commercial and corporate environments. The structure of this chapter is as follows: Section 5.1 introduces MLB techniques in indoor scenarios. Section 5.2 details the related work. Section 5.3 formulates the problem description. Section 5.4 describes some MLB mechanisms for corporate femtocell networks. In addition, Section 5.5 presents the design of innovative MLB mechanisms based on context information. Section 5.6 and Section 5.7 evaluates the proposed methods in a simulator and field trials, respectively. Finally, Section 5.8 summarizes the main conclusions of this chapter. 5.1 Introduction Most cellular communications take place at indoor environments (at work, home, shopping malls, etc.) [1] [10], especially in commercial and corporate scenarios. Hence, operators tend to deploy small cells, such as femtocells, indoors to enhance network capacity, to reduce outage coverage areas, to provide high-speed data traffic, etc. In this indoor scenarios, voice calls and data traffic, as well as local user densities, vary in temporal and spatial domain. Those situations lead to degraded indoor cellular networks because many people want to use their mobile devices at the same time, close to the same area and/or for a short period. For example, people waiting for a delayed flight at the boarding gate of the airport could collapse a femtocell for a while or, a 84 Indoor mobility load balancing techniques celebrity walking through a mall where everybody is interested in taking pictures and sharing them instantaneously in social networks or calling friends to share the experience. A simple solution to support these extreme situations could be to plan the network resources according to the peak traffic. Nevertheless, this solution would increase operators’ expenditure. The self-optimization use case of Mobility Load Balancing (MLB) has been proposed by 3GPP [45] to solve these situations. MLB use case aims to shift users from overloaded cells to other cells with spare resources. These MLB techniques tune network configuration parameters to reach a better configuration that alleviates the congestion situation. Focusing on indoor and femtocell environments, the following parameters are mainly modified: 1) femtocells handover margins to resize service areas and/or 2) femtocells transmission powers to resize cell coverage areas. Compared to traditional macrocells, femtocells are simple and efficient access points which present hardware restriction in the number of simultaneous active users. This means, the number of active users is usually limited from 2 to 64 by power of two (depending on the femtocell model [31]). As usual, the number of users that can be properly served also depends on the availability of radio resources of the cell. However, the limitation in the maximum number of connected users is commonly much more critical than the availability of resources, since the bandwidth normally supports higher number of simultaneous users than the femtocell is able to manage. SON mechanisms are commonly based on network alarms, counters and KPIs, or estimated information from radio propagation models. However, nowadays, smartdevices and its embedded sensors (e.g., accelerometers, gyroscopes, etc.) and applications, measure and provide additional context information such as users’ position, WiFi Service Set IDentifier (SSID), temperature, etc., that could be used by SON mechanisms to both accelerate their convergence and improve their performance. Focusing on users’ position, that information could be easily obtained at indoor environments thanks to the immediate interest on indoor LBS. That means, new key location-aware applications for emergency response, advertising, healthcare, domotics, etc. are showing up. Based on this, indoor positioning mechanisms have to evolve in order to be accurate and robust systems, being a hot topic in both academia and industry (e.g., Google Maps - Indoor). Additionally, classical MLB methods might not properly work in indoor femtocell networks due to the special characteristics of femtocells (restriction in the number of users per femtocell, unplanned deployments, short-range, etc.), indoor environments characteristics (multi-path reflections, occasional events, etc.) or users’ indoor mobility pattern (increase number of handovers, etc.). Hence, the development of new MLB Indoor mobility load balancing techniques 85 mechanisms based on context information from external mobile network sources might improve the performance in commercial and corporate environments. The main contributions of this chapter are: • The design of innovative MLB mechanisms in commercial and corporate LTE femtocell networks with open access. • The integration of context-aware information into novel MLB methods. • The reduction of temporal and focused overloaded situations at commercial and corporate indoor scenarios due to the high concentration of users in temporal and spatial domain. • The estimation of the impact of indoor positioning system accuracy on the designed location-aware MLB mechanisms. Finally, the proposed SON methods are discussed and compared in both, simulator and field trial scenarios to evaluate their capabilities. 5.2 Related work Among SON techniques and focusing on self-optimization mechanisms, several works have been centered on the macrocell case in both, literature and European projects [7] [8] [9]. By contrast, indoor environments present hard and difficult conditions to manage the cellular network due to the cell overlapping, lack of coverage, interference, etc. This implies a challenge for researchers and engineers in the development of SON mechanisms. In this sense, self-x techniques in femtocell networks are currently a hot topic [12] [13] [14] [85] [86] [87] [88] [89] [90]. MLB mechanisms have been also widely studied by both, academia and industry, at outdoor and indoor scenarios [4] [5] [6] [15] [16] [17] [91] [92] [93] [94] [95]. The most suitable solution to reduce or avoid cellular congestion situations is by resizing the cell coverage areas. It could be addressed from two ways: tuning physical parameters in the cell (e.g., antenna tilt [96] or pilot transmission power [94]) or changing parameters in Radio Resource Management (RRM) processes (e.g., handover, HO, [92] or cell reselection, CR, [97] parameters). The best configuration parameters could be found by formulating classical optimization problems [18]. However, information to build the analytic models is rarely available, thus, the operators use heuristic methods. Concerning MLB mechanisms in corporate femtocell environments, simple and lowcomplexity methods based on FLCs are proposed in [15], which investigated the 86 Indoor mobility load balancing techniques problem of re-distributing traffic demand between LTE femtocells. A similar fuzzy logic approach is presented in [16], where cell transmission power and handover margins are tuned to solve congestion problems on traffic distribution in enterprise LTE environments. A more computationally complex method [17] obtained better performance than previous studies through a fuzzy rule-based reinforcement learning system, while reference [95] proposed a distributed method to achieve automatic load balancing based on a flowing water method. However, none of the previous works analyzed the femtocell limitations in the maximum number of simultaneous connected users (macrocells or other type of small cells do not have such a restricted condition), which could degrade the network performance. In addition, those previous works were designed to solve localized and persistent congestion problems since a long time was usually necessary to obtain the optimized parameters, disregarding the challenge related to temporary congestion issues at indoor environments. The femtocell limitation in the maximum number of simultaneous users was taken into account in [88], where a handover algorithm based on the users’ speed and QoS was proposed. That work evaluated whether a handover was necessary or not according to the previous metrics (speed and QoS), but the algorithm was not analyzed under overload situations. Some other mechanisms introduced the use of context information in selfoptimization methods. More specifically, in location-awareness, the authors in [18] introduced users’ position into the MLB mechanisms to reduce handovers and call blocking rates in overloaded cells by modifying the coverage area. However, it was focused on outdoor UMTS macrocell networks. In [14], a method that adjusted hysteresis margins depending on an estimation of the distance from the cell to the terminal, reduces the number of redundant handovers while keeping the throughput of femtocells as high as possible. But, it did not take into account the number of active users in the femtocells. The work in [94] presented a dynamic power control for balancing data traffic in femtocell networks. It built Voronoi diagrams based on COST231 multi-wall radio propagation model to estimate the data traffic per femtocell based on the users’ position. However, the integration of this mechanism into a real environment was limited to the values calculated with the propagation model. Furthermore, it only analyzed the network load in terms of occupied radio resources (traffic volume). Other studies utilized users’ position in their SON mechanisms [98] [99] [100]. Nevertheless, those works are out of the scope of MLB use case. The MLB techniques proposed in this chapter are based on resizing cell areas by modifying cell transmission power. In this context, reference [100] introduced the fingerprinting technique as a method for cellular optimization. It compared the PRX measurements with a propagation model to predict the PRX. However, that work was oriented to outdoor environments. Conversely, the study in [94], as previously Indoor mobility load balancing techniques 87 mentioned, was focused on femtocell networks and presented a balancing data traffic method by resizing cell area based on PRX and users’ position. The authors in [101] and [102] analyzed the PRX in the handover decision process. Another work [103] applied a sliding window function on the PRX measurements to proceed with the decision. Although some of them analyzed the users’ position, none of them took into account the number of active users in the femtocells or the way the users are located in the scenario. Based on these works, the designed MLB mechanisms will focus on the prevention or reduction of temporary network congestion issues in open access LTE femtocells and commercial or corporate environments. This goal is accomplished by changing cell transmission power, hence, the cell service areas. In addition, the algorithms will be supported by classical mobile network indicators and context information such as users’ position to be able to improve the network performance in short period. 5.3 Problem description Femtocells are proposed as a solution to solve some of the current cellular challenges. Due to the proximity between femtocells and users, the user battery lifetime is increased and the user Quality of Experience (QoE) is enhanced. At the same time, operators also reduce CAPEX and OPEX. Nevertheless, femtocell networks present some shortcomings that must be addressed. Some of them are the unpredictable occurrences or occasional events that could provoke unexpected overload conditions in the network. These temporal and spatial variations, combined with the coverage holes and time-variant fading caused by reflections and obstacles, could negatively affect network performance. 5.3.1 Operators’ policy From the point of view of femtocell versus macrocell use, the operators could decide two different policies. On the one hand, the operator could be interested in a macrocell offload solution where macrocell data traffic hands over to femocells (when possible) to increase network capacity. Hence, once the femtocell is full (e.g., the maximum capacity of active users is reached), new incoming voice calls (e.g., VoLTE - Voice over LTE) that attempt to access are redirected to the macrocell, while the incoming data connections are accepted after handing over a voice traffic call to the macrocell. That situation could block many voice calls if the quality of the macrocell signal is poor indoors. 94 Indoor mobility load balancing techniques algorithm to prevent blocked calls. According to this, the new input indicator is related to bandwidth, by measuring the current cell capacity in radio resource terms. This indicator is hereafter described for LTE networks, although equivalent indicators can be defined for other technologies. Therefore, in PLS method, the load balance is performed based on the occupied PRB, whereas the previous PTS optimizer depended on blocked calls. According to this, the key network indicator for this system is represented as the following function, which is calculated for each cell: 𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐)= 𝐿𝑜𝑚𝐿(𝑚𝑒𝑐𝑐)−1 𝑁�𝐿𝑜𝑚𝐿(𝑖) 𝑁 𝑓=1 , (5.7) where 𝑁 is the number of neighboring cells and 𝐿𝑜𝑚𝐿(𝑖) is the ratio of occupied PRBs of the cell 𝑖, defined in equation (5.8). 𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐) is used as FLC input together with the femtocell transmission power deviation ∆𝛥𝑇𝑥(𝑚𝑒𝑐𝑐). 𝐿𝑜𝑚𝐿(𝑖)=𝑜𝑚𝑚𝑐𝑝𝑖𝑒𝐿_𝛥𝑒𝑒𝑜𝑐𝛥𝑚𝑒𝑒(𝑖) 𝑚𝑚𝑥 _𝛥𝑒𝑒𝑜𝑐𝛥𝑚𝑒𝑒(𝑖). (5.8) The method in [15] has been modified for solving temporary overloaded situations. The difference compared to [15] consists of changing the membership functions and fuzzy rules to converge to the optimal configuration in the shortest time. Hence, the range of outputs of the FLC is larger (from ±2 dB to ±6 dB). In this context, based on expert knowledge, some membership functions and fuzzy rules configurations have been evaluated. A large enough number of membership functions has been selected to achieve a reasonable level of detail while keeping the number of fuzzy sets small enough to build easy sets of fuzzy rules. Figure 5.4(a) shows the three membership functions of 𝜇𝑃(∆𝛥𝑇𝑥(𝑚𝑒𝑐𝑐)). It analyzes if femtocell transmission power deviation is very negative “Very Negative”, negative “Negative” or zero “Zero”. These functions were proposed by the PTS method and they are the same in PLS method to accomplish consistent and comparable results. The function 𝜇𝑦(𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐)), in Figure 5.4(b), depicts five membership functions to characterize the inputs in “Very Negative”, “Negative”, “Zero”, “Positive” and “Very Positive” according to the difference between the studied cell load ratio and the average load ratio of its neighboring cells, keeping the symmetry around 0 (i.e., the point where the load ratio between the studied cell and its neighbors is the same) to balance the load in the network. Each membership function is defined in an interval based on the expert knowledge, and for simplicity and computational efficiency, the selected membership functions are trapezoidal and triangular. Indoor mobility load balancing techniques 95 (a) Input 1 (b) Input 2 (c) Output Figure 5.4: Membership functions of PLS method. Table 5.1 defines the common-sense fuzzy rules in the form of “if-then” statements, following the syntax of equation (C.1) in Appendix C. These rules are built to balance the load in the network. In consequence, the more positive the 𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐) is (i.e., the studied cell is overloaded), the more negative the transmission power deviation should be (rules 10 and 11), and vice versa (rules 1 to 6). In case the network is balanced (𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓= "𝑍𝑒𝛥𝑜"), the algorithm tends to return to the initial transmission power configuration in order to improve SINR and reduce the number of PRBs per user as the CQI is increased (rules 7 to 9). In consequence, the cell would have more available free resources decreasing the probability of cell congestion. Based on the output function 𝜇𝑧(𝛿𝛥𝑇𝑥(𝑚𝑒𝑐𝑐)) shown on Figure 5.4(c), the femtocell transmission power is increased or decreased 𝛿𝛥𝑇𝑥(𝑚𝑒𝑐𝑐). The functions can take the 96 Indoor mobility load balancing techniques following labels and values: “Very Negative” and −6 dB, “Negative” and −3 dB, “Zero” and 0 dB, “Positive” and +3 dB and “Very Positive” and +6 dB. Finally, the femtocell transmission power 𝛥𝑇𝑥(𝑚𝑒𝑐𝑐) is tuned according to the output. Table 5.1: Fuzzy rules of PLS method. Loaddiff Operator ∆PTx δ PTx IF THEN 1 Very Negative AND Very Negative Very Positive 2 Very Negative AND Negative Very Positive 3 Very Negative AND Zero Positive 4 Negative AND Very Negative Very Positive 5 Negative AND Negative Positive 6 Negative AND Zero Positive 7 Zero AND Very Negative Very Positive 8 Zero AND Negative Positive 9 Zero AND Zero Zero 10 Positive AND - Negative 11 Very Positive AND - Very Negative This method has been defined focusing on an LTE network, although it could be easily extended to any other cellular technology. The PRB is bandwidth-related, therefore, an equivalent available bandwidth indicator can be used as input for any other cellular technologies. 5.4.1.3 Power User Sharing (PUS) Femtocells are usually designed to support from 2 to 64 users in connected mode, either voice or data traffic [32]. Rather than a wireless cellular bandwidth limitation, this characteristic is a restriction in the femtocell processing capability. For that reason, PRBs activity (as proposed in the PLS method) may not be always an appropriate indicator for balancing traffic in femtocell environments where most of the time there could be free resources to allocate user data but the femtocell is not able to process them due to the limit in the maximum number of active users. According to this, this PhD thesis proposes the PUS method, which considers the number of users in connected mode as the main indicator to offload temporary congested cells. In this case, the input indicator of the FLCs is defined as the ratio of active users, i.e., the number of simultaneous users in connected mode 𝑚𝑚𝑡𝑖𝑎𝑒_𝑐𝑒𝑒𝛥𝑒, to the Indoor mobility load balancing techniques 97 femtocell user limitation 𝑚𝑚𝑥_𝑐𝑒𝑒𝛥𝑒 (see equation (5.10)) of the studied cell 𝑈𝑒𝑒𝛥(𝑚𝑒𝑐𝑐), in relation to the same average ratio in its neighboring cells 𝑈𝑒𝑒𝛥(𝑖): 𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐)= 𝑈𝑒𝑒𝛥(𝑚𝑒𝑐𝑐)−1 𝑁�𝑈𝑒𝑒𝛥(𝑖) 𝑁 𝑓=1 , (5.9) where 𝑁 is the number of neighboring cells and 𝑈𝑒𝑒𝛥(𝑖) is defined for a cell 𝑖 as: 𝑈𝑒𝑒𝛥(𝑖)=𝑚𝑚𝑡𝑖𝑎𝑒_𝑐𝑒𝑒𝛥𝑒(𝑖) 𝑚𝑚𝑥_𝑐𝑒𝑒𝛥𝑒(𝑖). (5.10) This mechanism presents the same FLC structure as the previous PTS and PLS methods, but the inputs of the FLC are the new indicator, 𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐) as presented in equation (5.9), and the femtocell transmission power deviation, ∆𝛥𝑇𝑥(𝑚𝑒𝑐𝑐). The output of the FLC would increase/decrease the current femtocell transmission power. In order to get a moderate level of detail and straightforward fuzzy control rules, an acceptable number of membership functions have been selected for each indicator based on experienced human knowledge. The membership functions of this controller are shown in Figure 5.5. As it can be observed, the membership functions 𝜇𝑃(∆𝛥𝑇𝑥(𝑚𝑒𝑐𝑐)) are the same as the previous membership functions of the PTS and PLS methods (Figure 5.5(a)), and the membership functions, 𝜇𝑦(𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐)), of the new input, 𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐), are: “Very Negative”, “Negative”, “Zero”, “Positive” and “Very Positive” (Figure 5.5(b)). Notice that the shape of these membership functions are similar to those of the previous method. However, the interval of definition is adjusted by experts to get the best algorithm’s performance. For simplicity and computational efficiency, the implemented membership functions are triangular and trapezoidal. Fuzzy rules are depicted in Table 5.2 and aim to equalize the ratio of active users in the studied cell and the average ratio of its neighboring cells. According to this, the more positive 𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐) is (i.e., the ratio in the studied cell is high), the more negative the transmission power deviation should be (rules 8 and 9), and vice versa (rules 1 to 4). Once the network is equalized (𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐) = "𝑍𝑒𝛥𝑜"), the PUS method avoids very negative transmission power adaptations (label “Very Negative” - rule 5) in order to increase the end-users QoE. However, as this method is restricted by the number of users (instead of the available resources in the cell), returning to the default transmission power could increase the interference but the cell congestion would not be reduced. That is why “Negative” or “Zero” values of transmission power deviation (rules 6 and 7) do not return to the default value. 98 Indoor mobility load balancing techniques a) Input 1 b) Input 2 c) Output Figure 5.5: Membership functions of PUS method. The same output functions 𝜇𝑧(𝛿𝛥𝑇𝑥(𝑚𝑒𝑐𝑐)), as in previous algorithms, are defined for this one (illustrated on Figure 5.5(c)). Most of the time, this mechanism is suitable for achieving good performance in femtocell networks, above all in LTE deployments which support high-peak data rates up to 346 Mbps in the downlink and 85.5 Mbps in the uplink at 20 MHz and 4x4 MIMO. However, it could present some shortcomings when propagation channel is very poor (bad SINR) or users require high transmission rates (e.g., to watch streaming television in high definition). Indoor mobility load balancing techniques 99 Table 5.2: Fuzzy rules of PUS method. Userdiff Operator ∆PTx δ PTx IF THEN 1 Very Negative AND Very Negative Very Positive 2 Very Negative AND Negative Very Positive 3 Very Negative AND Zero Positive 4 Negative AND - Positive 5 Zero AND Very Negative Positive 6 Zero AND Negative Zero 7 Zero AND Zero Zero 8 Positive AND - Negative 9 Very Positive AND - Very Negative 5.4.1.4 Power Load and User Sharing (PLUS) The previous mechanisms, as explained, might not guarantee their proper operation in some situations. The shortcomings of each algorithm (PLS and PUS methods) could be complemented by the other, leaving outside the PTS method due to the long time needed to evaluate its main indicator (CBR). That means, the analysis of the two indicators (cell load and number of active users) is required to ensure an efficient MLB process in femtocell networks. According to this, the works described in the literature in the context of MLB at femtocell networks that did not analyze at least these two indicators, might not properly work. However, in case the operators’ policies and priorities are focused on voice traffic (e.g., VoLTE), the analysis of active users might be enough while for data traffic, the analysis of radio resources would be desirable. In the proposed PLUS method, both previously defined indicators (𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐) and 𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐)) are the inputs of the FLC to properly prevent or reduce occasional indoor congestions. For simplicity, this method does not include the cell transmission power deviation as an input indicator. The same membership functions proposed for PLS and PUS methods are implemented in this FLC to compare these methods under the same conditions (reader is referred to Figure 5.4(b) and Figure 5.5(b)). New fuzzy rules are defined based on these two indicators and the expert knowledge. Table 5.3 presents the set of control rules implemented in this system. These rules prioritize the indicator 𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓 over the 𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓 since a new user in a femtocell network usually implies a greater impact on the maximum allowed number of user than on the available resources. That means, the network resources assigned to 100 Indoor mobility load balancing techniques each user could be reduced in order to accept new users, even if some of them could be slightly dissatisfied (decrease the QoE, throughput, etc.). For example, under a “Very Negative” situation of 𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓 and a “Very Positive” situation of 𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓, it is preferred to decrease the transmission power of the studied cell (fuzzy output is “Negative”). Table 5.3: Fuzzy rules of PLUS method (Operator: 𝐴𝑁𝐼; output: 𝛿𝛥𝑇𝑥). Loaddiff Userdiff Very Negative Negative Zero Positive Very Positive Very Negative Very Positive Very Positive Positive Zero Zero Negative Very Positive Positive Positive Zero Zero Zero Very Positive Positive Positive Negative Very Negative Positive Zero Zero Negative Negative Very Negative Very Positive Negative Negative Very Negative Very Negative Very Negative The same constant output functions 𝜇𝑧(𝛿𝛥𝑇𝑥(𝑚𝑒𝑐𝑐)), as in the previous algorithms, are applied here to compare all the algorithms under the same conditions. Table 5.4: Summary of Fuzzy-based MLB Mechanisms. PTS* PLS PUS PLUS Input 1 ∆𝛥𝑇𝑥 ∆𝛥𝑇𝑥 ∆𝛥𝑇𝑥 𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓 Input 2 𝐶𝐶𝑈𝑏𝑓𝑓𝑓 𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓 𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓 𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓 Output 𝛿𝛥𝑇𝑥 𝛿𝛥𝑇𝑥 𝛿𝛥𝑇𝑥 𝛿𝛥𝑇𝑥 * Reference [16] A summary of the supported inputs and outputs of the MLB mechanisms is depicted in Table 5.4. 5.5 Methods for context-aware MLB in femtocell networks Context information could enhance the mobile network performance as it could support mechanisms in real-time, add extra information about the scenario, inform about future events, etc. However, current OAM architectures usually manage network Indoor mobility load balancing techniques 101 parameters in periods of fifteen minutes, hours or even days, or they are not able to integrate such as context information. It could be inefficient in dynamic scenarios such as indoor environments. In this sense, the local-centralized OAM architecture presented in Chapter 3 is a solution for implementing real-time self-management mechanisms supported by context information. The following subsections initially describe the information sources and subsequently, the proposed heuristic context-aware MLB methods are explained. These methods aim to mitigate temporal and focused overloaded situations at commercial and corporate indoor scenarios due to the high concentration of users in temporal and spatial domain, thanks to the knowledge of users’ position. Two context-aware MLB methods are proposed. The first method uses the geometrical distances between users, their distributions and their PRX to estimate the new femtocells transmission power. The second method updates femtocells transmission power based on the average PRX stored in a database when users’ position is available, in other case, the average PRX values from measurement reports are analyzed. Finally, a simple method of the energysaving (ES) use case [45] is proposed in coordination with a MLB mechanism. 5.5.1 System set-up The main network parameter that will be tuned is the femtocell transmission power. To address this operation, the PRX per terminal is used by the system to calculate the new cell transmission power. That information could be obtained or estimated from different sources (propagation models, measurement reports, etc.). Radio propagation models, as empirical mathematical formulas, characterize radio wave propagations. These models are very sensitive to pedestrians, obstacles, etc., which vary both in time and space, especially at indoor environments. Other sources such as measurement reports provide instantaneous information about PRX, while historical PRX values provide an average estimation of PRX. Thus, these PRX values could be different depending on the source, the influence of fading and shadowing, etc. 5.5.1.1 PRX from measurement reports Mobile devices provide valuable information about the network conditions through the Measurement Reports (MR). These reports are periodically sent to the cell and contain information about the channel quality (current mobile transmitted/received power, block error ratio of the data channel, etc.). This is vital to assist dynamic 102 Indoor mobility load balancing techniques network planning and RRM processes in power control decisions and handover. 3GPP specifications define this type of measurements in [79]. The PRX is included in the measurement reports, although named in a different way depending on the radio technology. For 2G deployments, it is called RxLev (Received Signal Level) whereas, for 3G, it is named as RSCP (Received Signal Code Power) and in LTE as RSRP (Reference Signal Received Power). That information is periodically forwarded to the OAM layer from the base stations. In practice, as aforementioned, the mobile network does not support continuous real-time communication to the OAM elements due to the amount of signaling data and the high-level location of the OAM in the hierarchy. However, the OAM architecture described in Chapter 3 placed some of its functions in lowest levels, making this realtime process viable. This is in line with distributed SON architectures proposed by 3GPP for future networks [2]. The PRX could be also monitored thanks to context sources such as smartphone apps (e.g., G-MoN application [78]). Unfortunately, some reported information could be incomplete due to terminal compatibility issues, i.e., measurement capabilities would depend on the terminal manufacturer. For example, the PRX from the serving cell is always reported but PRX from neighboring cells is missing. Even if the terminal is monitoring the PRX from its neighboring cells, the app is not able to report it. As a solution, the PRX information per terminal could be requested to a database with historical PRX values based on users’ positions. 5.5.1.2 Historical Path Loss Maps (HPLM) The PRX depends on several factors such as the propagation channel, the cell transmission power, the users’ positions, etc. However, since the signal Path Loss (PL) is independent of the cell transmission power, the PL information will be used instead of the PRX to create a database. PL values are calculated and stored into this database together with the users’ positions, to build the so named Historical Path Loss Maps (HPLM). The relationship between PRX and PL is given by: 𝛥𝑈𝑋𝑏𝑏𝑏𝑏(𝑥,𝑦) = 𝛥𝑇𝑋𝑏𝑏𝑏𝑏−𝛥𝐿𝑏𝑏𝑏𝑏(𝑥,𝑦), (5.11) where 𝛥𝑈𝑋𝑏𝑏𝑏𝑏(𝑥,𝑦) is the received power at (𝑥,𝑦) position from 𝑚𝑒𝑐𝑐, 𝛥𝑇𝑋𝑏𝑏𝑏𝑏 is the Equivalent Isotropically Radiated Power (EIRP) of 𝑚𝑒𝑐𝑐 and 𝛥𝐿𝑏𝑏𝑏𝑏(𝑥,𝑦) is the radio signal path loss at point (𝑥,𝑦) from 𝑚𝑒𝑐𝑐. The database contains the PL per position received from each cell. This information is calculated based on the MRs (i.e., PRX Indoor mobility load balancing techniques 103 values) performed by the terminals over time, the network configuration parameters (i.e., cells transmission power) and the users’ positions. Thanks to this HPLM database, the system will be able to get an estimation of the PRX per position based on equation (5.11). Building the HPLM database is a continuous process composed by two sets of information (Figure 5.6). On the one hand, the measurements reported from the active users are used to calculate the current PL from each cell 𝛥𝐿𝑏𝑏𝑏𝑏(𝑖,𝑗)𝑐, where coordinates (𝑖,𝑗)𝑐 are the position of user 𝑐. On the other hand, and synchronized with the MR, the active user position (𝑖,𝑗)𝑐 is supplied to the database by an external indoor positioning system (as a possible system, Chapter 4 presented an RFID-based indoor positioning system) [38]. With this data (measurements and positions) the average PL of a position and cell is updated according to the following equation: 𝛥𝐿 �𝑏𝑏𝑏𝑏(𝑖,𝑗)=1 𝐻(𝑖,𝑗)�𝛥𝐿𝑏𝑏𝑏𝑏(𝑖,𝑗)𝑎 𝑀(𝑓,𝑗) 𝑎=1 , (5.12) where 𝛥𝐿𝑏𝑏𝑏𝑏(𝑖,𝑗)𝑎 are the different path loss measurements over time at position (𝑖,𝑗) and 𝐻(𝑖,𝑗) is the total number of measurements at position (𝑖,𝑗) (coming from different users). That 𝛥𝐿 �𝑏𝑏𝑏𝑏(𝑖,𝑗) would be used to estimate the 𝛥𝑈𝑋 �𝑏𝑏𝑏𝑏(𝑖,𝑗), according to equation (5.11). a) Acquisition of information b) Estimation of PL Figure 5.6: HPLM scheme. The radio channel conditions, especially at indoor environments, suffer continuous changes due to the number of people, obstacles, etc. Therefore, the number of samples, 𝐻(𝑥,𝑦), to build the estimated PL value per position should be a configurable parameter and could be quite different depending on the scenario. For this work, the number of samples is limited in time to keep the propagation channel conditions updated: 110 Indoor mobility load balancing techniques Those situations would be managed thanks to the so-called Serving Cell Maps (SCM). Calculating SCM is a process composed of two steps. In the first step, the new coverage areas are defined based on geo-located PRX matrices supported by the HPLM database. This procedure is based on Thiessen Polygons [109] where, for each cell, an irregular polygon 𝐴𝑏𝑏𝑏𝑏(𝑏) that delimitates its serving cell area is generated: 𝐴𝑏𝑏𝑏𝑏(𝑏)={(𝑥,𝑦)∈(𝑋,𝑌)| 𝛥𝑈𝑋(𝑥,𝑦)𝑏𝑏𝑏𝑏(𝑏)>𝛥𝑈𝑋(𝑥,𝑦)𝑏𝑏𝑏𝑏(𝑗) ∀ 𝑗≠𝑘, (5.17) where (𝑋,𝑌) are the coordinates of the scenario. This process builds a network map as that presented in Figure 5.10(a), where for a certain femtocell (r) two disconnected coverage areas are presented. Therefore, the next step is to determine which the main area is. For that purpose, those areas 𝐴𝑏𝑏𝑏𝑏(𝑏) that are split into 𝑆 subareas, 𝐴𝑏𝑏𝑏𝑏(𝑏) 𝑓, must be analyzed. Those subareas are compared to find the biggest one in order to remove overshooting and keep the most suitable area: 𝐴′𝑏𝑏𝑏𝑏(𝑏)=𝑚𝑚𝑥�𝐴𝑏𝑏𝑏𝑏(𝑏) 𝑓� 𝐴𝑏𝑏𝑏𝑏(𝑏)=�𝐴𝑏𝑏𝑏𝑏(𝑏) 𝑓 𝑆 𝑗=1 }. (5.18) In this sense, those removed areas must be joined to other/s area/s. Therefore, each area 𝐴𝑏𝑏𝑏𝑏(𝑎)�𝑝≠𝑘 must be updated by including those discarded areas 𝐴𝑏𝑏𝑏𝑏(𝑏) 𝑓�𝐴𝑏𝑏𝑏𝑏(𝑏) 𝑓≠𝐴𝑏𝑏𝑏𝑏(𝑏) ′ according to: 𝐴′𝑏𝑏𝑏𝑏(𝑎)=𝐴𝑏𝑏𝑏𝑏(𝑎)+�𝐴𝑏𝑏𝑏𝑏(𝑎) ℎ 𝐻 ℎ=1 , (5.19) where 𝐻 is the number of new coverage subareas belong to 𝑚𝑒𝑐𝑐(𝑝) and 𝐴𝑏𝑏𝑏𝑏(𝑎) ℎ is calculated as equation (5.20) but focusing on those discarded areas and avoiding the PRX information of 𝑚𝑒𝑐𝑐(𝑘): 𝐴𝑏𝑏𝑏𝑏(𝑎) ℎ=�(𝑥,𝑦)∈�𝑋𝑏𝑏𝑏𝑏(𝑏) 𝑓,𝑌𝑏𝑏𝑏𝑏(𝑏) 𝑓�� 𝛥𝑈𝑋(𝑥,𝑦)𝑏𝑏𝑏𝑏(𝑎)>𝛥𝑈𝑋(𝑥,𝑦)𝑏𝑏𝑏𝑏(𝑞) ∀ 𝑞≠𝑝,𝑘}. (5.20) This process is repeated till all subareas are joined to a main cell. Finally, the proposed serving cells coverage areas are illustrated on Figure 5.10(b), where the small subarea of the red cell (r) is embedded into the yellow cell (y) area. Indoor mobility load balancing techniques 111 5.5.3.1.2 Neighbor cell maps A new virtual map, named Neighbor Cell Maps (NCM), is designed to estimate the neighboring cells of each serving cell. The aim of these maps is to create the proposed neighbor cell lists. (a) Proposed SCM (b) Proposed NCM Figure 5.11: Calculation of proposed neighbor cell list. For that purpose, the same procedure to build SCM is followed now but, removing the PRX values of the serving cell 𝑚𝑒𝑐𝑐(𝑘). Then, equations (5.17) to (5.20) generate the new NCM (see Figure 5.11(b)). Those new areas which cover the original area 𝐴′𝑏𝑏𝑏𝑏(𝑏) (from SCM) are denoted as 𝐴𝑏𝑏𝑏𝑏(𝑗→𝑏), where 𝑚𝑒𝑐𝑐(𝑗→𝑘) means that the cell 𝑚𝑒𝑐𝑐(𝑗) covers the area of the serving cell 𝑚𝑒𝑐𝑐(𝑘). Figure 5.11(b) depicts an example where three cells (𝑗= {𝑏,𝑦,𝑔}) are adjacent to the serving cell (𝑘= {𝛥}). Finally, the proposed neighbor cell list is created as follows: 𝐶𝑏𝑏𝑏𝑏(𝑏)={𝑚𝑒𝑐𝑐(𝑗) | 𝐴𝑏𝑏𝑏𝑏(𝑗→𝑏)∈𝐴′𝑏𝑏𝑏𝑏(𝑏)∀ 𝑗≠𝑘}. (5.21) 5.5.3.2 System implementation The proposed context-aware MLB system (based on VM or MR methods) ensures a consistent solution to those operators aiming at prioritizing voice traffic users over data traffic users in indoor femtocell environments, as the impact to the end user is more frustrating in voice calls than in data traffic. For that purpose, the system provides free resources for voice traffic when the indoor network is temporary congested, whereas data traffic is handed over to macrocells or suffers outage for a while (that would be managed by the AC or the Schedulers). Additionally, the ratio of occupied radio resources is considered negligible in comparison with the ratio of active users in LTE networks due to the high-peak data rates supported and the very low 112 Indoor mobility load balancing techniques number of active users. For this reason, cell load is only analyzed based on the ratio of active user (see equation (5.10)). The global scheme of the proposed heuristic rule-based mechanism is depicted in Figure 5.12. It shows an iterative mechanism to estimate the new transmission power variations of the femtocells, 𝛿𝛥𝑇𝑥𝑏𝑏𝑏𝑏. The system iterates while it evaluates that any femtocell could be overloaded with the calculated new transmission power variation. That new network configuration would be estimated based on the VMs. Once the system achieves a balance situation, those new transmission power variations are set in femtocells. The system inputs are the cellular network KPIs, the PRX and the position of the terminals. The system output is a vector, 𝛿𝛥𝑇𝑥 � � � � � � � � � �  , with the transmission power variation that should be applied to each femtocell. Additionally, another vector, 𝐹, indicates the activity of the femtocells (whether a femtocell has active users or not). Figure 5.12: Context-aware MLB scheme. Note that, when users’ positions are not available or the indoor positioning system statistical parameters are unsuitable for improving the performance of the system due to high position error, PRX and cell data are directly read from the MRs and KPIs. Additionally, two thresholds are defined to consider a serving cell is overloaded and its neighboring cells are low-loaded: • 𝑆𝑎ℎ is the minimum ratio of users (see equation (5.10)) to consider a serving cell as highly loaded. Its range goes from 0 (cell is considered highly loaded) to 1 (cell is considered highly loaded). • 𝑇𝑎ℎ is the maximum average ratio of users in the neighboring cells of the studied cell to consider the neighboring cells as low loaded and, consequently, available to catch users to offload the studied cell. Its range goes from 0 to 1. These thresholds are configured based on a sensitivity study or on the operators’ experience, policies or priorities. The initial condition 𝑆𝑎ℎ>𝑇𝑎ℎ should be mandatory to avoid possible ping-pong effects and oscillations in the MLB process. Indoor mobility load balancing techniques 113 The main modules of the MLB method are described in the next subsections according to Figure 5.12. As an iterative system, once the new femtocells transmission powers are estimated, this procedure analyzes the network again based on these estimations to detect possible overloaded femtocells with the new configuration. 5.5.3.2.1 Estimation of users to HO The aim of this module is to provide an estimation of the number of users that should leave the selected overloaded cell, 𝑐𝑏𝑏𝑏𝑏, to hand over towards a target cell. The scheme of this module is depicted in Figure 5.13. Firstly, the system examines the ratio of users per femtocell 𝑘 (see equation (5.10), hereafter referred as 𝑐𝑏𝑏𝑏𝑏(𝑏)), where 𝑚𝑚𝑡𝑖𝑎𝑒_𝑐𝑒𝑒𝛥𝑒𝑏𝑏𝑏𝑏 could be either 1) the number of active users in the femtocell (for the MR method) or 2) the number of users in the coverage area of the femtocell based on the SCM when users’ positions are available (for the VM method). Figure 5.13: Scheme of estimation of users to HO. All 𝑐𝑏𝑏𝑏𝑏(𝑏) are ranked to get the maximum ratio, 𝐿𝑏𝑏𝑏𝑏, and the highest overloaded cell. Hereafter that cell will be called studied cell. The procedure could return to this point to select the next ranked cell when the studied cell could not be offloaded in this iteration (see next subsection). The condition 𝐿𝑏𝑏𝑏𝑏> 𝑆𝑎ℎ makes the algorithm continue, in other case, this module finishes with 𝑐𝑏𝑏𝑏𝑏= 0 (no users should leave the studied cell in this iteration). In case the condition is accomplished, users are classified per service. Due to the interest of satisfying voice call users, a basic classification of traffic into two main services is taken into account: voice connections and other services. In case there are users with other services (data traffic), these users are forced to handover to macrocells one by one after evaluating the initial condition again, 𝐿𝑏𝑏𝑏𝑏 > 𝑆𝑎ℎ. Note L cell > S th u cell =0 No Ratio of estimated users per serving cell ( L cell) L nc < T th Handover to macrocell Ratio of estimated users at neighbor cells ( L nc) Estimation of leaving users Data traffic (no voice)? Yes Yes No Yes No u cell Proposed neighbor cell list ( C cell) 114 Indoor mobility load balancing techniques that, in most cases, there is no macrocell coverage indoors. Hence, most of these data traffic users could be in outage for a while till a femtocell is available. Since normally there is a high overlapping between femtocells because of their unplanned deployments, the studied cell would usually include a lot of cells in the neighbor cell lists. However, when users’ positions are available, these lists would be changed by the proposed neighbor cell lists of the studied cell, 𝐶𝑏𝑏𝑏𝑏, based on NCM, described in “5.5.3.1.2 Neighbor cell maps”. It would reduce overshooting issues. Now, it is time to determine whether the neighboring cells have space to allocate new users or not. In this sense, the average of the ratio of users in the neighboring cells is calculated (same as the summation term of equation (5.9)): 𝐿𝑐𝑏=1 𝑁𝐶𝑏𝑏𝑏𝑏∙� 𝑐𝑏𝑏𝑏𝑏(𝑓) 𝑁𝐶𝑎𝑎𝑓𝑓 𝑓=1 , (5.22) where 𝑐𝑏𝑏𝑏𝑏(𝑓) is calculated according to equation (5.10), 𝑖 refers to the identification of the neighboring cells and 𝑁𝐶𝑏𝑏𝑏𝑏is the length of that list. After that, the situation could be: • 𝐿𝑐𝑏≥𝐿𝑏𝑏𝑏𝑏, which means that the studied cell has less camped users than the average camped users of its neighboring cells. Therefore, no users should leave that cell because the situation out of this cell is the same or worst. • 𝐿𝑐𝑏<𝐿𝑏𝑏𝑏𝑏, which means that the situation of the studied cell could be critical but it could be solved as, in average, its neighboring cells present lower number of camped users. In consequence, some users of the studied cell will hand over to the most suitable neighboring cells. To ensure good QoE, avoid ping-pong effects and continue with the offloading of the studied cell, the condition 𝑇𝑎ℎ>𝐿𝑐𝑏 should be accomplished. Otherwise, there are no close offloaded cells to share and move users towards. This situation could be solved by offloading those neighboring cells in the following iterations. Hence, the system would return to the first module and select the next ranked 𝑚𝑒𝑐𝑐 in terms of the ratio of active users (𝑐𝑏𝑏𝑏𝑏). That new 𝑚𝑒𝑐𝑐 would be the new studied cell. Note that, if the system is not able to solve this unlikely situation, an event would be trigger to warn the network engineers about this problematic situation. Engineers will study this case and evaluate the planning of the deployment of additional femtocells (if this case is repeated several times in the same place) or doing nothing because it is an isolate case or because 𝑆𝑎ℎ and 𝑇𝑎ℎ should be reconfigured (bad configuration or very restricted thresholds). Indoor mobility load balancing techniques 115 Finally, the number of users, 𝑐𝑏𝑏𝑏𝑏, that should leave the studied cell are calculated: 𝑐𝑏𝑏𝑏𝑏=𝑚𝑚𝑡𝑖𝑎𝑒_𝑐𝑒𝑒𝛥𝑒𝑏𝑏𝑏𝑏−𝑚𝑒𝑖𝑐{max _𝑐𝑒𝑒𝛥𝑒𝑏𝑏𝑏𝑏∙�𝐿𝑏𝑏𝑏𝑏 + ∑𝑐𝑏𝑏𝑏𝑏(𝑓) 𝑁𝐶 𝑓=1 𝑁𝐶𝑚𝑒𝑐𝑐+ 1 �}, (5.23) where 𝑚𝑒𝑖𝑐 function gets the nearest integer towards infinity, 𝑚𝑚𝑡𝑖𝑎𝑒_𝑐𝑒𝑒𝛥𝑒𝑏𝑏𝑏𝑏 could be either 1) the number of active users in the femtocell (for the MR method) or 2) the number of users in the coverage area of the femtocell based on the SCM when users’ positions are available (for the VM method), 𝑚𝑚𝑥_𝑐𝑒𝑒𝛥𝑒𝑏𝑏𝑏𝑏 is the total number of users that the femtocell could allocate, 𝐿𝑏𝑏𝑏𝑏 is the ratio of users per cell and 𝑁𝐶𝑏𝑏𝑏𝑏 is the number of neighboring cells. 5.5.3.2.2 Calculation of 𝜹𝜹𝑻𝜹 � � � � � � � � � � �  This module decides the transmission power variation that must be applied to each femtocell to move users and to avoid overloaded cells in the network. The scheme is illustrated in Figure 5.14. Figure 5.14: Scheme of calculation of 𝛿𝛥𝑇𝑥 � � � � � � � � � �  block. Once the candidate cell to be offloaded has been selected and the number of users that should leave the cell calculated, it is necessary to determine the PRX of users in that cell. The system proposes two ways to obtain the PRX depending on whether the users are located or not: • PRX from HPLM: users’ positons are required to estimate the value of its PRX. In this case, equation (5.11) estimates those values of PRX for each user’ position and femtocell, based on the stored information of HPLM and equations (5.12) and (5.14). Calculation of Δ u cell > 0 Yes Estimation of δPTx cell Remove user No Δ > 0 No Estimation of δPTx target Yes δPTx • max(u cell ) • Reset parameters 116 Indoor mobility load balancing techniques • PRX from MR: the PRX is directly acquired from the mobile reports. It provides actual data about the propagation channel. When there is no information about the PRX from a given neighboring cell, the PRX for such cell is set to -130 dBm. This process starts with the initialization of 𝛿𝛥𝑇𝑥 � � � � � � � � � �  to zero or, in case the system had already iterated, with its previous value. Then, the calculation of ∆ is performed. These parameters are calculated by elementary equations: additions, and operators: maximum and minimum. It would reduce the computational complexity and costs. For each user 𝑔(𝑚) (where 𝑚= {1,2, … 𝑚𝑚𝑡𝑖𝑎𝑒_𝑐𝑒𝑒𝛥𝑒𝑏𝑏𝑏𝑏}) of the studied cell, the PRX value from its serving cell, 𝛥𝑈𝑋𝑏𝑏𝑏𝑏 𝑔(𝑎), and the maximum PRX value from its neighboring cells, 𝑈𝑆𝑆𝐼𝑐𝑏(𝑎) 𝑔(𝑎), are defined. Being 𝐼𝑚(𝑚) the neighboring cells with the highest value of PRX for the studied user 𝑚. Remind that this process could be done, either with information from MRs or, based on the VMs (SCM and NCM). The following two vectors composed of those values are created: 𝛥𝑈𝑋𝑏𝑏𝑏𝑏 � � � � � � � � � � � � � � �  =�𝛥𝑈𝑋𝑏𝑏𝑏𝑏 𝑔(1),𝛥𝑈𝑋𝑏𝑏𝑏𝑏 𝑔(2)…𝛥𝑈𝑋𝑏𝑏𝑏𝑏 𝑔�𝑐𝑏𝑎𝑓𝑎𝑏_𝑐𝑐𝑏𝑃𝑐𝑎𝑎𝑓𝑓��, (5.24) 𝛥𝑈𝑋𝑐𝑏 � � � � � � � � � � � � �  =�𝛥𝑈𝑋𝑐𝑏(1) 𝑔(1),𝛥𝑈𝑋𝑐𝑏(2) 𝑔(2)…𝛥𝑈𝑋𝑐𝑏(𝑐𝑏𝑎𝑓𝑎𝑏_𝑐𝑐𝑏𝑃𝑐) 𝑔�𝑐𝑏𝑎𝑓𝑎𝑏_𝑐𝑐𝑏𝑃𝑐𝑎𝑎𝑓𝑓��. (5.25) After that, the minimum value of the difference between these two vectors is saved: ∆𝑎𝑐𝑃𝑔𝑏𝑎=min�𝛥𝑈𝑋𝑏𝑏𝑏𝑏 � � � � � � � � � � � � � � �  −𝛥𝑈𝑋𝑐𝑏 � � � � � � � � � � � � �  �, (5.26) where 𝑡𝑚𝛥𝑔𝑒𝑡 refers to the neighboring cell that minimizes previous equation. That neighboring cell 𝐼𝑚(𝑚) which makes ∆𝑎𝑐𝑃𝑔𝑏𝑎 minimum, is selected (𝑡𝑚𝛥𝑔𝑒𝑡=𝐼𝑚(𝑚), possible target cell for that user). In case that ∆𝑎𝑐𝑃𝑔𝑏𝑎 is positive, the transmission power variation 𝛿𝛥𝑇𝑥𝑎𝑐𝑃𝑔𝑏𝑎 applied to that target cell is calculated as follows in equation (5.27), where 𝛿𝛥𝑇𝑥𝑎𝑐𝑃𝑔𝑏𝑎 is equal to ∆𝑓𝑎𝑡𝑡𝑎𝑓 2 the first iteration because initial values of 𝛿𝛥𝑇𝑥𝑎𝑐𝑃𝑔𝑏𝑎 are set to zero. Otherwise, ∆𝑎𝑐𝑃𝑔𝑏𝑎 is negative, this step is skipped and the transmission power variation 𝛿𝛥𝑇𝑥𝑎𝑐𝑃𝑔𝑏𝑎 is not estimated. ∆𝛥𝑇𝑥𝑎𝑐𝑃𝑔𝑏𝑎 =𝑚𝑚𝑥�∆𝛥𝑇𝑥𝑎𝑐𝑃𝑔𝑏𝑎 ,∆𝑎𝑐𝑃𝑔𝑏𝑎 2�. (5.27) Indoor mobility load balancing techniques 117 That user is removed from vectors 𝛥𝑈𝑋𝑏𝑏𝑏𝑏 � � � � � � � � � � � � � � �  and 𝛥𝑈𝑋𝑐𝑏 � � � � � � � � � � � � �  and 𝑐𝑏𝑏𝑏𝑏 decreases one unit. The loop is repeated till 𝑐𝑏𝑏𝑏𝑏 reaches the zero value, then, the studied cell transmission power variation 𝛿𝛥𝑇𝑥𝑏𝑏𝑏𝑏 is calculated: 𝛿𝛥𝑇𝑥𝑏𝑏𝑏𝑏=−�𝑜𝑖𝑖𝑒𝑒𝑡+𝑚𝑚𝑥�𝛿𝛥𝑇𝑥𝑎𝑐𝑃𝑔𝑏𝑎 � � � � � � � � � � � � � � � � � � � � � � �  ��, (5.28) where 𝑜𝑖𝑖𝑒𝑒𝑡 is the parameter involved in the handover process (e.g., event A3 in LTE [79]) defined to reduce ping-pong handovers, etc. and 𝛿𝛥𝑇𝑥𝑎𝑐𝑃𝑔𝑏𝑎 � � � � � � � � � � � � � � � � � � � � � � �  is the vector composed of all the target neighboring cells transmission power variations. Note that the decrease in the studied cell transmission power is truncated by 𝜏𝑎𝑓𝑐, as well as the neighboring cells transmission power has an upper limit of 𝜏𝑎𝑐𝑃. This condition is required to guarantee the QoS and to accomplish the maximum transmission power of the cells, respectively. Finally, the output of this module is the vector: 𝛿𝛥𝑇𝑥 � � � � � � � � � �  =�𝛿𝛥𝑇𝑥𝑏𝑏𝑏𝑏,𝛿𝛥𝑇𝑥𝑎𝑐𝑃𝑔𝑏𝑎 � � � � � � � � � � � � � � � � � � � � � � �  �. (5.29) The system emulates how the new network configuration, 𝛿𝛥𝑇𝑥 � � � � � � � � � �  , impacts on the deployed cells and determine whether it would be well-balanced or another cell needs to be analyzed because there are still overload issues (𝐿𝑏𝑏𝑏𝑏> 𝑆𝑎ℎ). In order to do that, new values of PRX must be estimated per user (𝛥𝑈𝑋′). To determine those values of 𝛥𝑈𝑋′, the variable 𝛥𝑇𝑋𝑏𝑏𝑏𝑏 of equation (5.11) is replaced by 𝛥𝑇𝑋𝑏𝑏𝑏𝑏+𝛿𝛥𝑇𝑥𝑏𝑏𝑏𝑏, i.e., the new EIRP that should be configured in that cell. According to 𝛥𝑈𝑋′ values and going back to the first block “5.5.3.2.1 Estimation of users to HO”, the new number of active users per 𝑚𝑒𝑐𝑐𝑓 is estimated based on the number of users that accomplish: {𝛥𝑈𝑋′𝑏𝑏𝑏𝑏𝑓>𝛥𝑈𝑋′𝑏𝑏𝑏𝑏𝑗 ∀ 𝑖≠𝑗}. With the availability of users’ positions, new SCM and NCM are built to estimates the serving areas and the proposed neighboring cell list. 5.5.3.2.3 Cell activity Once the system has determined the network would be well-balanced (𝐿𝑏𝑏𝑏𝑏< 𝑆𝑎ℎ), the last block provides the configuration as an output file. It contains information about cells transmission power variation, 𝛿𝛥𝑇𝑥 � � � � � � � � � �  , and the activity of the cells, 𝐹= {𝑖𝑏𝑏𝑏𝑏(1), 𝑖𝑏𝑏𝑏𝑏(2), … 𝑖𝑏𝑏𝑏𝑏(𝑏)}. The latter indicates when a cell has camped users: 𝑖𝑏𝑏𝑏𝑏=�1𝑚𝑒𝑐𝑐 ℎ𝑚𝑒 𝑚𝑚𝑚𝑝𝑒𝐿 𝑐𝑒𝑒𝛥𝑒 0𝑜𝑡ℎ𝑒𝛥𝑤𝑖𝑒𝑒 . (5.30) 118 Indoor mobility load balancing techniques 5.5.3.3 Load balancing and energy savings coordination As MLB mechanisms are femtocell transmission power related, the modification of this parameter might be highly susceptible to logical or parametric dependencies with other SON algorithms. As a consequence, many conflicts could arise (e.g., tuning the same network parameter) during mobile network operation due to these dependencies, undermining the stability of the network performance. The coordination of SON algorithms is an important challenge that must be addressed to ensure the highest network performance [110]. Those cells with no close users could turn off its radio propagation activity and standby their internal procedures by keeping the cell in dormant mode or switched off. As a consequence, a substantial reduction of mobile network energy consumption could be reached. This use case is referred as Energy-Savings (ES) [45]. Since it is power related, the coordination between this mechanism and other power related SON methods is mandatory to avoid potential conflicts in the network parameters adaptation, ensuring the system robustness and reliability [110]. Based on this, a location-based ES algorithm and its coordination with proposed MLB mechanisms are developed to ensure users’ accessibility and retainability, at the same time reducing the overall power consumption. These mechanisms analyze the network indicators and the information about the positions of the terminals (see the scheme in Figure 5.15). Figure 5.15: Coordinated location-based self-optimization scheme. This mechanism switches on/off or turns into dormant mode the femtocells depending on the users’ position, the network KPIs and the information provided by the MLB mechanism. In dormant mode, the femtocell is recovered in few seconds, Indoor mobility load balancing techniques 119 whereas it takes tens of seconds when it is switched off. This algorithm will be triggered once the MLB algorithm had finished. 5.5.3.3.1 System implementation The ES system is illustrated in Figure 5.16, which is mainly composed of two modules. The first module computes the reference distance based on the SCM. The second module introduces a FLC to evaluate the status of each femtocell: • Reference distance: This module calculates an indicator named reference distance for each user and cell, which is proportional to the distance from the users’ position to the studied cell. Figure 5.16: Energy-savings scheme. For that purpose, a process similar to that defined for NCM (see Figure 5.11) is applied in this method for each studied cell. In this case, instead of removing the studied cell, the femtocells of its proposed neighbor cell list (see equation (5.21)) are removed from the network. In consequence, a new layer is created with three different areas: 1) the original studied cell coverage area is defined as “blue area”, 2) the new extended coverage area of that cell is the “green area” and 3) the rest of the scenario is denoted as “red area” (see Figure 5.17). After that, a vector 𝐸𝑏𝑏𝑏𝑏 � � � � � � � �  is obtained with the reference distances of all users in the scenario for each studied cell. It is built as: 𝐸𝑏𝑏𝑏𝑏 � � � � � � � �  =�𝑒𝑏𝑏𝑏𝑏(𝑔1),𝑒𝑏𝑏𝑏𝑏(𝑔2)… 𝑒𝑏𝑏𝑏𝑏(𝑔𝑐)�, (5.31) where 𝑔𝑃 identifies the user, 𝐼 is the total number of active users in the scenario and 𝑒𝑏𝑏𝑏𝑏(𝑔𝑃) is described as follows: 126 Indoor mobility load balancing techniques (a) Classical coverage areas (b) Proposed cell areas (SCM) Figure 5.21: Estimated SCM. To ensure that all positions are always simultaneously covered by at least two femtocells (it would be a requirement of the cellular-indoor positioning system detailed in the field trial), the minimum value of transmission power variation is limited to -30 dB. The maximum increment is set to 10 dB as the femtocells maximum transmission power is restricted to 13 dBm (initially 3 dBm). 5.6.2 Simulation results Initially, the fuzzy-based MLB mechanisms are assessed. Then, the proposed context-aware SON methods are discussed. Indoor mobility load balancing techniques 127 5.6.2.1 Fuzzy-based MLB mechanisms Firstly, a sensitivity study is performed to analyze the performance of the algorithms with regard to the changes in the membership functions. Secondly, the global network and hotspot (most congested cell) simulation results are presented and described in a specific scenario: femtocells are limited to maximum 8 connected users and 1.4 MHz bandwidth. Thirdly, the study is extended to different femtocells’ capacity deployments (max. 4, 8, 16 and 32 users) and bandwidth (1.4, 3 and 5 MHz). 5.6.2.1.1 Sensitivity study Initially, five sets of membership functions with different interval of definition are adjusted by experts for each input indicator (𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓 and 𝑈𝑒𝑒𝛥𝑏𝑓𝑓𝑓). These intervals change in steps of 0.025 in the range of ±0.05 (keeping the shape of the membership function and the symmetry) according to the ones illustrated in Figure 5.4(b) and Figure 5.5(b). Additionally, another five sets of membership functions with different intervals, out of these recommendations, were evaluated. A sensitivity study is performed in the previous scenario where the maximum number of active connections is restricted to 4 users and the femtocell bandwidth is 1.4 MHz. Table 5.7 shows the performed sensitivity study where the UDR indicator has been evaluated. On the one hand, the analysis shows that the algorithm’s performance was not very sensitive to the set of membership functions proposed by experts (see the evaluation of Set 1 to Set 5). On the other hand, membership functions different to the recommendations of experts (Set 6 to Set 10) show (as expected) lower system performance. Therefore, it is assumed that membership functions that significantly differ from experts’ recommendations would probably decrease the algorithm’s performance. Table 5.7: Sensitivity analysis (UDR [%]). Max. 4 users/femtocell (1.4MHz) Set 1 Set 2 Set 3 Set 4 Set 5 Set 6 Set 7 Set 8 Set 9 Set 10 PLS 3.6458 3.6594 3.6487 3.6624 3.6548 5.157 5.2368 5.4861 5.1958 5.3682 PUS 1.2351 1.241 1.2115 1.2254 1.2388 4.1253 4.3672 4.6985 4.1687 4.8372 Finally, the set of membership functions with the lowest UDR (Set 3) is implemented (functions presented in Figure 5.4 and Figure 5.5). Note that, these membership functions are also implemented in other deployments where femtocells present different capacities (8, 16 and 32 users) and frequencies (3 MHz and 5 MHz). 128 Indoor mobility load balancing techniques 5.6.2.1.2 Assessment of the results: 8 users at 1.4 MHz The first analysis corresponds to the situation in which all femtocells limit is set to 8 active users and the network bandwidth is 1.4 MHz, i.e., it supports 6 PRBs. Both, the global network and the hotspot performance are assessed in these tests. Figure 5.22(a) illustrates the UDR indicator during one hour simulation. This indicator is collected every minute as vertical lines depict. This figure represents the network analysis when there is no optimization method implemented at this scenario (No OPT - blue stem) and the performance of the four MLB algorithms. The average value during the simulation is also shown for each method (horizontal lines). The non-optimized situation (blue stem with circles), which represents the absence of MLB methods, shows no congestion issues at the beginning and at the end of the evaluation, i.e., no users are dissatisfied. However, the network displays a high percentage of dissatisfied users during the overloaded period (from minute 30 to 50). Sometimes, the ratio of dissatisfied users reaches UDR=50%, i.e., half of the users that attempt to access the network in that minute are rejected or connected users are dropped. On average, this network indicator is 10% (horizontal blue line), which is higher than the values typically accepted by mobile operators. (a) Network performance (max. 8 users/femto and 1.4 MHz) (b) Hotspot performance (max. 8 users/femto and 1.4 MHz) Figure 5.22: UDR performance (1.4 MHz). Indoor mobility load balancing techniques 129 The PTS method (red stem with triangles) starts to reduce the blocked calls after detecting them, therefore, the UDR decreases as well. Nevertheless, due to the traffic fluctuations, peaks of UDR are obtained. Thanks to this mechanism, the average UDR is reduced to about 4%. However, as previously explained, this mechanism needs a large period of time to converge. The PLS mechanism (green stem with squares) shows improved results because it makes use of the occupied radio resources as input, decreasing the number and the value of dissatisfied users’ situations. The horizontal green line in Figure 5.22(a) shows an improvement over 7% from the non-optimized situation and 1% from the PTS method, getting an average value of UDR below 3%. Nevertheless, its 𝐿𝑜𝑚𝐿𝑏𝑓𝑓𝑓(𝑚𝑒𝑐𝑐) input does not reflect the most restricted factor for femtocells because it usually offers enough bandwidth to serve more users than the femtocell is able to support in connected mode. Thus, there are free radio resources and the algorithm is not triggered but the femtocell is totally congested (no more users are accepted). The PUS method (magenta stem with inverse triangles) reduces the users’ dissatisfaction. However, the performance of the algorithm decreases when the required user’s QoE is very high (huge amount of radio resources are demanded by the users). With this method there are only problems for an instant (minute 35) where few users are connected to the network but a lot of radio resources are required. Average UDR is lower than 1% (horizontal magenta line). The PLUS method (yellow stem with stars), as expected, provides the best results. In this case, the algorithm is aware of the amount of free space to access the femtocell both, in terms of active users and radio resources. The figure shows the same problem as with the PUS method for an instant (minute 35), where few users are connected to the network but a lot of radio resources are required. However, the UDR is lower because it also optimized the parameters according to the occupied radio resources. In average during one hour (horizontal yellow line), the value of UDR is around 0.2%, which complies with the usual operators’ requirements. UDR indicator at the most overloaded cell (hotspot) is shown in Figure 5.22(b). UDR evolves in the same way as in Figure 5.22(a), although it is higher in all cases. In average, around 30% of users are dissatisfied, while the optimization PTS, PLS, PUS and PLUS methods get around 18, 7, 2 and 1% of UDR respectively. 5.6.2.1.3 Assessment of the results: Other configurations The previous study has been extended to other femtocell deployments (different number of maximum connected users) and different bandwidth. The performance in 130 Indoor mobility load balancing techniques terms of the average value of UDR is depicted in Figure 5.23 for the different use cases. (a) 1.4 MHz (6 PRBs) (b) 3 MHz (15 PRBs) (c) 5 MHz (25 PRBs) Figure 5.23: Average UDR performance The minimum LTE bandwidth (1.4 MHz) is depicted in Figure 5.23(a). The UDR is reduced when the femtocell limitations in the maximum number of active users is increased. For the scenarios of maximum 4 and 16 active users per femtocell, the same explanation of previous subsection (8 active users at 1.4 MHz) is valid. Conversely, for the scenario of maximum 32 active users per femtocell, depending on the experienced 0 2 4 6 8 10 12 4816 32 UDR [%] Maximum users/femto NO OPT PTS PLS PUS PLUS 0 2 4 6 8 10 12 4 8 16 32 UDR [%] Maximum users/femto NO OPT PTS PLS PUS PLUS 0 2 4 6 8 10 12 4 8 16 32 UDR [%] Maximum users/femto NO OPT PTS PLS PUS PLUS Indoor mobility load balancing techniques 131 users’ quality and demand, sometimes the femtocell allows less than 32 simultaneously connected users because there are no available resources. Likewise, at another time, the femtocell allows 32 users because there are free radio resources. The reason is related to the RRM configuration. According to this, all the algorithms have comparable average UDR value except the PLUS method. Those situations overload femtocells adjacent to the hotspot and, many handover attempts fail (dropped connections appear) because those adjacent cells are already fully occupied (active users or radio resources). It means that the PUS method will not properly work all the time, the same for the PLS method. However, this issue is successfully fixed thanks to the PLUS method. Additionally, commercial femtocells could avoid this problem because most of them work with wider bandwidth (normally > 3 MHz). Similar simulations have been carried out with higher bandwidth (3 MHz), as Figure 5.23(b) depicts. This time, the maximum 32 users per femtocell deployment has enhanced UDR and, in particular, the PUS method presents much better performance compared to the 1.4 MHz case because the network available radio resources are higher. For the other use cases, the network performance is similar to the previous bandwidth (1.4 MHz) deployments. Finally, the network bandwidth was increased to 5 MHz. Figure 5.23(c) shows the overall network improvement achieved thanks to the PLUS method. It should be noticed that the higher the network bandwidth, the worse the PLS method works. This is due to the fact that the network bandwidth is increased but the occupied radio resources are the same (for most network services). An indicator which concerns operators is the HO signaling load (measured by the UHR). The higher the number of handovers the network manages, the more expenses the operator should afford. At Figure 5.24, results have shown that, in average, there is around one handover per user for all the methods. This is related to both pilot and data transmission powers are jointly tuned. Hence, changing and resizing femtocell areas do not increase signaling load, in consequence operator’s expenses are the same. To summarize, all methods have reduced users’ dissatisfaction while the network signaling load is held. At indoor networks where different bandwidth and types of femtocells are analyzed, both femtocell radio resources and maximum number of active users supported by the femtocell must be considered in MLB methods. Proposed PLS and PUS methods provide good results, but their performance depends on the predominant kind of traffic in the femtocells. The proposed PLUS method, however, led to the best network and hotspot performance for all situations with fast traffic fluctuations. 132 Indoor mobility load balancing techniques (a) 1.4 MHz (6 PRBs) (b) 3 MHz (15 PRBs) (c) 5 MHz (25 PRBs) Figure 5.24: Average UHR performance. 5.6.2.2 Context-aware load balancing mechanisms Firstly, a sensitivity study is performed to analyze the performance of the algorithms with regard to the changes in the defined thresholds. Secondly, the assessment of UD, VM and MR methods is detailed in four scenarios (maximum femtocell capacity: 4, 8, 16 and 32 active users at 5 MHz bandwidth). Thirdly, the study is extended to analyze the impact of users’ position error in the proposed context-aware mechanisms. Finally, the results of the coordination of the VM and MR methods and the ES method are discussed. 0.0 0.2 0.4 0.6 0.8 1.0 1.2 4 8 16 32 UHR Maximum users/femto NO OPT PTS PLS PUS PLUS 0.0 0.2 0.4 0.6 0.8 1.0 1.2 4 8 16 32 UHR Maximum users/femto NO OPT PTS PLS PUS PLUS 0.0 0.2 0.4 0.6 0.8 1.0 1.2 4 8 16 32 UHR Maximum users/femto NO OPT PTS PLS PUS PLUS Indoor mobility load balancing techniques 133 5.6.2.2.1 Sensitivity study The performance of the proposed methods would depend on the configuration of their thresholds. These thresholds could be defined based on: • The operators’ experience, policies or priorities in this field: expert engineers propose different configuration parameters based on their knowledge in this kind of networks. Then, these thresholds could be slightly modified to reach the optimal configuration. • A sensitivity study of the UDR indicator: the femtocell deployment is simulated with different thresholds to estimate the lowest values of UDR. This procedure is the one followed to get the thresholds of the proposed mechanisms. UD method Figure 5.25 shows the average values of UDR depending on 𝑆𝑎ℎ and 𝑀𝑎ℎ in steps of 10 and a deployment of femtocells with maximum 32 active users simultaneously at 5 MHz. The minimum value of UDR is obtained when the ratio of connected users, 𝑆𝑎ℎ, and the average ratio of occupied radio resources, 𝑀𝑎ℎ, is over 50 and 60, respectively. Under this configuration, the algorithm reduces the UDR to 1.6%. Note that the performance of the system is not very sensitive to the parameter selection (it would be around 2.5% of UDR) when the value of the selected parameters is below 100 (100 means the location-aware method is not triggered). In consequence, this system would not require a complex study to select the optimal parameters. Figure 5.25: UD method - Sensitivity study. 134 Indoor mobility load balancing techniques VM method Figure 5.26 shows the average values of UDR depending on 𝑆𝑎ℎ and 𝑇𝑎ℎ in steps of 10 and a deployment of femtocells with maximum 32 active users simultaneously at 5 MHz. The minimum value of UDR is obtained when the ratio of femtocell connected users, 𝑆𝑎ℎ, is over 70 and the average ratio of neighbors femtocell connected users, 𝑇𝑎ℎ, is below 60. Under this configuration, the algorithm reduces the UDR to 1.1%. Similar to UD sensitivity study, this system would not require a complex study to select the optimal parameters as the performance of the system is not very sensitive to the parameter selection (it would be around 2% of UDR) when the value of the selected parameters is below 100 (100 means the location-aware method is not triggered). Figure 5.26: VM method - Sensitivity study. MR method Figure 5.27 shows the average values of UDR depending on 𝑆𝑎ℎ and 𝑇𝑎ℎ in steps of 10 and a deployment of femtocells with maximum 32 active users simultaneously at 5 MHz. The minimum value of UDR is obtained when the ratio of femtocell connected users, 𝑆𝑎ℎ, is over 50 and the average ratio of neighbors femtocell connected users, 𝑇𝑎ℎ, is below 40. Under this configuration, the algorithm reduces the UDR to 1.8%. Similar to previous methods, this system would not require a complex study to select the optimal parameters as the performance of the system is not very sensitive to the parameter selection (it would be around 2.5% of UDR) when the value of the selected parameters is below 100 (100 means the method is not triggered). Indoor mobility load balancing techniques 135 Figure 5.27: MR method - Sensitivity study. 5.6.2.2.2 Assessment of the results The proposed algorithms are compared to the non-optimized network (baseline) and the PLUS algorithm described in subsection “5.4.1.4”. The challenge faced is threefold. Firstly, the UDR must be as low as possible. Secondly, a good QoE is expected to support the communication. Finally, the number of HOs should not be highly increased to keep the signaling data. According to the sensitivity study of the UDR indicator carried out for each method in the airport scenario, the optimal values for theses thresholds are: 𝑆𝑎ℎ=50 and 𝑀𝑎ℎ=60 for UD method, 𝑆𝑎ℎ=70 and 𝑇𝑎ℎ=60 for VM method and 𝑆𝑎ℎ=50 and 𝑇𝑎ℎ=40 for MR method. In order to accomplish an extensive and complete study, the evaluation of the algorithms was performed in four different indoor deployments depending on the femtocell capacity (maximum 4, 8, 16 and 32 active users). The number of users in the scenario has been established according to the analyzed cell capacity, e.g., for a femtocell capacity limit of 4 active users, 500 users per hour are simulated, whereas for a femtocell capacity limit of 8, a population of 1000 users per hour is defined. As aforementioned, femtocells need a period to set the configuration file (i.e., to change the transmission power) which normally takes tens of seconds. For this reason, the algorithms are triggered every minute while hotspots are created every hour at different places (see Figure 5.29).