Detection and compensation methods for self-healing in self-organizing networks

Universidad de Málaga Escuela Técnica Superior de Ingeniería de Telecomunicación Programa de Doctorado en Ingeniería de Telecomunicación TESIS DOCTORAL Detection and Compensation Methods for Self-Healing in Self-Organizing Networks Autora: Isabel de la Bandera Cascales Directores: Raquel Barco Moreno Matías Toril Genovés 2017 AUTOR: Isabel de la Bandera Cascales http://orcid.org/0000-0003-4228-3494 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 Programa de Doctorado en Ingeniería de Telecomunicación Universidad de Málaga Por la presente, Dra. Dª. Raquel Barco Moreno y Dr. D. Matías Toril Genovés, profesores doctores del Departamento de Ingeniería de Comunicaciones de la Universidad de Málaga, CERTIFICAN: Que Dª. Isabel de la Bandera Cascales, Ingeniera de Telecomunicación, ha realizado en el Departamento de Ingeniería de Comunicaciones de la Universidad de Málaga bajo su dirección, el trabajo de investigación correspondiente a su TESIS DOCTORAL titulada: “Detection and Compensation Methods for Self-Healing in Self-Organizing Networks” En dicho trabajo se han propuesto aportaciones originales para la gestión de problemas en redes móviles. En particular, se han propuesto métodos para la detección y compensación de diferentes fallos. Además, se ha desarrollado una herramienta de simulación de la tecnología celular LTE para la evaluación de las técnicas propuestas. Los resultados expuestos han dado lugar a publicaciones en revistas, patentes y aportaciones a congresos. Por todo ello, consideran que esta Tesis es apta para su presentación al Tribunal que ha de juzgarla. Y para que conste a efectos de lo establecido, AUTORIZAN la presentación de esta Tesis en la Universidad de Málaga. En Málaga, a de de . Fdo: Raquel Barco Moreno, Matías Toril Genovés UNIVERSIDAD DE MÁLAGA ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA DE TELECOMUNICACIÓN Reunido el tribunal examinador en el día de la fecha, constituido por: Presidente: Dr. D. Secretario: Dr. D. Vocal: Dr. D. para juzgar la Tesis Doctoral titulada Detection and Compensation Methods for Self-Healing in Self-Organizing Networks realizada por Da. Isabel de la Bandera Cascales y dirigida por la Dra. Da. Raquel Barco Moreno y el Dr. D. Matías Toril Genovés, acordó por otorgar la calificación de y para que conste, se extiende firmada por los componentes del tribunal la presente diligencia. Málaga a de de El Presidente: El Secretario: El Vocal: Fdo.: Fdo.: Fdo.: A mis niños. Acknowledgements At last this day has come and I am close to finishing this stage. It has been a large and intense journey which would have been very different without all the people that I have met during these years. Firstly, I would like to express my most sincere gratitude to my supervisors, Raquel and Matías. Many thanks to Raquel for giving me the opportunity to become part of this exceptional team, for being available at all times for any problem that I had, and for her support and affection during all these years. Many thanks to Matías for his guidance, for all the hours spent with the simulator and for his knowledge and encouragement that have helped me in each step. I would also like to thank Salva, who introduced me to this amazing world of mobile communications for the first time. I wish to thank all my office mates that have been with me during all these years. There is no doubt that with their support and sense of humor this work has been more pleasant. Special mention to Emil, David and Ana for all the discussions and good advice and for all the shared laughs. I would like to express my special thanks to Pablo. He is the only one who has been with me since the beginning and the main support. Thanks for his patience, for helping me make every decision, for sharing with me all his knowledge and for assisting me in writing papers. I know that this work is also yours. I must also acknowledge the financial support given by the projects mentioned below, together with the research group TIC-102 Ingeniería de Comunicaciones and Universidad de Málaga. They made this work possible and allowed me to present results and exchange knowledge and skills in conferences and journals. I would like to strongly thank my family for their love and support. Thanks to my parents for their motivation and trust during these years and all my life. I know that I would not have achieved many of my successes in life without them. Thanks to Cristina, my favorite sister. Her opinion is always the most important even if it is usually the opposite to mine. There is no doubt that I will always ask you for advice. And thanks to Javier for wanting to be part of this family. I could not finish these lines without thanking my beautiful family. These years of intense work have also brought me the most important thing of all my life, my beautiful children. All of this work and everything else is for them. Thanks to my husband for his support, his motivation, 7 on analytical and simulated models. However, the main challenges are now linked to devising SON solutions which are based on data from live networks. This is particularly interesting in the area of Self-Healing. In this context, the availability of historical data is crucial to understand how the network behaves under normal and faulty conditions or how the faults impact on the Quality-of-Experience. The main goal of this thesis is to develop effective detection and compensation algorithms for mobile networks. Firstly, a Cell Outage Detection (COD) algorithm based on handover statistics is proposed. In spite of its simplicity, the proposed algorithm allows to detect cell outages in any network immediately after collecting network performance indicators. Secondly, a novel Cell Outage Compensation (COC) methodology is proposed. The conceived method adapts the compensation action to the specific degradation produced by the cell outage. Once the cell outage is detected, an analysis is carried out to find out the degradation caused in the neighboring cells. Afterwards, different COC algorithms can be applied to each neighboring cell depending on the kind of degradation. Thirdly, state-of-the-art proposes compensation solutions for cell outages. Conversely, in this thesis a cell problem different to an outage is analyzed: a weak coverage problem. Specifically, a Cell Degradation Compensation (CDC) algorithm based on handover margin modifications is proposed. Many of the methods proposed in this thesis have been designed and validate based on real networks data. Unfortunately, although it is interesting to evaluate the proposed methods in live networks, this is not always possible. Operators are sometimes unwilling to test algorithms in live networks, especially when test needs changing configuration parameters. For this reason, part of this thesis is dedicated to the implementation of a dynamic system-level simulator that allows the evaluation of the proposed algorithms. vi Resumen Uno de los elementos clave en la definición de los recientes estándares de comunicaciones móviles del 3rd Generation Partnership Project (3GPP), LTE (Long Term Evolution) y LTEAdvanced, es la consideración de funciones que se puedan ejecutar de manera automática. Este tipo de redes se conocen como redes Auto-Organizadas (Self-Organizing Networks, SON). Las funciones SON permiten hacer frente al importante incremento en tamaño y complejidad que han experimentado las redes de comunicaciones móviles en los últimos años. El número de usuarios es cada vez mayor y los servicios requieren gran cantidad de recursos y altas tasas de transmisión por lo que la gestión de estas redes se está convirtiendo en una tarea cada vez más compleja. Además, cuando las redes de quinta generación (5G) se implanten, la complejidad y el coste asociado a estas nuevas redes será todavía mayor. En este contexto, las funciones SON resultan imprescindibles para llevar a cabo la gestión de estas redes tan complejas. El objetivo de SON es definir un conjunto de funcionalidades que permitan automatizar la gestión de las redes móviles. Mediante la automatización de las tareas de gestión y optimización es posible reducir los gastos de operación y capital (OPEX y CAPEX). Las funciones SON se clasifican en tres grupos: AutoConfiguración, Auto-Optimización y Auto-Curación. Las funciones de Auto-Configuración tienen como objetivo la definición de los distintos parámetros de configuración durante la fase de planificación de una red o después de la introducción de un nuevo elemento en una red ya desplegada. Las funciones de Auto-Optimización pretenden modificar los parámetros de configuración de una red para maximizar el rendimiento de la misma y adaptarse a distintos escenarios. Las funciones de AutoCuración tienen como objetivo detectar y diagnosticar posibles fallos en la red que afecten al funcionamiento de la misma de manera automática. Cuando un fallo es detectado en una celda este puede ser recuperado (función de recuperación) o compensado (función de compensación). Uno de los principales desafíos relacionado con las funciones SON es el desarrollo de métodos eficientes para la automatización de las tareas de optimización y mantenimiento de una red móvil. En este sentido, la comunidad científica ha centrado su interés en la definición de métodos de Auto-Configuración y Auto-Optimización siendo las funciones de Auto-Curación las menos exploradas. Por esta razón, no es fácil encontrar algoritmos de detección y compensación realmente eficientes. Muchos estudios presentan métodos de detección y compensación que producen buenos resultados pero a costa de una gran complejidad. Además, en muchos casos, los algoritmos de detección y compensación se presentan como solución general para distintos tipos vii de fallo lo que hace que disminuya la efectividad. Por otro lado, la investigación ha estado tradicionalmente enfocada a la búsqueda de soluciones SON basadas en modelos analíticos o simulados. Sin embargo, el principal desafío ahora está relacionado con la explotación de datos reales disponibles con el objetivo de crear una base del conocimiento útil que maximice el funcionamiento de las actuales soluciones SON. Esto es especialmente interesante en el área de las funciones de Auto-Curación. En este contexto, la disponibilidad de un histórico de datos es crucial para entender cómo funciona la red en condiciones normales o cuando se producen fallos y como estos fallos afectan a la calidad de servicio experimentada por los usuarios. El principal objetivo de esta tesis es el desarrollo de algoritmos eficientes de detección y compensación de fallos en redes móviles. En primer lugar, se propone un método de detección de celdas caídas basado en estadísticas de traspasos. Una de las principales características de este algoritmo es que su simplicidad permite detectar celdas caídas en cualquier red inmediatamente después de acceder a los indicadores de funcionamiento de la misma. En segundo lugar, una parte importante de la tesis está centrada en la función de compensación. Por un lado, se propone una novedosa metodología de compensación de celdas caídas. Este nuevo método permite adaptar la compensación a la degradación específica provocada por la celda caída. Una vez que se detecta un problema de celda caída, se realiza un análisis de la degradación producida por este fallo en las celdas vecinas. A continuación, diferentes algoritmos de compensación se aplican a las distintas celdas vecinas en función del tipo de degradación detectado. En esta tesis se ha llevado a cabo un estudio de esta fase de análisis utilizando datos de una red real actualmente en uso. Por otro lado, en esta tesis también se propone un método de compensación que considera un fallo diferente al de celda caída. En concreto, se propone un método de compensación para un fallo de cobertura débil basado en modificaciones del margen de traspaso. Por último, aunque es interesante evaluar los métodos propuestos en redes reales, no siempre es posible. Los operadores suelen ser reacios a probar métodos que impliquen cambios en los parámetros de configuración de los elementos de la red. Por esta razón, una parte de esta tesis ha estado centrada en la implementación de un simulador dinámico de nivel de sistema que permita la evaluación de los métodos propuestos. viii Acronyms 3G 3rd Generation 3GPP 3rd Generation Partnership Project 5G 5th Generation AM Acknowledge Mode AMC Adaptive Modulation and Coding ARQ Automatic Repeat Request AS Access Stratum AWGN Additive White Gaussian Noise BC Best Channel BLER Block Error Rate CAPEX Capital Expenditures CBR Call Blocking Ratio CCO Capacity and Coverage Optimization CDC Cell Degradation Compensation CDD Cell Degradation Detection CDMA Code-Division Multiple Access ix CDR Call Dropping Ratio CM Configuration Management Parameters CMC Connection Mobility Control COC Cell Outage Compensation COD Cell Outage Detection COM Cell Outage Management CPU Computer Processing Unit CQI Channel Quality Indicator eNB evolved Node B EPA Extended Pedestrian A EPC Evolved Packet Core EPS Evolved Packet System ETU Extended Typical Urban E-UTRAN Evolved Universal Terrestrial Radio Access Network EVA Extended Vehicular A FLC Fuzzy Logic Controller GSM Global System for Mobile communications GPRS General Packet Radio Service HARQ Hybrid Automatic Repeat Request HO Handover HOH Handover Hysteresis HOM Handover Margin HOSR Handover Success Rate x HPR HO ping-pong Ratio inHO incoming HO KPI Key Performance Indicators LDF Large Delay First LTE Long-Term Evolution MAC Medium Access Control MBMS Multimedia Broadcast Multicast Service MCS Modulation and Coding Scheme MIMO Multiple-Input Multiple-Output MME Mobility Management Entity NAS Non-Access Stratum NGMN Next Generation Mobile Networks OFDM Orthogonal Frequency Division Multiplexing OFDMA Orthogonal Frequency Division Multiplex Access OPEX Operational Expenditures OSS Operations Support System PDCP Packet Data Convergence Control PDSCH Physical Downlink Shared Channel PDU Packet Data Unit PF Proportional Fair PM Performance Management Parameter PRB Physical Resource Block P-GW Packet Data Network Gateway xi QAM Quadrature Amplitude Modulation QoS Quality-of-Service QPSK Quadrature Phase Shift Keying RAC Radio Admission Control RACH Random Access Channel RAT Ratio Access Technology RBC Radio Bearer Control RE Resource Element RLC Radio Link Control RR Round Robin RRC Radio Resource Control RRM Radio Resource Management RS Reference Signal RSRP Reference Signal Receive Power RSRQ Reference Signal Receive Quality RSSI Received Signal Strength Indicator SC-FDMA Single Carrier Frequency Division Multiple Access SDU Service Data Unit S-GW Serving Gateway SINR Signal to Interference plus Noise Ratio SISO Single Input Single Output SON Self-Organizing Networks TB Transport Block xii TTI Transmission Time Interval TTT Time-To-Trigger UE User Equipment UMTS Universal Mobile Telecommunications System US Uncorrelated Scattering VoIP Voice over IP WLAN Wireless Local Area Network WSS Wide-Sense Stationary WSSUS Wide-Sense Stationary Uncorrelated Scattering xiii xiv Chapter 1 Introduction This first chapter introduces the main topics of this thesis, presenting the motivation, the objectives and the document structure. 1.1 Motivation During the last decades, the mobile communications industry has experienced a significant increase, leading to a constant evolution of the related technologies. At the same time, the number of users demanding mobile services has experienced an exponential increase, as 95% of the global population live in an area that is covered by a mobile cellular network [1]. Thus, the mobile terminals (e.g., smartphones) have become an essential element for most people, which use them in different areas of life, not only as a basic instrument in their jobs, but also for fun. This fact has lead to a growing complexity and variety of the demanded services. To cope with the large number of users and the complexity of existing services, the developed technologies have had to evolve quickly. The first digital networks (i.e., Global System for Mobile communications, GSM) provided a voice service of high quality and led to the first important expansion of the mobile networks. However, users were soon demanding other types of services and these networks resulted insufficient to satisfy the growing demand. Thus, the 3rd generation (3G) of mobile communications (e.g., Universal Mobile Telecommunications System, UMTS) appeared. This standard, developed by the 3G Partnership Project (3GPP) group, offered data services in addition to the voice service. The 3G networks continued evolving in order to improve network performance and to offer more complex services. Currently, a new standard (i.e., Long Term Evolution, LTE) is being deployed to cope with the enormous demand of services. The key benefits of LTE can be summarized in improved system performance, higher data rates and spectral efficiency, reduced latency and power consumption, enhanced flexibility of spectral usage and simplified network architecture [2]. In such a scenario, network operators have two main objectives: to satisfy the large traffic 1 CHAPTER 2. TECHNICAL BACKGROUND Figure 2.1: LTE overall architecture [2]. –Access stratum (AS) security control –Idle state mobility handling –EPS bearer control •Serving Gateway (S-GW), which is the user plane gateway to the E-UTRAN. The S-GW provides these functions: –Mobility anchor point for mobility between LTE cells or cells from other 3GPP technologies –Termination of user plane packets for paging reasons –Packet forwarding, routing, and buffering of downlink data for UEs that are in LTEIDLE state •Packet Data Network Gateway (P-GW), which is the user plane gateway to the packet data network. The P-GW is responsible for policy enforcement, charging support, and user’s IP address allocation. It also serves as a global mobility anchor for mobility between 3GPP and non-3GPP access networks. The P-GW is the edge router between EPS and external packet data networks. The E-UTRAN consists of one node only, the eNB (eNodeB), which is the interface to the UE. eNBs are connected to neighboring eNBs through the X2 interface and with the core network elements through the S1 interface. The eNB hosts these functions: •Radio resource management (RRM) •IP header compression and encryption •Selection of MME at UE attachment •Routing of user plane data towards S-GW 8 2.1. OVERVIEW OF THE LTE SYSTEM •Scheduling and transmission of paging messages and broadcast information •Measurement and measurement reporting configuration for mobility and scheduling Fig. 2.2 summarizes the main functions included in each element of the LTE system. Among these functions, one of the most important function is the RRM. The purpose of RRM is to ensure the efficient use of the available radio resources in order to provide large variety of services to a large number of users, whilst also ensuring Quality of Service (QoS) requirements. Most RRM functions are included in the link and network layer. Figure 2.2: Functional Split between E-UTRAN and EPC [2]. 2.1.2 Physical layer Downlink and uplink transmission in LTE are based on the use of multiple access technologies, namely Orthogonal Frequency Division Multiple Access (OFDMA) for the downlink, and SingleCarrier Frequency Division Multiple Access (SC-FDMA) for the uplink. OFDMA is a variant of Orthogonal Frequency Division Multiplexing (OFDM). OFDM makes use of a large number of closely spaced orthogonal subcarriers that are transmitted in parallel. Each subcarrier is modulated with a conventional modulation scheme (such as QPSK, 16QAM, or 64QAM) at a low symbol rate. The combination of thousands of subcarriers enables data rates similar to conventional single-carrier modulation schemes in the same bandwidth. This technique creates high peak-to-average signals, which is the major drawback in the uplink. Alternatively, SC-FDMA is the transmission scheme chosen for the LTE uplink. This technique combines the low peak-to-average ratio of single-carrier transmission systems, such as GSM and Code-Division Multiple Access (CDMA), with the multi-path protection and flexible frequency allocation of 9 CHAPTER 2. TECHNICAL BACKGROUND OFDMA. In LTE, a Resource Element (RE) is the smallest unit in the physical layer, consisting of one OFDM or SC-FDMA symbol in the time domain and one subcarrier in the frequency domain. These REs are grouped in a Physical Resource Block (PRB) that is the smallest unit that can be scheduled for transmission. A PRB physically occupies 0.5 ms (1 slot) in the time domain and 180 kHz in the frequency domain. The number of subcarriers per RB and the number of symbols per RB vary as a function of the cyclic prefix length and subcarrier spacing. In particular, a PRB comprises 12 subcarriers at a 15 kHz spacing. In addition to the user data, reference signals and other control data are included in each PRB. Another important feature of LTE is the use of multiple antenna techniques, which are used to increase network coverage and capacity. Such a technique is known as Multiple-Input, Multiple-Output (MIMO) technique. MIMO increases spectral capacity by transmitting multiple data streams simultaneously in the same frequency and time, taking full advantage of the different paths in the radio channel. 2.1.3 Link layer The link layer is divided into three sublayers: Medium Access Control (MAC), Radio Link Control (RLC) and Packet Data Convergence Protocol (PDCP). The main services and functions of the MAC sublayer include: •Mapping between logical channels and transport channels •Multiplexing/demultiplexing of MAC Service Data Units (SDUs) belonging to one or different logical channels into/from Transport Blocks (TBs) delivered to/from the physical layer on transport channels •Scheduling information reporting •Error correction through Hybrid Automatic Repeat reQuests (HARQ) •Priority handling between logical channels of one UE •Priority handling between UEs by means of dynamic scheduling •Multimedia Broadcast Multicast Service (MBMS) identification •Transport format selection •Padding The RLC sublayer includes the following functions: •Transfer of upper layer Packet Data Units (PDUs) •Error Correction through Automatic Repeat reQuest (ARQ) •Concatenation, segmentation and reassembly of RLC SDUs 10 2.1. OVERVIEW OF THE LTE SYSTEM •Re-segmentation of RLC data PDUs •Reordering of RLC data PDUs •Duplicate detection •Protocol error detection •RLC SDU discard •RLC re-establishment Finally, the main services and functions of the PDCP sublayer include: •Header compression and decompression •Transfer of user data •In-sequence delivery of upper layer PDUs at PDCP re-establishment procedure for RLC Acknowledge Mode (AM) •Duplicate detection of lower layer SDUs at PDCP re-establishment procedure for RLC AM •Retransmission of PDCP SDUs at handover (HO) for RLC AM •Ciphering and deciphering •Timer-based SDU discard in uplink •Ciphering and Integrity Protection (control plane) •Transfer of control plane data (control plane) As for the RRM, the link layer includes three main functions: HARQ, Adaptive Modulation and Coding (AMC) and user scheduling. HARQ is a technique for ensuring that data is sent reliably from one network node to another, identifying when transmission errors occur and facilitating retransmission from the source. AMC is the mechanism used for link adaptation to improve data throughput in a fading channel. This technique changes the downlink Modulation and Coding Scheme (MCS) based on the channel conditions of each user. When the link quality is good, the system can use a higher order modulation scheme (more bits per symbol) or less channel coding, which results in higher data rates. When link conditions are poor because of problems such as signal fading or interference, the system can use a lower modulation depth or stronger channel coding to maintain acceptable margin in the radio link budget. The scheduler is part of the MAC layer and controls the assignment of uplink and downlink resources. Uplink and downlink scheduling are separated in LTE and uplink and downlink scheduling decisions can be taken independently of each other. The basic principle for the downlink scheduler is to dynamically determine, in each 1 ms interval, which terminal transmits or receives data and on which resources. 11 CHAPTER 2. TECHNICAL BACKGROUND 2.1.4 Network layer The main services and functions of the Radio Resource Control (RRC) sublayer include: •Broadcast of System Information related to the AS and the NAS •Paging •Establishment, maintenance and release of an RRC connection between the UE and EUTRAN •Security functions including key management •Establishment, configuration, maintenance and release of point to point radio bearers •Mobility functions: cell selection and reselection, HO •Establishment, configuration, maintenance and release of radio bearers for MBMS services •QoS management functions •UE measurement reporting and control of the reporting •NAS direct message transfer to/from NAS from/to UE The RRC layer includes most of the main functions for the radio resource management. The most significant are the following: •Radio Bearer Control (RBC). This function is in charge of the establishment, maintenance and release of radio bearers. When setting up a radio bearer for a service, RBC takes into account the overall resource situation in E-UTRAN, the QoS requirements of in-progress sessions and the QoS requirement for the new service. RBC is also concerned with the maintenance of radio bearers of in-progress sessions at the change of the radio resource situation due to mobility or other reasons. RBC is involved in the release of radio resources associated with radio bearers at session termination, HO or at other occasions. •Radio Admission Control (RAC). The task of RAC is to admit or reject the establishment requests for new radio bearers. For this purpose, RAC takes into account the overall resource situation in E-UTRAN, the QoS requirements, the priority levels and the provided QoS of in-progress sessions and the QoS requirement of the new radio bearer request. The goal of RAC is to ensure high radio resource utilization (by accepting radio bearer requests as long as radio resources available) while ensuring proper QoS for in-progress sessions (by rejecting radio bearer requests when they cannot be accommodated). •Connection Mobility Control (CMC). CMC is concerned with the management of radio resources in connection with idle or connected mode mobility. In idle mode, the cell reselection algorithms are controlled by setting parameters (thresholds and hysteresis values) that define the best cell and/or determine when the UE should select a new cell. Likewise, E-UTRAN broadcasts the values of parameters configuring UE measurement and reporting procedures. In connected mode, the mobility of radio connections has to be supported. HO 12 2.2. SELF-ORGANIZING NETWORKS decisions may be based on UE and eNB measurements. In addition, HO decisions may take other inputs into account, such as neighboring cell load, traffic distribution, transport and hardware resources and operator defined policies. In particular, the physical layer measurements considered in the mobility decisions are: –Reference Signal Receive Power (RSRP). This measurement is the most basic of the UE physical layer measurements, consisting of the linear average (in watts) of the downlink Reference Signals (RS) across the channel bandwidth. Since the RS exist only for one symbol at a time, the RSRP measurement is made only on those REs that contain cell-specific RS. –Reference Signal Receive Quality (RSRQ). This measurement provides an indication of signal quality and is defined as the ratio of RSRP to the E-UTRA carrier Received Signal Strength Indicator (RSSI). RSSI represents the entire received power, including the desired power from the serving cell as well as cochannel interference and other sources of noise. Measuring RSRQ becomes particularly important near cell edge when an HO is performed to the next cell. 2.2 Self-Organizing Networks Mobile communication networks are increasing in size and complexity significantly. Thus, the automation of network management tasks has become a key element in the definition of new generation mobile networks. The addition of automation features is carried out by implementing SON functions. The main objective of SON is to reduce OPEX and CAPEX, while providing services with a certain quality requirements. The main benefits of introducing SON functions in cellular networks are [50]: •To reduce installation time and costs •To reduce OPEX by reducing manual efforts when monitoring, optimizing, diagnosing, and healing the network •To reduce CAPEX due to better use of network infrastructure and spectrum resources •To improved network performance •To improved user experience SON functions can be implemented in different network elements. Depending on the selected network elements, there are three main alternatives: centralized, distributed and hybrid [51]. In a centralized SON architecture, the algorithms are executed at the network management level. The main benefit of this approach is that the SON algorithms can take information from all parts of the network into consideration. In the case of a distributed SON architecture, SON algorithms are implemented in the network nodes. Thus, the computational load of executing algorithms is distributed across network nodes. A major drawback is the fact that the nodes need to exchange information directly with each other. Finally, the hybrid SON alternative implements part of the SON algorithm in the network management system and another part in the network elements. 13 CHAPTER 2. TECHNICAL BACKGROUND Most SON functions share an initial phase of collecting network data that are used to make decisions. Network information can be of different types depending on the source [10]: •Configuration management parameters (CM). This information consists of the current configuration settings of network elements. •Performance management parameters (PM). This information consists of counters reflecting the performance of network elements, which are reported periodically and aggregated at different levels. They can be related to traffic load, resource availability, etc. •Alarms. These are messages generated by the network elements when a failure occurs. •Mobile connection traces. This data consists of very detailed information of events generated by specific UEs, which can be collected from most network elements. •Real time monitoring. Some network elements provide online measurements, such as traffic load or HOs. •Drive tests. Field measurements performed in a specific area with specialized equipment. •KPIs. These indicators are calculated by combining other counters or measurements to provide a meaningful performance measure. They are the measurements more frequently used as input of SON functions. •Context information. This is information related to the environment, such as type of area (e.g., urban area) or typical weather in the cell (e.g., rainy). According to the 3GPP definitions [8, 9], the SON functions can be grouped into three categories, shown in Fig. 2.3: •Self-Configuration. The objective of this functionality is to reduce the amount of human intervention in the overall installation process by providing "plug and play" functionality in network elements, such as the eNBs. Some tasks of Self-Configuration systems are: the detection of the transport link and establishment of a connection with the core network elements, to download and upgrade to the latest software version, to set up the initial configuration parameters including neighbor relations, to perform a self-test and to set the node to operational mode. The main use cases defined in Self-Configuration are: –Automated configuration of physical cell identity. This SON use case provides an automated configuration of a newly introduced cell physical identity. When a new eNB is brought into the field, a physical cell identity needs to be selected for each of its supported cells, avoiding collision with respective neighboring cells. –Automatic Neighbor Relation function. This use case allows an eNB to build and maintain neighbor relationships. This function is also important when the network is expanding and even in mature networks. •Self-Optimization. Self-Optimization functions aim at maintaining adequate service quality and network performance after changes in network operating conditions with a minimum of manual intervention from the operator. These functions monitor and analyze performance 14 2.2. SELF-ORGANIZING NETWORKS Self-Configuration -Detectionoftransportlink -EstablishmentofaconnectionwithEPC -Downloadandupgradesoftware -Set-upinitialconfiguration -Self-test - AutomatedConfigurationofPhysicalCellIdentity - AutomaticNeighborRelationfunction Self-Optimization -Coverageandcapacityoptimization -EnergySavings -InterferenceReduction -Mobilityrobustnessoptimization -MobilityLoadbalancingoptimization -RACHOptimization -Inter-cellInterferenceCoordination Self-Healing -Informationcollection -Faultdetection -Diagnosis -Faultrecovery -Faultcompensation -CellOutageCompensation -CellOutageDetection Figure 2.3: SON functions and use cases. data and automatically trigger optimization actions when necessary. The main use cases in Self-Optimization are: –Coverage and capacity optimization. A typical operational task is to optimize the network based on coverage and capacity criteria. Coverage optimization has usually a higher priority than capacity optimization. To provide an optimal coverage, users should establish and maintain connections with acceptable or default service quality, according to operator’s requirements. It implies that coverage is continuous and users are unaware of cell borders. The coverage must therefore be provided in both, idle and active mode for both, uplink and downlink. Coverage optimization algorithms must take the impact on capacity into account. Since coverage and capacity are linked, a trade-off between the two goals may also be a subject of optimization. –Energy savings. This use case is based on switching off a cell when its capacity is no longer needed and re-activate it on a need basis in order to reduce operational expenses. –Interference reduction. The objective of this use case is to improve the capacity 15 CHAPTER 2. TECHNICAL BACKGROUND through interference reduction by switching off those cells that are not needed at some point of time (e.g., home eNBs when the user is not at home). –Mobility robustness optimization. The objective of this use case is to automatically modify the HO configuration parameters in order to reduce the degradation of the service performance caused by a non-optimal configuration of HO parameters. –Mobility load balancing optimization. This use case is based on the optimization of cell reselection and/or HO parameters to cope with the unequal traffic load distribution and to minimize the number of HOs and redirections needed to achieve the load balancing. –Random Access Channel (RACH) optimization. The objective of this use case is to optimize the RACH configuration, which has a significant impact on the system performance, such as call setup delays, call setup rate, HO success rate, HO delays, etc. –Inter-cell interference coordination. The objective of this use case is to reduce or avoid the interference between PRBs in uplink and downlink by a coordinated usage of available radio resources in adjacent cells. This coordination is carried out by prioritizing users in the different cells. The main benefit is an improved signal quality and higher user data throughput. •Self-Healing. Self-Healing functions aim to mitigate as far as possible the performance degradation caused by a problem in a network. Firstly, performance failures that can affect the network should be detected and diagnosed without human intervention. Then, the failure should be recovered or compensated. According to [10], the main functions in Self-Healing are the following: –Information collection. As described before, the first phase of a SON algorithm should be to collect information from the network in order to analyze the current status. The more complete the information is, the faster the failures are identified and solved. –Fault detection. The objective of this function is to identify cells with problems. Detection includes identifying cells with service outage (COD) and also cells with service degradation (Cell Degradation Detection, CDD). –Diagnosis. This function consists of identifying the fault cause of the problem and then determining the repair actions to eliminate that cause. –Fault recovery. This function carries out the execution of the identified repair actions to solve the detected failure. –Fault compensation. This function tries to minimize the degradation caused by the fault until it is definitely solved. The compensation can be applied when the fault is an outage (COC) or when a cell is degraded (CDC). This function acts in parallel with the diagnosis function. COD and COC are two of the main use cases in Self-Healing defined by the 3GPP. These two use cases are addressed in this thesis. In a global context where different failures may occur 16 2.2. SELF-ORGANIZING NETWORKS in a network, the detection function may be one of the easiest function in Self-Healing. In many cases, the problematic cell can easily be identified by some specific alarms and KPIs. However, there are certain faulty situations (e.g., cell outages) that may be harder to detect, as they do not generate any alarms [10]. This thesis deals with the latter case, proposing a new COD algorithm. Likewise, a novel COC methodology and a novel CDC algorithm are proposed. 17 CHAPTER 3. SIMULATION TOOL Lastly, an antenna gain is also defined. Fig. 3.5 shows the horizontal and vertical pattern calculated as discussed before. Figure 3.5: Horizontal and vertical antenna pattern. Alternatively, a real horizontal and vertical antenna pattern can be included in the simulator instead of the aforementioned patterns. Fig. 3.6 shows an example of real antenna pattern. Figure 3.6: Realistic horizontal and vertical antenna pattern. 3.2.3 Spatial traffic distribution Users can be spatially distributed in a uniform or non-uniform way over the scenario. In the uniform case, users are located in any point of the scenario with the same probability. However, to reproduce a realistic situation, it is recommended to use a non-uniform distribution. The typical spatial distribution in urban areas can be described by a log-normal distribution at a cell level [68]. In this work, the traffic is created following this distribution with a central hotspot as in [69]. Fig. 3.7 shows the probability of starting a call in any location of the scenario 24 3.2. SIMULATOR GENERAL STRUCTURE when an uniform scenario is used. It is observed that the spatial traffic distribution has a central peak, creating a congested area with higher traffic density in the center of the scenario. Users can be static or not. In the latter case, spatial traffic distribution is slightly affected by the mobility model, as explained in the next section. Figure 3.7: Spatial traffic distribution. 3.2.4 Mobility model The proposed mobility model does not impose any constraint on user direction, as users can freely move over the scenario. More precisely, the mobility model considers random constant paths for users in the simulation scenario. Users move at constant speed, set to 3, 10 or 50 km/h. This model also includes the effect of the wrap-around technique when the uniform scenario is used, which means that when a user reaches the limit of the original scenario, it appears in the correct position of the scenario. In the case that the realistic scenario is used, a random direction is calculated and assigned to each new user. Along the iterations, users move with the assigned direction and the configured speed. The mobility model takes into account scenario boundaries. 3.2.5 Traffic model The simulator includes two types of service: a voice service and a packet data service. The voice service modeled in the simulator is VoIP. This service is defined as a source generating packets of 40 bytes every 20 ms [70], reaching a bit rate of 16 kbps. As will be described later, the radio resource allocation in the simulator is performed for time intervals of 10 ms. For this reason, the voice service has been implemented as users that transmit packets of 20 bytes every 10 ms. For data, the service model is similar to the full buffer service. This service is modeled by users with an infinite amount of information to transmit. Thus, a user will transmit with the maximum bit rate if there are available PRBs. This service is mainly conceived for throughput measurements. Service duration is finite (i.e., an random number of seconds). 25 CHAPTER 3. SIMULATION TOOL 3.3 Physical layer 3.3.1 Channel model The mobile radio channel can be described as a time-varying linear filter [71]. Therefore, it can be represented in the time domain by its impulse response, h(τ, t), where τstands for delay of each path in h, and the amplitude of each path varies with time t. Also, the channel can be characterized by the time-variant transfer function, H(f, t), which is related with impulse response through the Fourier transform with respect to the delay variable τ. When the behavior of the channel is randomly time variant, the above-mentioned channel functions become stochastic processes. A realistic approach to the statistical characterization of such a channel may be accomplished in terms of correlation of channel functions since it enables channel output autocorrelation to be determined. Channel autocorrelation functions are related through Fourier transform as well. For typical physical channels, time fading statistics can be assumed stationary over short periods of time and channel correlation function is invariant under a translation in time t, thus being categorized as wide-sense stationary (WSS). In addition, frequency-selective behavior is stationary in frequency fbeing the autocorrelation function invariant under frequency translations. This condition is termed uncorrelated scattering (US), and most practical channels satisfy it fairly well. Autocorrelation functions of wide-sense stationary uncorrelated scattering (WSSUS) channels exhibit the property that the time-variant transfer function autocorrelation is stationary both in time tand frequency fvariables, i.e., its value does not depend on the absolute time or frequency considered but only on the time or frequency shift between time or frequency points of observation. As a consequence, a WSSUS channel can be simulated generating the impulse response, h(τ, t), with stationary variation in time tfor each path and no cross-correlation between different values of delay τ(i.e., generating independent stochastic processes for different paths). Stationarity is achieved by applying Doppler filters to the amplitude time tvariation on each path. These filters perform spectrum shaping according to Doppler effect experimented by any radio signal propagating from a transmitter to a moving receiver (or vice versa). Afterwards, the frequency transfer function, H(f, t), can be computed easily by applying the Fourier transform to the impulse response with respect to the delay variable. To provide the possibility of simulating non-constant speed mobiles, fading realizations cannot be performed over time as an independent variable. Alternatively, space variables have to be used so that channel varies according to the current position of the mobile at each iteration of simulation. Therefore, a fading channel spatial grid has been generated. This grid provides channel responses for every physical position in the simulated scenario, regardless of mobiles speed. Narrow band fading grid is generated to get a Lord Rayleigh universe [72]. In other words, following Clarke’s model [71], a spatial bidimensional complex Gaussian variable is filtered by a 26 3.3. PHYSICAL LAYER bidimensional Doppler filter. The bandwidth of 2-D Doppler filter can be obtained as a function of spatial grid resolution and wavelength size. Once narrowband channel behavior for each spatial position is obtained, extension to wideband is possible performing the same procedure for every path in power delay profiles described in the specification for Extended Typical Urban (ETU), Extended Pedestrian A (EPA) and Extended Vehicular A (EVA) channels in [73]. Thus, different (uncorrelated) Rayleigh universes are generated for each delay in wideband channel scenario. This results in a distance-variant impulse response h(τ, d)(autocorrelation) of the channel instead of a time-variant impulse response h(τ, t)described in [71] as one of the four system functions for complete WSSUS channel characterization. The only difference is the time to distance (tto d) variable change made. A realization of the function is shown in Fig. 3.8. Figure 3.8: Generated bidimensional channel impulse response for ETU channel model in [73]. Since the simulator requires channel realizations for different frequency bands (corresponding to OFDM subcarriers), the distance-variant impulse response has to be transformed into a distance-variant transfer function H(f, d)at each position, by applying Fourier transform with respect to delay variable τ. An example of this function can be seen in Fig. 3.9. The only remaining step is to extend the space variable dof the generated function H(f, d)to a bidimensional (x, y)space variable, obtaining H(f, x, y), a tridimensional function that provides frequency response for each spatial position given by coordinates, xand y. 3.3.2 Radio propagation channel The radio propagation calculations are made from a set of pre-computed matrices. Thus, it is not necessary to perform the calculations during the simulation. For the definition of the precomputed propagation matrix, the scenario is divided into a grid, whose resolution is given by 27 CHAPTER 3. SIMULATION TOOL Figure 3.9: Generated distance-variant transfer function for ETU channel model in [73]. the correlation distance of the slow fading (50 m). To compute the propagation loss for a user, it is only necessary to read the index in the matrix corresponding to the position occupied by the user in the scenario relative to every base station and then interpolate it with other values of the matrix depending on the relative position in the grid. The propagation matrices include pathloss and slow fading. The radio propagation model is the COST 231 extension of Okumura-Hata model [74]. This model is applicable for frequencies in the range from 1500 to 2000 MHz. This model is valid for cell radius between 1 and 20 kilometers, base stations height between 30 and 200 meters and user terminal height between 1 and 10 meters. Equation 3.4 shows the radio propagation model COST 231 - Hata for cities: L= 46.3 + 33.9log fc−13.82 log hb+ (44.9−6.55 log hb)log r−a(hm) [dB],(3.4) a(hm)=3.2(log (11.75hm))2−4.97 ,(3.5) where fcis the carrier frequency, hbis the height of the base station, hmis the height of the user terminal and ris the distance between the user terminal and the base station. In the developed simulator, the effective height of the base station or eNB antenna has been set to 30 m, while the effective height of the user terminal antenna has been set to 1.5 m. With these assumptions and setting the carrier frequency to 2 GHz, the expression for the pathloss as a function of the distance is given by: L= 137.79 + 35.22 log r[dB],(3.6) 28 3.4. LINK LAYER From propagation losses, it is possible to study the link quality experienced by each user in terms of SINR, which is explained in the next section. 3.4 Link layer 3.4.1 SINR calculation The SINR is a representative measurement of the link quality that the user is experiencing. To calculate the SINR in the simulator, it is first necessary to compute the interference received by each user. It is assumed that intra-cell interference is negligible in LTE, because the scheduler assigns different frequencies and time slots to each user. Thus, only co-channel inter-cell interference due to interfering cells using the same subcarriers is considered. This requires estimating the signal level received from all interfering cells. To calculate the interference from each base station to the terminal, the channel response is not taken into account, but only the pathloss and slow fading. The SINR calculation for a given subcarrier k,γk, is computed using the expression proposed in [75], γk=P(k)ׯ G×N N+Np×RD NSD/NST ,(3.7) where P(k)represents the frequency-selective fading power profile value for the kth subcarrier, ¯ G includes the propagation loss, the slow fading, the thermal noise and the experienced interference, Nis the Fast Fourier Transform size used in the OFDM signal generation, Npis the length of the cyclic prefix, RDindicates the percentage of maximum total available transmission power allocated to the data subcarriers, NSD is the number of data subcarriers per Transmission Time Interval (TTI) and NST is the number of total useful subcarriers per TTI. If it is assumed that the multipath fading magnitudes and phases are constant over the observation interval, the frequency selective fading power profile value for the kth subcarrier can be calculated using the expression: P(k) =  paths X p=1 MpApexp (j[θp−2πfkTp]) 2 ,(3.8) where pis the multipath index, Mpand θprepresent the amplitude and the phase values of the multipath fading respectively, Apis the amplitude value corresponding to the long-term average power for the pth path, fkis the relative frequency offset of the kth subcarrier within the spectrum, and Tpis the relative time delay of the pth path. In addition, the fading profile is assumed to be normalized such that E[P(k)] = 1. The value of ¯ Gis calculated from the expression: 29 CHAPTER 3. SIMULATION TOOL ¯ G=Pmax gn(UE)×gUE P LUE,n×SHUE,n Pnoise +PN k=1,k6=nPmax ×gk(UE)×gUE P LUE,k×SHUE,k ,(3.9) where gn(UE)is the antenna gain of the serving base station in the direction of the user UE, gUE is the antenna gain of the user terminal, Pnoise is the thermal noise power, PLUE,k is the propagation loss between the user and the eNB k,SHUE,k is the loss due to slow fading between the user and the eNB kand Nis the number of interfering eNBs considered. A PRB is the minimum amount of resources that can be scheduled for transmission in LTE. As a PRB comprises 12 subcarriers, it is necessary to translate those SINR values previously calculated for each subcarrier into a single scalar value. This can be made using the Exponential Effective SIR Mapping, which is based on computing the effective SIR by the equation: SIReff =−βln 1 Nu Nu X k=1 exp −γk β!,(3.10) where βis a parameter that depends on the MCS used in the PRB [76] assuming that all subcarriers of the PRB have the same modulation and Nuindicates the number of subcarriers used to evaluate the effective SIR. The values of βhave been chosen so that the block error probability for all the subcarriers are similar to those obtained for the effective SIR in an Additive White Gaussian Noise (AWGN) channel [77]. The value of βfor a particular MCS is shown in Table 3.1. Once the SINR has been calculated, the Block Error Rate (BLER) showing the connection quality can be derived. There exist curves that establish the relationship between the values of SINR and BLER defined for an AWGN channel for every modulation and coding rate combination. These curves can also be used to calculate the BLER because inter-cell interference is equivalent to AWGN as the value of βhas been selected for this purpose. Then, given the value of BLER and taking into account the MCS used in the transmission, it is possible to calculate the value of throughput, Ti, for each user as follows: Ti= (1 −BLER(SINRi)) ×Di TTI ,(3.11) where Diis the data block payload in bits [78], which depends on the MCS selected for the user in that time interval and BLER(SINRi)is the value of BLER obtained from the SINR. 3.4.2 Link adaptation Before explaining the link adaptation function, the 3GPP standardized indicator known as CQI is described. Such an indicator represents the connection quality in a subband of the spectrum. The resolution of the CQI is 4 bits, although a differential CQI value can be transmitted to reduce the CQI signaling overhead. Thus, there is only a subset of possible MCS corresponding to a CQI value [79]. QPSK, 16QAM and 64QAM modulations may be used in the transmission 30 3.4. LINK LAYER Table 3.1: Values of βdepending on the Modulation and Coding Scheme Modulation Coding βfactor QPSK 1/3 1.49 QPSK 2/5 1.53 QPSK 1/2 1.57 QPSK 3/5 1.61 QPSK 2/3 1.69 QPSK 3/4 1.69 QPSK 4/5 1.65 16QAM 1/3 3.36 16QAM 1/2 4.56 16QAM 2/3 6.42 16QAM 3/4 7.33 16QAM 4/5 7.68 64QAM 1/3 9.21 64QAM 2/5 10.81 64QAM 1/2 13.76 64QAM 3/5 17.52 64QAM 2/3 20.57 64QAM 17/24 22.75 64QAM 3/4 25.16 64QAM 4/5 28.38 scheme. In the simulator, the CQI is reported by the user to the base station in each iteration (i.e., 100 ms). Based on CQI values, the link adaptation module selects the most appropriate MCS to transmit the information on the Physical Downlink Shared Channel (PDSCH) depending on the propagation conditions of the environment. To quantify the link quality for each user and subband of the spectrum, the CQI index is used to provide this information. If the experienced BLER value is required to be smaller than a specific value given by the service, it is possible to establish a SINR-to-CQI mapping that allows to select the most appropriate MCS from a given SINR value [53]. The standard 3GPP defines a 5-bit MCS field of the downlink control information to identify a particular MCS. This leads to a greater variety of possible MCSs. For simplicity, the developed LTE simulator includes only the same set of MCS given by the CQI index. From the SINR value, the CQI index is calculated and the MCS can be determined for the next time interval. In the developed simulator, Fig. 3.10 and 3.11 have been used for the SINR-to-CQI mapping. Once the MCS is selected, the throughput can be estimated based on the efficiency corresponding to each MCS [80]. 3.4.3 Resource scheduling The resource scheduling can be decomposed into a time-domain and frequency-domain scheduling. On the one hand, it is necessary to determine which user transmits at the following time 31 CHAPTER 3. SIMULATION TOOL Figure 3.10: SINR-to-BLER mapping, [53]. Figure 3.11: SINR-to-CQI mapping for a BLER<10%, [53]. interval. On the other hand, the frequency-domain scheduler selects those subcarriers within the system bandwidth whose channel response is more suitable for user transmission. For this purpose, the channel response for each user and for each subcarrier of the system bandwidth has to be estimated. Such a piece of information is given by the channel realizations generated in the initialization phase of the simulation, assuming a perfect estimation of the channel response. To select the most appropriate frequency subband for the user, the CQI index is used. The developed simulator includes different strategies for radio resource scheduling. In all of them, the CQI parameter gives the information of the channel quality experienced by each user. 32 3.4. LINK LAYER Likewise, scheduling is done for each cell at each iteration following the configured strategy [81]. The scheduling algorithms implemented in the simulator are: •Best Channel Scheduler (BC): in this scheduler, both time-domain and frequency-domain scheduling are done for a more efficient use of resources. At each iteration, all users are sorted based on the quality experienced for each PRB, which is obtained from CQI values. Once users are sorted, allocation procees until there are not available radio resources or no more users to transmit. The resource allocation is made by the following expression: ˆ i[n] = arg max i{rik[n]},(3.12) where ˆ iis the selected user iand rik is the estimated achievable throughput for PRB k and user iobtained from the CQI. This scheduling algorithm maximizes the overall system efficiency because the resource allocation is done looking for the combinations PRB-user with better channel conditions. The disadvantage of this algorithm is that harms users with bad channel conditions. Thus, if a user is far from the serving eNB or it is under a deep fading for a large period of time, it cannot be scheduled and it can suffer significant delays. •Round Robin to Best Channel Scheduler (RR-BC): this scheduler uses different strategies for time-domain and frequency-domain scheduling. For time-domain scheduling, the Round Robin method is applied. Thus, users are selected cyclically without taking into account the channel conditions experienced by each of them. Then, each PRB is assigned to the user with a higher potential transmission rate for that PRB. Transmission rate is estimated based on the user’s CQI value for each PRB. At each iteration, and for each base station, the expressions to be evaluated are: ˆ i[n+ 1] = (ˆ i[n] + 1) mod Nu(3.13) and ˆ k[n] = arg max k{rik[n]},(3.14) where ˆ iis the selected user, Nuis the number of users and ˆ krepresents the PRB selected. In this case, the goal is to maximize system efficiency, but trying not to harm users with unfavorable channel conditions. •Large Delay First to Best Channel Scheduler (LDF-BC): this scheduler is similar to the previous one only differing in time-domain scheduling. In this case, instead of cyclically selecting the users, they are sorted by the time they have spent without transmitting. Thus, if for some reason, such as a fading prolonged in time, the user has not been allocated in previous iterations, then he will get a higher priority in the current iteration. As in the previous case, at each iteration and for each base station, the allocation is carried 33 CHAPTER 3. SIMULATION TOOL Figure 3.13: User trace for a PBGT_HO execution. in order to favor the HO ping-pong occurrence. Congestion control The aim of the congestion control algorithm is to limit the admission of new connections to ensure the quality of the ongoing connections. In this test, two simulations are carried out. First, a simulation with the congestion control algorithm inactive is executed. The considered service is voice. The selected scheduler is RR-BC because the main requirement of the voice service is a small transmission delay. For convenience, the overall network load is configured to produce a high value of CBR and CDR to ease the analysis of the functionality. The second simulation uses the same configuration but the congestion control is activated and the overall network load is set to 80%. Fig. 3.15 and 3.16 show CBR and CDR results of the 57 cells in the scenario, respectively. In addition, the average values of both indicators is given in Table 3.3. Based on these results, it can be concluded that the congestion control algorithm achieves a significant reduction in the CDR, at the expense of increasing the CBR. 40 3.6. EVALUATION OF SYSTEM PERFORMANCE Figure 3.14: User trace for an HO ping-pong event. 1 2 0 10 20 30 40 50 60 0 2 4 6 8 10 12 14 Cells CBR(%) Congestioncontrolactive Congestioncontrolinactive Figure 3.15: CBR. Analysis of schedulers In order to study the different scheduler strategies included in the simulator, throughput and delay measurement statistics are analyzed. Fig. 3.17 shows the achieved average user throughput 41 CHAPTER 3. SIMULATION TOOL 1 2 0 10 20 30 40 50 60 0 1 2 3 4 5 6 7 8 Cells CDR(%) Congestioncontrolactive Congestioncontrolinactive Figure 3.16: CDR. Table 3.3: Global results of CBR and CDR Congestion control inactive Congestion control active CBR 3.14% 8.66% CDR 5.44% 3.31% per SINR for each scheduler, compared to the upper bound of system capacity obtained with the Shannon formula [53]. It can be seen that the BC and PF strategies achieve higher values of throughput than the RR-BC and LDF-BC. Fig. 3.18 and 3.19 show the cumulative density function (CDF) of transmission delay. From the figure, it is deduced that RR-BC and LDF-BC schedulers obtain the best results. According to these results, it can be concluded that BC achieves the higher throughput values at the expense of penalizing users with the lowest signal quality, which are usually located at the cell edge. RR-BC and LDF-BC aim to avoid large delays for the worst users, resulting in lower average throughput values. PF achieves good throughput values without incrementing the transmission delay of the worst users. The difference between BC and PF is more noticeable when comparing the average cell throughput, shown in Fig. 3.20. Values obtained for the PF algorithm are lower than that obtained by the BC because, even if throughput per SINR is similar for both schedulers, PF allows bad users to transmit to avoid large delays for these users. Such a different behavior has a large impact on the SINR distribution, shown in Fig. 3.21. The signal quality experienced by all users (i.e., users that transmit and users that do not transmit) can also be studied by analyzing pilot SINR measurements, as shown in Fig. 3.22. It is observed that the mode (i.e., the most frequent value) is 0 dB. In general, the obtained pilot SINR values are high. The reason is that the simulations have been carried out with a traffic load of 100%. This result is similar for every schedulers. 42 3.7. CONCLUSIONS Figure 3.17: Throughput per SINR for different scheduler strategies. Figure 3.18: Transmission delay distribution due to the lack of resources. 3.7 Conclusions In this chapter, a computationally-efficient dynamic system-level LTE simulator has been described. This simulator includes the main characteristics of the radio access technology as well as common radio resource management algorithms for improving the spectral efficiency. The simulator is conceived for benchmarking SON algorithms. Thus, the simulator has been implemented to have a low computational cost. A simulation consists of iterations to evaluate the 43 CHAPTER 3. SIMULATION TOOL Figure 3.19: Transmission delay distribution due to the lack of quality. Figure 3.20: Global throughput per cell. modification of network parameters performed by SON algorithms (e.g., a cell outage compensation algorithm). For a detailed analysis, the simulator provides several indicators at physical and link layers to check the connection quality of mobile users. Such indicators are used as an input to radio resource management functions such as link adaptation and dynamic scheduling. To obtained reliable results, it is essential that these indicators reflect the behavior of the real network accurately. For this purpose, an OFDM channel model has been implemented in the simulator to characterize the time and frequency variation of the radio transmission link for each 44 3.7. CONCLUSIONS Figure 3.21: Histogram of SINR distribution: (a) BC and (b) PF. Figure 3.22: Pilot SINR distribution. user. The main functions of radio resource management have also been described. At network level, the main functions are admission control and mobility management. Finally, several experiments have been carried out with the simulator. Results presented in this chapter include a detailed analysis of classical scheduling, HO and congestion control schemes. 45 CHAPTER 3. SIMULATION TOOL 46 Chapter 4 Cell Outage Detection This chapter presents a novel cell outage detection algorithm based on incoming HOs statistics. The proposed algorithm is able to detect outages even when the base station is affected, since it is based on measurements collected by neighboring cells. Note that, in this case, no performance indicators from the problematic cell would be available. Section 4.1 introduces the related work found in the literature. The following section, Section 4.2, describes the problem formulation. Sections 4.3 and 4.4 present the proposed algorithm and its evaluation. Finally, Section 4.5 summarizes the conclusions. 4.1 Related work One of the fundamental use cases in Self-Healing is Cell Outage Management (COM) [84]. COM comprises COD and COC functions. A cell is in outage when it cannot carry traffic due to a failure. In this situation, it is very important to identify the cell in outage as soon as possible to minimize the effects in the network. There are several methods to implement COD. In most cases [27, 28], the COD algorithm monitors KPIs and alarms reported by cells to determine if they experience problems. Specifically, in [28], the authors propose a method to detect different types of network performance degradation including cells in outage. With this methodology, it is possible to detect outage situations only if the eNB can supply KPIs from the cell in outage to the OSS (Operations Support System). However, if the outage affects the eNB, no KPIs will be available from the cell in outage. When this occurs, the only way to detect an outage in a certain cell is using information provided by its neighboring cells. In this line, in [29] the authors present a COD algorithm to detect outages based on the analysis of the KPIs reported by each cell which allows to determine if any of its neighboring cells is in outage. The effectiveness of this algorithm depends on the 47 CHAPTER 4. CELL OUTAGE DETECTION severity of problems caused in other cells by the cell outage. This is an important limitation of the algorithm because, in many cases, a cell in outage does not cause a performance degradation in the neighboring cells. The COD algorithm proposed in [30] is based on the neighbor cell list reports. This algorithm can detect cell outages even when the eNB is affected, since the detection is based on user measurements from the neighboring cells. User measurements are also used in [31, 32]. The algorithm presented in [31] is able to detect an outage in a femtocell scenario based on signal level measurements. In [32], the authors propose an algorithm to detect cells in outage based on user measurements combined with location information. However, the use of user measurements is the main drawback of all these approaches because the use of traces limits the bandwidth of the system and operators are unwilling to activate them. In this chapter, a COD algorithm that overcomes the previous problems is presented. 4.2 Problem formulation Fig. 4.1 shows different cases of cell outage. In some cases, the fault causing the outage affects only the related cell. In this situation, the related eNB can provide the KPIs from this cell indicating that the cell is not available due to a problem. In other cases, the outage affects the whole eNB. If the latter occurs, there are no KPIs available in the OSS from any cell of the site in outage. Therefore, if the detection algorithm is based on monitoring the value of different KPIs for each cell, this outage situation cannot be detected. Other possible detection algorithms can be based on the lack of KPIs for a certain cell. However, such a lack of KPIs does not always indicate an outage problem. For instance, an eNB-OSS connection failure may cause a lack of KPIs although the related cell is still active. Cellinoutage Siteinoutage eNB-OSSfailedconnection Cellwithoutproblems KPIs eNB Figure 4.1: Cases of cell outage To assess different COD algorithms, an LTE system model is proposed here. The considered system model (Cell Outage Model) allows to simulate diverse outage situations so that different detection algorithms can be tested. Moreover, the proposed system model also includes situations that can be detected as outage problems even when the cell operates properly to show the limitations of the algorithms. 48 4.3. CELL OUTAGE DETECTION BASED ON INCOMING HANDOVER STATISTICS The proposed Cell Outage Model includes different cases of cell outage, Fig. 4.1: •Cell outage that does not affect the eNB. In this case there are KPIs available from the cell in outage, although most KPIs are likely to be zero. The eNB indicates that the cell is not active using KPIs related to availability (availability KPIs). In some cases, the cell is automatically locked due to a problem. However, there are also other situations in which the cell is switched off by the operator for maintenance tasks or due to energy saving reasons. In this latter case, switched off cells should not be considered as cells in outage. •Site outage. In this situation the eNB is affected by the fault so that all cells covered by this eNB are impacted. As a consequence, no KPIs can be collected from the OSS. •Cell is not in outage, but there is a failure in eNB-OSS connection. The Cell Outage Model implemented here includes this fault because this situation can be erroneously detected as an outage problem by some detection methods. When an eNB loses the connection with the OSS the KPIs cannot be collected, however, the site is serving traffic. The Cell Outage Model can be applied to any type of LTE RAN simulator to evaluate different COD algorithms. 4.3 Cell Outage Detection based on incoming handover statistics The proposed COD algorithm allows to detect a cell in outage even when the eNB is affected and KPIs from that cell are not available. For this purpose, the algorithm is based on neighbor measurements. Specifically, the proposed algorithm monitors the number of incoming HOs (inHO) on a per-cell basis. If this number becomes zero for a certain cell, the algorithm selects the cell as a candidate cell in outage. The algorithm includes a configurable parameter, called measurement period, which determines the time period between two executions of the algorithm. When configuring this parameter, a tradeoff exists between detection delay and statistical significance. The lower the measurement period is, the faster the detection is. However, this parameter has to be high enough to ensure that the collected KPIs are statistically significant and depends on the periodicity of updating KPIs in the OSS. Fig. 4.2 shows a flow diagram of the proposed COD algorithm. In the first stage, the number of inHO on a per-cell basis is calculated based on HO statistics collected on per-adjacency basis. If it is the first measurement period in which the algorithm is activated, the next step is to wait for the next measurement period to draw conclusions from the comparison of consecutive periods. Otherwise, the second stage is executed for each cell of the network. Finally, in the third stage, the detected cell outages list is obtained and the algorithm is stopped until the next measurement period. The decision to determine if a cell is in outage is made in the second stage of the algorithm. 49 CHAPTER 4. CELL OUTAGE DETECTION significant impact on the overall network performance. Another false negative situation is when a cell suffers outages whose duration is less than the algorithm execution period. Hence, the false negative rate of the proposed algorithm can be calculated as the probability of occurrence of outages in a cell with no traffic and outages with a duration less than the algorithm execution period. Most detected cell outages are problems that have lasted a few hours but there are some cases of outage that have affected a cell during many hours even days. The detected cases have been confirmed by the operator. Fig. 4.5 - 4.8 illustrate how the algorithm works. All figures show the value of three KPIs for a certain cell during a day. The selected KPIs are the number of connections, the inactivity time indicating the number of seconds per hour that the cell has been inactive and the number of inHO. Firstly, Fig. 4.5 shows an outage case that does not affect the eNB. In this situation, there are KPIs available from the cell in outage. It can be seen that, when the number of inHO becomes zero, the availability KPI (i.e., inactive time) indicates that the cell is not active for the whole hour (3600 s). At the same time, the number of connections in that cell becomes zero too. Fig. 4.6 presents the behavior of the KPIs when a cell is in outage and there are no KPIs available. In this situation, the only way to detect the outage is using the number of inHO. The last two figures represent situations when no outages affect the network. In Fig. 4.7, a potential false positive situation is shown (hour 14), when the number of inHO becomes zero. in the absence of availability KPIs, this would indicate that the cell is in outage. However, the availability KPI shows that the cell has not been inactive at any time. Moreover, the figure shows that the cell is carrying traffic during all day. Such information is used by the proposed algorithm to discard this as a cell outage. The last figure, Fig. 4.8, presents the case of an eNB-OSS connection failure. It is observed that there are some hours with no KPIs available from the cell (i.e., number of connections and availability KPI are not available at certain hours). However, it is still possible to collect the number of inHO at these hours, since it is calculated based on neighbor measurements. This indicator shows that the considered cell has a nonzero number of inHO, which is clear indication that the cell is carrying traffic and does not have an outage problem. 4.5 Conclusions A COD algorithm has been proposed in this chapter. The algorithm is based on the number of inHO, which is available as a counter in the network management system. The algorithm allows to detect a cell outage even when KPIs from the considered cell are not available. A set of simulations and real network tests have been carried out to evaluate the proposed detection method. Results show that, unlike previous approaches, the proposed algorithm is able to detect most outage situations in a real network. The main limitation of the algorithm is that the cell outages that affect cells with very low traffic cannot be detected. However, these outage 56 4.5. CONCLUSIONS 0 5 10 15 20 25 0 2000 4000 6000 8000 10000 12000 Hour Number of connections Inactive time [s] inHO Figure 4.5: Cell in outage with available KPIs. 0 5 10 15 20 25 0 1000 2000 3000 4000 5000 6000 7000 Hour Number of connections Inactive time [s] inHO Figure 4.6: Cell in outage with no KPI available. situations would affect a small population and would thus have a low impact on the overall network performance. 57 CHAPTER 4. CELL OUTAGE DETECTION 0 5 10 15 20 25 0 50 100 150 200 250 300 350 Hour Number of connections Inactive time [s] inHO Figure 4.7: Cell without problems. 0 5 10 15 20 25 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 Hour Number of connections Inactive time [s] inHO Figure 4.8: eNB-OSS connection failure. 58 Chapter 5 Cell Outage Compensation This chapter is devoted to the COC methodology. The proposed COC scheme presents an important improvement of the COC function by adapting different COC strategies to different cell outage situations. For this purpose, a detailed analysis of the faulty situation is first carried out to classify the degradation produced by the cell outage in the neighboring cells. Then, a different COC algorithm is applied to each affected neighboring cell. In addition, some COC algorithms based on changing HO parameters are presented. The chapter is organized as follows. Section 5.1 introduces the related work. Section 5.2 presents the first phase of the COC methodology, namely the cell outage analysis. Three methods are described to classify outages depending on the degraded metrics in neighboring cells. In addition, a method for estimating the lost traffic due to a cell outage is proposed to quantify the load that cannot be absorbed by neighboring cells. Section 5.3 presents the study of new COC algorithms that can be adapted to the specific outage problem. Finally, Section 5.4 summarizes the conclusions. 5.1 Related work Most Self-Healing use cases defined by the 3GPP [9] are related to the COM. This function is composed of COD and COC. The COD function aims to detect the problematic situation, while COC aims to reduce the degradation caused by a failure in a cell until the fault is solved. The compensation can be made by modifying different configuration parameters in the network. Such parameters usually belong to the neighboring cells. All the modifications carried out by the compensation algorithm must be reverted when the network failure is solved. Several works that cope with these two issues can be found in the literature. As for the COD functionality, Chapter 4 includes a detailed analysis of the state-of-the-art related to this function. 59 CHAPTER 5. CELL OUTAGE COMPENSATION In the literature, a major difference between different COC algorithms is the control parameter used for the compensation action. Thus, the main control parameters presented in the literature are: antenna tilt ([33, 34]), uplink target received power level (P0) ([33, 35]), RS power ([33, 37]) and transmission power of the base stations ([36]). It is important to point out that operators usually are unwilling to modify the transmission power of the base stations, since these changes may produce coverage holes. Specifically, in [33], the authors evaluate the tunning of different parameters (i.e., reference signal power, P0 and antenna tilt) and analyze their effectiveness in different scenarios. In that study, the cell outage causes a coverage hole, but the compensation can be made focusing on improving coverage or throughput. The obtained results show that tunning P0 and antenna tilt are the most effective strategies in improving coverage, while P0 is the most effective in improving throughput. The objective in [34] is to increase the coverage area of the compensating cells in the outage area. Parameter changes are based on propagation measurements. The COC algorithm presented in [35] improves the coverage and signal quality in the outage area. The authors of [36] apply a COC algorithm to an irregular network (i.e., cellular network where base station locations, power levels and coverage areas are highly inhomogeneous). Finally, a distributed COC algorithm is presented in [37]. Compared to centralized schemes, the distributed algorithm can enhance management efficiency and remain active when the management node fails. It is also possible to find in the literature compensation methods that modify more than one parameter at the same time. In [38, 41], an algorithm for compensating an outage by simultaneously modifying the antenna tilt and the transmission power of the base station based on reinforcement learning is proposed. In [38], the authors present a comparison between the modification of both parameters and the compensation using each parameter separately. As expected, the conclusion is that the method that combines both parameters obtains the best results. In [41], the compensation algorithm is applied to a heterogeneous network. The same set of configuration parameters is used by the authors of [40] that present a COC algorithm based on fuzzy logic. There are also other works that implement COC algorithms applied to different technologies. In [39], the authors present a fuzzy logic algorithm that compensates outage problems by modifying the transmission power of several access points in a Wireless Local Area Network (WLAN). In addition to the previous works, some authors propose to apply Coverage and Capacity Optimization (CCO) algorithms to cell outage problems. In most cases, the modified parameter is the antenna tilt [42, 43, 44, 45]. The authors of [42] present an algorithm with two phases. The objective of the first phase is the antenna tilt optimization. In the second phase the antenna tilt is fine-tuned to adapt to network dynamics such as the failure of a neighboring cell. In [43], different strategies based on reinforcement learning are proposed. Each proposed strategy is analyzed in different network situations: deployment, normal operation and cell outage. The authors of [44] present a heuristic variant of the gradient ascent method that allows to optimize the antenna tilt even in a cell outage situation. Finally, there are other works that simultaneously modify more than one parameter in order to improve the results. For instance, in [45], an optimization 60 5.2. IMPROVING CELL OUTAGE MANAGEMENT THROUGH DATA ANALYSIS method based on the joint modification of antenna tilt and a cell reselection offset is proposed. All previous approaches present an important limitation. These works assume that the main effect produced by the cell outage is a coverage hole and they do not consider other possible effects, such as congestion or mobility problems. Consequently, most state-of-the-art COC algorithms aim to cover the outage area by increasing the coverage area of the neighboring cells by modifying the antenna tilt or the transmission power of the base stations. However, in some network conditions, a cell outage problem may not lead to a coverage hole. Consequently, the above COC algorithms may not compensate the caused degradation. 5.2 Improving Cell Outage Management through data analysis This section is organized as follows. Section 5.2.1 introduces the problem formulation. Section 5.2.2 describes the proposed methods for cell outages analysis. The proposed method for estimating the lost traffic due to a cell outage is presented in Section 5.2.3. Finally, some guidelines to improve the existing COC mechanisms are included in Section 5.2.4. 5.2.1 Problem formulation Before determining the overall impact of a cell outage, an accurate detection of this event is required. This is carried out by the COD functionality [27, 41]. In general, by monitoring alarms and KPIs from each cell it is possible to detect most cell outages. However, there are some particular cases where KPIs are not available, but the cell is still alive. In these cases, extra information such as neighbor cell lists or HO-related data is needed to discriminate both cases. For example, in the algorithm described in Chapter 4 when the number of inHO measured in neighboring cells becomes zero, the cell under study is likely to be in outage. Once cell outages are detected, a powerful COC scheme should distinguish between outage situations by analyzing the impact on neighboring cells. This is justified by the fact that some cases will require more attention than others from the operator perspective. For example, depending on the amount of traffic absorbed in the affected area, neighboring cells could become overloaded or not. A slight increase in the traffic of neighboring cells may happen in deployments with high cell densification or in areas with low offered traffic. In addition, in early deployment stages, a cell outage may produce important coverage holes which would negatively impact on user accessibility and retainability. In contrast, in mature deployments, the new scenario after a cell outage may lead to mobility problems in neighboring cells, especially between those with overlapped coverage areas. Moreover, the type of scenario may determine the effects caused by a cell outage, so that it is important to consider the type of affected cells (e.g., if the cell outage occurs in a heterogeneous network). Thus, it is clear that the same failure (i.e., a cell outage) may produce different effects in neighboring cells that should be compensated in a different way. Fig. 5.1 shows the effects produced in two different neighboring cells by the same cell outage. Specifically, the figure shows KPIs related to traffic (no. of connections), bad coverage (no. of 61 CHAPTER 5. CELL OUTAGE COMPENSATION measurements with RSRP level below a certain threshold) and cell load (Computer Processing Unit (CPU) processing load) for the two selected neighboring cells on an hourly basis. It can be seen that the cell outage that occurs in hours 59-62 produces a degradation in the coverage and an increase in traffic Neighbor 1, whereas Neighbor 2 suffers a congestion problem. In this situation, a COC strategy based on changing antenna tilt can improve the bad coverage issue in Neighbor 1 but would fail to solve the effect on processing load in Neighbor 2. 0 20 40 60 80 100 0 0.5 1 1.5 2 2.5 x104num_Connect Hour 0 20 40 60 80 100 0 500 1000 1500 2000 2500 bad_Coverage Hour 0 20 40 60 80 100 0 10 20 30 40 avg_CPU Hour 0 20 40 60 80 100 0 0.5 1 1.5 2x104num_Connect Hour 0 20 40 60 80 100 0 50 100 150 200 250 bad_Coverage Hour 0 20 40 60 80 100 0 10 20 30 40 avg_CPU Hour Neighbor1 Neighbor2 Figure 5.1: Degraded KPIs from two neighboring cells in case of cell outage. In this section, an analysis of cell outages based on identifying degraded KPIs in neighboring cells is presented. In a first (offline) approach, a method for analyzing historical records of cell outages is proposed. Then, various methods to analyze cell outages in a real-time (online) SON manner are presented and evaluated taking the offline approach as a baseline. 5.2.2 Analysis of cell outages Offline analysis method In Self-Healing, a common technique to manage anomalies in the network is to study the correlation between different metrics, such as alarms or KPIs. In particular, several works using the correlation for COD algorithms can be found in the state-of-the-art [85, 86]. The basis of these algorithms is the recognition of the characteristic behavior of faults by correlating the KPIs with each other. However, in this work, the correlation is not used to detect cell outages, which is done by the COD algorithm explained in Chapter 4. Alternatively, the correlation is used to analyze the effects produced by the cell outage in the neighboring cells. Since the cell outage is a 62 5.2. IMPROVING CELL OUTAGE MANAGEMENT THROUGH DATA ANALYSIS special fault where KPIs from the problematic cell are lost, the analysis of KPIs must necessarily be performed in neighboring cells. The first step in correlation-based Self-Healing is the selection of metrics to be correlated. This information should be as complete as possible, covering different aspects of the network (e.g., coverage, signal quality, capacity, etc.). A representative set of KPIs to analyze degradation in cell outages is the following: •num_Connect : it measures the number of established connections in a cell at network layer, so that it is an estimation of the carried traffic. •bad_Coverage: it is based on the number of UE reports indicating a low signal level received from the serving cell. A high value means that there is a lack of coverage in that cell. •inter_RAT_HO: it reflects the number of calls that have executed an HO to another cell belonging to a different radio access technology (RAT). This may reveal, for example, a lack of coverage in a certain RAT. •HO_PP: it measures the number of HO ping-pong events per cell produced in the network. As a consequence of a wrong configuration of mobility parameters, an HO ping-pong event occurs when the same user experiences two (or more) consequtive HOs that take place between the source and the target eNB and vice versa. This problem may significantly decrease the performance of HOs, while increasing network signaling load unnecessarily. •avg_RSSI : this KPI calculates the average value of the RSSI, which reflects the received power level including not only the desired signal but also background noise and interference. •avg_CPU_Load: it measures the average load of the CPU in the eNB. This KPI may indicate hardware congestion. •num_Drops: it counts the number of dropped calls in a cell as an indication of user retainability problems. •Failed_Conn_Estab: it measures the number of failed connection establishments as an indication of user accessibility problems. The above KPIs are currently measured on an hourly basis by the OSS. In the proposed method the time evolution of these metrics is used to find potential degradations on neighboring cells due to the impact of cell outages occurred in the network. The procedure to detect KPI degradations is shown in Fig. 5.2, where it is assumed that the duration of cell outages is only a few hours for clarity. The aim of the proposed algorithm is to generate a set of reference signals for each analyzed KPI that includes a certain level of degradation independently of the real values of the KPI. These estimated signals will be compared to the current values of the KPI to determine if there are degradations due to the outage. The algorithm is divided in four phases: 63 CHAPTER 5. CELL OUTAGE COMPENSATION Linearcorrelation coefficientcalculation Is> Threshold? DegradedKPI Non-degradedKPI Yes No 1 Outage 2 3 4 Maximum DefineObserved Signal Scalesignal EstimateSignalwith normalbehaviour Outage 0 10 20 30 40 0 0.6 1.2 1.8 x104 hour num_Connect 0 10 20 30 40 0 0.6 1.2 1.8 x104 DegradedKPI 0 10 20 30 40 0 1 2 3 4x104 0 10 20 30 40 0 1 2 3 4x105Outage Outage DefineReferenceSignals hour hour hour num_Connect num_Connect num_Connect Figure 5.2: Flow diagram of the proposed cell outage analysis method. 1Define Observed Signal. For each neighboring cell, a set of samples of each KPI in a period of two days is collected. One day should comprise the cell outage and the other day should not include any outage. 2Estimate signal with normal behavior. The previous signal is modified in order to generate a signal with no effects due to the cell outage. For this purpose, the samples that fall within the cell outage interval are substituted by samples at the same hours of the other day. These samples should not be affected by the cell outage. 3Scale signal. The samples at the hours of the cell outage are scaled to simulate a possible degradation produced by the outage. In particular, two different scaling factors (e.g., 3 and 30) are applied to cover both slight and strong degradations. The resulting signals are the reference signals. 4Calculate correlation. The linear correlation coefficient is calculated between the observed and the reference signals. This is made for each scaling factor. The maximum correlation value for the used scaling factors is selected. If the correlation value is above a specific threshold (e.g., 0.80), the KPI of the analyzed neighboring cell is considered to be degraded by the cell outage. 64 5.2. IMPROVING CELL OUTAGE MANAGEMENT THROUGH DATA ANALYSIS The time needed to complete the phases above is negligible compared to the KPI reporting period (typically, 1 hour), so that this analysis does not increase significantly the reaction time of the compensation process. In the proposed algorithm, the length of the signals has been carefully selected, since the correlation values are very sensitive to this parameter. More specifically, the time length of the signals should be consistent with the duration of cell outages. For example, an excessive time length of the signals compared to the duration of cell outages could mask the impact of the cell outage on the correlation values. Another issue is related to the selection of days in the step 1 of the algorithm. Depending on the day of the week, the traffic pattern may be different, especially between business days and weekend. Since hours at different days are correlated, the selected days of the week should also be consistent in relation to the traffic pattern. To avoid such an issue, in this work, only time intervals during the business days have been analyzed. The above-described algorithm has been applied to 22 real cases of LTE macro cell outages. The selected dataset corresponds to an LTE network composed of 8000 cells approximately. The information about different KPIs is collected every hour. The inputs of the algorithm are the above mentioned KPIs, which are measured in the neighboring cells of each cell in outage. To consider a cell as neighbor, the number of HOs between both cells must be higher than zero. To simplify the analysis, only the three most degraded neighboring cells (i.e., those with higher number of degraded KPIs) are selected. In Fig. 5.3, a histogram of the different degraded patterns that have been found in neighboring cells is depicted. As observed, the most repeated pattern (i.e., pattern 1) is the one without any degradation in the KPIs. This reveals that in many cases cell outages only have an impact on none, one or two neighboring cells. Pattern 2 is given by the degradation in num_Connect and bad_Coverage, which are the typical effects of cell outages in traditional single-RAT networks, i.e., an increase of traffic in neighboring cells and the creation of a coverage hole (reflected in a bad coverage statistics). One of the most repeated patterns, pattern 4, is given by a degradation only in HO_PP. This may be common in networks with a high degree of cell overlapping. In these cases, a cell outage may lead to a situation of a lack of dominant cell, where the number of unnecessary HOs would be increased. Pattern 5 presents a typical situation of multi-RAT networks, where the traffic can be mainly absorbed by other RATs (especially if cells are co-sited), impacting on Inter_RAT_HO rather than on num_Connect in the neighboring cell. Patterns 3 and 6 are subcategories of pattern 2, which are typical effects of cell outages, as previously mentioned. As the impact of a cell outage on neighboring cells becomes more severe, the number of degraded KPIs is also higher. An example of this is represented by patterns 11, 13, 23, 24, 26 and 27. In these cases, a coverage hole is created affecting both bad_Coverage and Inter_RAT_HO. In addition, a significant amount of traffic is absorbed by the neighboring cells so that num_Connect is increased. The coverage hole and the increased traffic in neighboring cells raises the likelihood of call dropping, which is illustrated by the increase in num_Drops. In some cases (see patterns 18, 24 and 26), the cell outage forces neighboring cells to carry an excessive amount of traffic, so that user accessibility is deteriorated. As a result, both Failed_Conn_Estab and num_Connect will be affected. Most of the remaining patterns are combinations of the previously analyzed 65 CHAPTER 5. CELL OUTAGE COMPENSATION network conditions, different situations may occur in the problematic area. In some cases, when the level of overlapping between the neighboring cells is low, the cell outage may produce a coverage hole. Affected users can only recover their connection if some compensating action is taken. Conversely, if the level of overlapping between the neighboring cells is high, it is possible that most affected users could recover their communication by establishing a new connection with a neighboring cell without any compensating action. In this situation, the cell outage may not result in a coverage hole, since the neighboring cells absorb most of the traffic from the faulty cell. However, depending on the amount of absorbed traffic, neighboring cells could become overloaded. In addition, the outage may produce new neighbor relationships. If HO parameters are not correctly configured between these new neighbors, a mobility problem could be also produced by the outage. The aim of the compensation algorithm in these two last cases (i.e., cell overload and mobility problem) will not be to recover the lost traffic, but to mitigate the overload or mobility problem. In this section, three different situations are considered: •Cell outage that results in a coverage hole (Coverage_outage): This situation occurs in scenarios where the overlapping areas between cells are very limited and/or the outage affects a large geographical area (e.g., when a whole site is down). Thus, the problematic area presents an important number of users that lose their connection. •Cell outage that produces a congestion problem (Load_outage): In this case, when the outage occurs, most users move to neighboring cells. This situation leads to a congestion problem due to the traffic absorbed by the neighboring cells. •Cell outage that produces a mobility problem (Mobility_outage): When a cell outage occurs and neighboring cells cover the problematic area, new neighbor relationships appear. If mobility parameters are not well configured between these new neighbors, different mobility problems can occur. One of these problems is the HO ping-pong [8]. A real network usually presents a very irregular layout. In such a scenario, the level of overlapping between neighboring cells or the load of the cells may be very different depending on the considered area. For this reason, when a cell or a group of cells (e.g., a whole site) is in outage, the neighboring cells of the cells in outage may present different types of degradation. When this occurs, the COC algorithm to be applied to each neighboring cell should be different. In this work, a scenario that combines more than one of the outage situations described above is also considered. 5.3.2 System model The following paragraphs describe the control parameters and system measurements, which are later used as outputs and inputs of the compensation algorithm. 72 5.3. ADAPTIVE CELL OUTAGE COMPENSATION IN SELF-ORGANIZING NETWORKS Control parameters As described before, one of the most commonly used parameters in COC is the antenna tilt angle. Tunning this parameter is effective when the outage causes a coverage hole in the network. In order to consider the antenna tilt modifications as part of the COC methodology, the vertical antenna radiation pattern (AV(θ)) should be modeled [87]. The following expression represents the model considered in this work: AV(θ) = −min[12(θ−θetilt θ3dB )2, SLAv], where −90◦≤θ≤90◦ (5.6) where θis the angle of inclination between the user and the eNB, θ3dB is the vertical halfpower beamwidth, θetilt is the electrical antenna downtilt (i.e., the angle of inclination of the transmitting antenna with respect to the horizontal plane) and SLAvis the side lobe level in dB relative to the maximun gain of the main beam. However, if the negative effects experienced by the neighboring cells are not related to a coverage degradation, antenna tilt modifications may not produce any improvement. For these other situations, HO parameter modifications are considered. One of the most widely used HO algorithms is that based on the A3 event defined by the 3GPP [88]. This event determines the condition that must be fulfilled to execute an HO, as (RSRPj)≥(RSRPi+HOM(i, j)) ,(5.7) where RSRPiand RSRPjare the RSRP measured by the user from cells iand j, respectively, and HOM(i, j)is the HOM defined between the cell iand neighbor j. The condition (5.7) must be fulfilled for a certain time period given by the TTT parameter. A specific value of HOM(i, j)and symmetric HOM(j, i)allow to determine a certain HO hysteresis (HOH) that avoids unnecessary HOs. HOH can be calculated as follows: HOH(i, j) = HOM(i, j) + HOM(j, i) = HOH(j, i).(5.8) In addition, these two parameters (i.e., HOM(i, j)and HOM(j, i)) can be used to apply a certain offset (HOoffset) that modifies the HO performance with load balancing purposes. Thus, HOM modifications can be made with different objectives: to modify the HOH or the HOoffset. On the one hand, HOM(i, j)and the symmetric HOM(j, i)can be jointly tuned to change HOoffset for load balancing purposes. In this case, the two margin parameters are modified with the same magnitude but opposite sign [89]. This kind of changes allows to modify the serving area of cells iand j, while maintaining HOH to avoid unnecessary HOs in the overlapped area between both cells. Alternatively, to adjust the HOH, the two margin parameters can be 73 CHAPTER 5. CELL OUTAGE COMPENSATION modified with the same magnitude and the same sign. Fig. 5.5 illustrates different HO performance depending on the value of HOM parameters. Fig. 5.5(a) shows that a low value of HOM favors a user to perform an HO to a neighboring cell although it may produce an increase in unnecessary HOs (and, consequently, an increase in HOs ping-pong) due to signal fluctuations. Conversely, Fig. 5.5(b) shows that HOM may be configured to a higher value in order to avoid these unnecessary HOs. In this thesis, HOoffset modifications are used in case of a Load_outage situation and HOH changes are used in case of Mobility_outage. RSRP Time HOM (a) RSRP Time (b) Cell A CellB Cell A CellB ChangetoB ChangetoB Changeto A ChangetoB Figure 5.5: HO process for different values of HOM parameter: (a) small and (b) large. System measurements A set of performance indicators has been selected in order to analyze the proposed algorithms. Some of these KPIs constitute the inputs of the algorithms and allow to decide when a control parameter modification is needed and when the performed changes lead the network to the optimal performance. The selected KPIs are the following: •Accessibility. This KPI indicates the capability of a cell to accept new connections. When a user requests a new connection to a certain cell, it may be blocked if the cell does not have enough available resources. The following expression indicates how to calculate this KPI: Accessibility = 1 −Nblocked Nattempts ,(5.9) where Nblocked is the number of blocked connections and Nattempts is the total number of connections attempts. •Retainability. This indicator represents the capability of a cell to maintain active connections under different environment conditions. Its calculation depends on the number of dropped connections in a cell. When a user abnormally loses its connection due to connec74 5.3. ADAPTIVE CELL OUTAGE COMPENSATION IN SELF-ORGANIZING NETWORKS tion quality or coverage problems, it is considered a dropped connection. The retainability can be expressed as the ratio: Retainability =Nsucc Ndrops +Nsucc ,(5.10) where Nsucc is the number of successfully finished connections and Ndrops is the number of dropped connections. •HO ping-pong Ratio, HPR. This KPI shows the percentage of HO ping-pong occurred in a certain adjacency, calculated as the ratio between the number of HO ping-pong and the total number of HO executed. This KPI can be calculated as follows: HP R =NHO_P P Nexecuted ,(5.11) where NHO_P P is the number of HOs ping-pong occurred in a certain adjacency and Nexecuted is the number of HO successfully executed in that adjacency. In particular, the NHO_P P for a certain adjacency (i, j)is obtained from the total number of outgoing HOs from cell iand the number of outgoing HOs from cell jthat return as incoming HO to cell iin a certain time period (ping-pong period). •Percentage of blocked connections due to a lack of coverage (Block_Cov). This indicator measures the percentage of connections affected by a lack of coverage. A user is considered to be out of the coverage area of a cell if the best signal level received from that cell is below the minimum required signal value. If the user is out of the coverage area of the strongest cell in the scenario, the user is considered to be blocked due to a lack of coverage. The received signal strength is represented by the RSRP level. The Block_Cov for a certain cell is obtained by dividing the number of blocked connections due to a lack of coverage by the total number of connections measuring that cell with the highest RSRP value in the scenario. The following expression represents this KPI: Block_Cov =Nblocked Nmeasured ,(5.12) where Nblocked is the number of blocked connections due to a lack of coverage in a cell and Nmeasured is the total number of users that measured the cell as the strongest cell. 5.3.3 Cell Outage Compensation methodology Fig. 5.6 shows the different phases of the proposed COC methodology. The COC methodology is triggered when a new cell outage problem is detected. Such a detection is carried out by the COD functionality presented in Chapter 4. When the COD algorithm detects a new cell 75 CHAPTER 5. CELL OUTAGE COMPENSATION outage problem, the cell outage analysis is carried out. This phase aims to analyze the kind of degradation produced in the neighboring cells by the cell outage. The details of this phase have been presented in the previous section. Depending on the kind of degradation detected, the most appropriate COC algorithm is selected for each neighboring cell. Network Outage Detection Outage Analysis Compensation Algorithm KPIs Parameter modifications CellOutageCompensationMethodology Figure 5.6: Cell Outage Compensation methodology. The following sections are devoted to each phase of the methodology. Cell outage analysis In order to apply an adaptive COC method, it is essential to classify each cell outage situation when it is detected. The proposed method checks whether there is any degradation in the neighboring cells at the time of the cell outage and which KPIs are most affected. With that aim, the KPIs of the neighboring cells are correlated with a reference signal, as described in Section 5.2.2, which represents the cell availability along time of the cell in outage. Such a reference signal can be defined as the unit step function with the discontinuity corresponding to the first hour of the cell outage. In this work, the Pearson correlation coefficient (eq. 5.1) is used. In addition to the correlation coefficient, the KPI value is compared to a pre-defined threshold. This threshold allows to determine if the detected degradation is severe enough to be compensated. A different threshold must be defined for each KPI. This threshold can be automatically calculated by the average of the specific KPI over a long time in a cell or over many cells in the network, assuming that most of the time there are no problems in the cells. An alternative method for the analysis, commonly used by operators, might be to directly compare the value of the KPI against the threshold, without considering the correlation. However, in a live network, 1) KPIs usually experience large fluctuations (i.e., spurious values) during normal behavior, and 2) the normal values can vary from cell to cell and also depending on network conditions. The correlation method avoids false positives and correctly detects degradation patterns in the neighboring cells’ performance. Then, the defined threshold allows to determine whether this degradation is enough to be considered problematic. Table 5.3 shows how the degradation produced by a cell outage in a certain neighboring cell is classified based on the affected KPIs in the three types described in Section 5.3.2: Cov76 5.3. ADAPTIVE CELL OUTAGE COMPENSATION IN SELF-ORGANIZING NETWORKS erage_outage,Load_outage and Mobility_outage. For clarity, in the following analysis of the three situations, it is considered that a cell outage affects all the neighboring cells in the same way. When a Coverage_outage has occurred, a coverage hole is produced in the outage area. In this situation, most users in the coverage area of the faulty cell lose their connection. The neighboring cells absorb only a small part of the affected users. The main degradation detected in the neighboring cells is an increase in Block_Cov. The only way to increase the number of users absorbed by the neighboring cells is extending their coverage area. The most appropriate solution for this situation is to perform tilt modifications in order to cover the outage area. Therefore, in the case of Coverage_outage, the COC algorithm should be based on tilt modifications (COC_TILT) and the selected cells for the compensation action should be the neighboring cells affected by the coverage hole. When the outage problem is a Load_outage, the neighboring cells absorb most of the affected connections. Thus, the percentage of users out of coverage tends to be very low. However, a congestion problem may affect the neighboring cells when the amount of absorbed traffic is high enough. In this case, an increase of the coverage area would not significantly affect the percentage of absorbed traffic since most traffic has already been absorbed and would only worsen the congestion problem. In this situation, the aim of the compensation algorithm should be to mitigate the congestion problem. The most suitable method for this scenario is a COC algorithm based on HOoffset (COC_HOoffset). The aim of this algorithm is to reduce the serving area of the affected cells and increase the serving area of their own neighboring cells. Thus, the congestion may be reduced. The selected cells in this case should be the neighboring cells of the cell in outage (i.e., cells affected by congestion) and their own neighboring cells. Finally, a cell outage may produce a mobility problem in its neighboring cells (i.e., Mobility_outage). As in the case of Load_outage, this kind of cell outage may not result in a coverage hole. Most of the affected traffic can be absorbed by the neighboring cells without any compensation action. Unlike Load_outage situation, in this case, the neighboring cells do not experience a congestion problem. Since the neighboring cells cover the outage area, new neighbor relationships may appear. This situation may produce mobility problems if the HO parameters are not correctly configured between these new neighboring cells. One typical mobility problem that may appear is the HO ping-pong problem due to a wrong value of the HOH. HOH changes (COC_HOH) allow to reduce the number of HOs ping-pong. The selected cells for the compensation are the affected neighboring cells of the cell in outage. As described above, in these two last cases (i.e., COC_HOoffset and COC_HOH), the adjusted control parameter is the HOM. A cell outage may affect each neighboring cells in a different way. In this case, one cell outage may produce different outage situations depending on the considered neighboring cell. Likewise, a neighboring cell may suffer different types of degradation simultaneously due to the cell outage. The COC algorithm should adapt the compensation action of each neighboring cell 77 CHAPTER 5. CELL OUTAGE COMPENSATION Table 5.3: Outage analysis Type of outage Degraded KPI Parameter Coverage_outage Block_Cov Tilt Load_outage Accessibility HOoffset Mobility_outage HPR HOH to the detected outage situation. In the case that a cell experiences different types of degradation, the COC actions should be prioritized. Cell Outage Compensation algorithms This section presents the three compensation algorithms applied in this work to the cell outage problem. All the algorithms are based on fuzzy logic. This technique is especially suitable to be applied to cellular networks since it allows to take decisions from imprecise information. In addition, the description of control actions in linguistic terms in fuzzy logic systems favors that the operator’s experience can be easily applied to the problem. In particular, three Fuzzy Logic Controllers (FLC) have been defined, one for each compensation algorithm. All of them are designed according to the Takagi-Sugeno approach [90]. Fig. 6.2 presents the main blocks in an FLC. The first stage, fuzzifier, is in charge of transforming the numerical input values into fuzzy inputs. This mapping is based on membership functions, . Several membership functions are defined for each input of the FLC, determining the degree of membership of any numerical input to different fuzzy sets. A linguistic term represents each defined fuzzy set. In this work, the terms High, Medium and Low are used. The values of the degree of membership is a real value between 0 and 1. The next stage is the inference engine which is based on a set of IF-THEN rules. The definition of these rules is based on the knowledge and experience of human experts, and with the ultimate goal of performing the appropriate compensation. Rules determine different situations that can occur with the corresponding action that the FLC should execute. Based on these rules, the inference engine calculates the output fuzzy sets. Specifically, depending on the input fuzzy sets, different rules may be activated with different strength (α). The activation strength of a rule k,αk, is calculated by the product operator: αk=µ(input1)·µ(input2).(5.13) Each rule produces a certain output (e.g., a constant value with an associated linguistic term, o). As result, a fuzzy output, α·o, is generated for each rule. Finally, the defuzzifier calculates the output crisp value from the results of the previous stage. In particular, depending on the rules’ outputs and the activated rules with the corresponding activation strength , the final crisp output is obtained as a weighted average as follows: 78 5.3. ADAPTIVE CELL OUTAGE COMPENSATION IN SELF-ORGANIZING NETWORKS output = N P i=1 αi·oi N P i=1 αi ,(5.14) where Nis the number of rules, oiis the output of rule iand αiis the activation strength of the rule i. Fuzzifier Inference engine Defuzzifier Rules Inputs Output Figure 5.7: Block diagram of a Fuzzy Logic Controller. The three algorithms presented hereafter differ in the considered inputs and outputs, the fuzzy sets and the rules. The configuration details for each method are explained in the following paragraphs. COC_TILT algorithm In this case, the aim of the algorithm is to increase the coverage area of the neighboring cells in order to cover a coverage hole. The FLC is executed separately for each selected neighboring cell (i.e., there is a FLC for each neighbor). The considered inputs are the Block_Cov and the Accessibility. The output of the FLC is the increment that must be applied to the antenna tilt of each neighboring cell. Fig. 5.8 shows the membership functions defined for each input. Three fuzzy sets have been defined for Block_Cov (i.e., Low, Medium and High) and two fuzzy sets have been defined for Accessibility (i.e., Low and High). The values selected for the different thresholds in the membership functions are similar to typical limits accepted by network operators. Table 5.4 presents the set of control rules, where L is Low, M is Medium and H is High. As for the fuzzy outputs, Negative means a decrease of the antenna tilt of 1◦(i.e., uptilt), Null means no change in antenna tilt and Positive means an increase of the antenna tilt of 1◦(i.e., downtilt). For instance, rule 1 can be read as: ’IF (Block_Cov is Low) AND (Accessibility is Low) THEN (∆Tilt is Positive)’. Each rule has been defined with a certain objective. Rules 4 and 6 are activated when the neighboring cell is suffering coverage problems but not congestion. Rule 1 is activated when the absorbed traffic begins to produce a congestion problem in the neighboring 79 CHAPTER 5. CELL OUTAGE COMPENSATION cell. Rules 3 and 5 avoid changes so that Accessibility is not deteriorated. Finally, rule 2 is in charge of maintaining the compensation situation once it is achieved. The obtained output crisp value is a real value that is rounded to -1, 0 or 1 (standing for Negative, Null or Positive, respectively). Finally, the resulting tilt value obtained by adding the increment to the previous tilt value is limited to [0◦−12◦] to avoid excessive changes. 11 Low High Low High 0.02 0.04 0.95 0.97 Block_Cov Accessibility (a) (b) Medium 0.06 µ(Block_Cov) µ(Accessibility) Figure 5.8: Input membership functions for (a) Block_Cov and (b) Accessibility. Table 5.4: Control Rules for COC_TILT algorithm No Block_Cov Accessibility ∆Tilt 1 L L Positive 2 L H Null 3 M L Null 4 M H Negative 5 H L Null 6 H H Negative COC_HOoffset algorithm This algorithm aims to change the service area of cells without modifying the hysteresis. This means that all changes to HOM(i, j)should be applied with equal magnitude and opposite sign to HOM(j, i), as explained in Section 5.3.2. The aim of this algorithm is to balance the load between the neighboring cells of the cell in outage (which have absorbed the traffic from the outage area) and their own neighboring cells. These last cells should not be neighbors of the cell in outage. Hereafter, each neighboring cell of the cell in outage is referred to as ’serving cell’ and its neighboring cells are referred to as ’adjacent cells’. The FLC is executed for each pair of serving and adjacent cell (i.e., there is a FLC per adjacency). The considered inputs are the Accessibility for the serving cell, s, the Accessibility for the adjacent cell, a, the current value of the HOM parameter, HOM(s, a), and the Retainability for the adjacent cell. Fig. 5.9 shows the membership functions defined in this case. Two fuzzy sets have been defined for both KPIs: Low and High. The values selected for the different thresholds related to Accessibility and Retainability are again typical limits accepted by network operators. As for the HOM(s, a), the parameter is considered High if the value is above 1 and Low if the value is below -1. A Low HOM(s, a)value facilitates the HO from the serving cell to the adjacent cell, thus offloading the serving cell. Table 5.5 presents the set of control rules, where L is Low and H is High. Rule 5 is 80 5.3. ADAPTIVE CELL OUTAGE COMPENSATION IN SELF-ORGANIZING NETWORKS activated when a load balance is needed, regardless of the current HOM value. In the case that the adjacent cell experiences problems related to Accessibility (rules 1 and 2) or Retainability (rules 4 and 5), a Positive change is applied only if the situation is produced by an excessive modification of HOM. Finally, rule 6 maintains the compensation situation once it is achieved. The output of the FLC, ∆HOM, represents the modification to be applied to HOM(s, a). The same modification with opposite sign must be applied to HOM(a, s). The resulting output increment values are rounded to -1, 0 or 1 dB (Negative, Null, Positive, respectively) and the HOM values are limited to [-12 - 12] dB. 1Low High -1 1 HOM(s,a) (a) (b) 1Low High 0.90 0.95 Accessibility/ Retainability µ(Accessibility/Retainability) µ(HOM(s,a) Figure 5.9: Input membership functions for (a) Accessibility and Retainability and (b) HOM(s, a). Table 5.5: Control Rules for COC_HOM algorithm No Acc Acc HOM Ret ∆HOM (s) (a) (s, a) (a) (s, a) 1 - L H - Null 2 - L L - Positive 3 - H H L Null 4 - H H L Positive 5 L H - H Negative 6 H H - H Null COC_HOH algorithm The aim of this algorithm is to modify the HOH in order to improve the HO performance between the neighboring cells when a cell outage occurs. To modify the hysteresis level, the modifications should be applied to HOM(i, j)and HOM(j, i)with the same sign and magnitude. The FLC is executed for each degraded adjacency between the neighboring cells of the cell in outage. The considered inputs are the HP R(i, j)per adjacency and the Retainability for cell iand for cell j. Fig. 5.10 presents the membership functions. As in the previous case, two fuzzy sets have been defined: Low and High. The values for the different thresholds have been selected according to the performance of the considered scenario in a normal situation. Table 5.6 shows the set of control rules, where L is Low and H is High. Rule 8 produces an increase of HOM when HPR is high. Rules 4, 6 and 7 are activated when the compensation action must be reverted if any cell 81 CHAPTER 5. CELL OUTAGE COMPENSATION 2 4 6 8 10 12 14 0 0.1 0.2 0.3 Simulationloops HPR 2 4 6 8 10 12 14 0.95 0.96 0.97 0.98 0.99 1 Retainability HPR(Tilt) HPR(HOoffset) HPR(HOH) Retainability(Tilt) Retainability(HOoffset) Retainability(HOH) Normal Fault Compensation Figure 5.14: Sensitivity analysis for Mobility_outage case. Fig. 5.15 shows the results obtained in the application of the COC_TILT algorithm to a Coverage_outage problem. The figure presents the average value for the neighboring cells for Block_Cov and Accessibility KPIs. The cell outage failure affects cells 10, 11 and 12, belonging to the same site (Fig. 5.11) and the selected neighboring cells are neighbors in the first tier (i.e., cells 8, 15, 28 and 29). During compensation, the minimum achieved value for the antenna tilt is 6◦. In the normal situation, this site covers a large geographical area. When the cell outage occurs, an important percentage of users suffer lack of coverage. When compensation is activated, this percentage decreases significantly. This decrease is achieved by uptilting the degraded neighboring cells to absorb the users in the outage area. While neighboring cells are accepting new users, its Accessibility is slightly degraded. The proposed algorithm achieves a balanced situation with a significant decrease of the Block_Cov and a slight degradation of the Accessibility. Fig. 5.16 presents the results for the Load_outage problem. The cells in outage are cells 1, 2 and 3. In this case, the outage failure causes a congestion problem in cells 11, 13, 63 and 66 (i.e., Neighbors Group A). The set of cells selected to carry out the compensation are 12, 14, 15, 61, 62, 64 and 65 (i.e., Neighbors Group B). The applied algorithm (i.e., the COC_HOoffset algorithm) performs parameter changes similar to a load balancing algorithm. Thus, the main KPIs that should be considered in this test are Accessibility and Retainability of both groups of cells (i.e., Accessibility of cell in Group A and Retainability of cells in Group B), Fig. 5.16. It can be seen that the algorithm improves the Accessibility of the neighboring cells of Group A (i.e., cells with congestion problems) without degrading the Retainability of the cells used for the compensation (i.e., Neighbors Group B). The maximum modification applied to HOM is 8 dB (i.e., 10 dB for HOM(i, j)and -6 dB for HOM(j, i)). 88 5.3. ADAPTIVE CELL OUTAGE COMPENSATION IN SELF-ORGANIZING NETWORKS 5 10 15 20 25 30 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 Simulationloops Block_Cov 5 10 15 20 25 30 0.92 0.94 0.96 0.98 1 Accessibility Block_Cov Accessibility Normal Fault Compensation Figure 5.15: Results for the COC_TILT algorithm in the Coverage_outage case. 5 10 15 20 25 30 0.9 0.95 1 Simulationloops Accessibility(Group A) 5 10 15 20 25 30 0.9 0.95 1 Retainability(GroupB) Accessibility(Group A) Retainability(GroupB) Normal Fault Compensation Figure 5.16: Results for the COC_HOoffset algorithm in the Load_outage case. Finally, the COC_HOH algorithm has been tested. This algorithm is applied to a Mobility_outage failure. As described in the previous section, for this test, the HOH for the neighboring cells is configured to zero, so that the initial hysteresis level is too low. Fig. 5.17 presents the average value of HPR and Retainability for the degraded neighboring cells (i.e., cells 10, 32, 63 and 66 in the first tier). By increasing the HOH value between the affected cells, it is possible to reduce the HPR. A value closer to that in the normal situation is achieved in the first iteration of the algorithm avoiding a degradation in Retainability. The maximum value for the HOH at the end of the tuning process is 2 dB. 89 CHAPTER 5. CELL OUTAGE COMPENSATION 5 10 15 20 25 30 0.04 0.08 0.12 0.16 0.2 Simulationloops 5 10 15 20 25 30 0.95 0.96 0.97 0.98 0.99 1 Retainability Normal Fault Compensation HPR Retainability HPR Figure 5.17: Results for the COC_HOH algorithm in the Mobility_outage case. For clarity, in the previous tests, it has be assumed that a cell outage affects in the same way all the neighboring cells. Thus, different strategies have been analyzed. In the last experiment, a more realistic situation is presented, where different neighboring cells show different effects from the outage. Specifically, a cell outage in cells 10, 11 and 12 is studied. This cell outage may cause different degradation in each neighboring cell. Once the specific degradation is detected for each neighboring cell, the corresponding COC algorithm is applied in the specific neighbor. The configuration is similar to that used in the Coverage_outage test, but with a new hotspot near cells 13 and 15. These conditions allow to simulate a cell outage causing a coverage hole in the outage area and a congestion problem in cells 13 and 15. Once the cell outage occurs, the analysis of the degradation is carried out. For each neighboring cell, the correlation of the reference signal and each considered KPI (i.e., Block_Cov, Accessibility and HPR) is calculated. Table 5.8 shows the obtained results. According to these results and considering degradation when the correlation coefficient is higher than 0.9, it can be concluded that cells 8, 13, 15 and 29 may be affected by the coverage hole; cells 13 and 15 may be suffering a congestion problem; and cells 3, 8 and 13 may experience a mobility problem. These results must be completed with the comparison to the pre-defined thresholds. To classify a certain neighboring cell as degraded cell, the value of Block_Cov has to be higher than 0.06, the value of Accessibility has to be lower than 0.96 and the value of HPR has to be higher than 0.2. Thus, after comparing to the thresholds, the analysis determines that cells 8 and 29 are affected by the coverage hole, while cells 13 and 15 experience a congestion problem. The rest of neighbors in the first tier are not significantly affected. Fig. 5.18 and Fig. 5.19 show the obtained results. Specifically, Fig. 5.18 shows the average value of the KPIs Block_Cov and Accessibility for cells 8 and 29. For these cells, the COC_TILT algorithm is applied to reduce Block_Cov by uptilting these cells. This is achieved at the expense of reduction of Accessibility. To avoid this negative effect, the COC_HOoffset could be applied, once Accessibility degradation appears. 90 5.4. CONCLUSIONS Fig. 5.18 does not consider this alternative. Table 5.8: Correlation results for the analysis phase Block_Cov Accessibility HPR Cell 2 0.08 0 0.47 Cell 3 0.69 0 -0.94 Cell 8 0.99 -0.61 -0.92 Cell 13 0.98 -0.97 -0.91 Cell 15 0.95 -0.98 -0.75 Cell 28 0.61 0 -0.67 Cell 29 0.95 -0.61 -0.54 5 10 15 20 25 30 0 0.05 0.1 0.15 Simulationloops Block_Cov 5 10 15 20 25 30 0.85 0.9 0.95 1 Accessibility Block_Cov Accessibility 55 Normal Fault Compensation Figure 5.18: Results for coverage hole problem. Simultaneously, cells 13 and 15 are affected by a congestion problem (Fig. 5.19). Following the same nomenclature that in previous tests, these two cells are considered Neighbors Group A and the set of cells used to reduce the congestion are Neighbors Group B (i.e., cells 2, 3, 14, 29, 40, 42, 69, 72). In this case, the COC_HOoffset algorithm is applied to compensate the congestion problem. Along simulation loops, Accessibility of the affected cells is increased but a slight degradation of Retainability of the cells from Group B is produced. However, the achieved Retainability values remain in an acceptable margin. 5.4 Conclusions This chapter has focused on the COC function. A novel COC methodology has been presented. The method adapts the compensation action on a per-neighbor basis according to the degradation produced by the cell outage in each neighboring cell. The adaptation consists of automatically 91 CHAPTER 5. CELL OUTAGE COMPENSATION 5 10 15 20 25 30 0.92 0.94 0.96 0.98 1 Simulationloops Accessibility 5 10 15 20 25 30 0.92 0.94 0.96 0.98 1 Retainability Retainability(GroupB) Accessibility(Group A) Normal Fault Compensation Figure 5.19: Results for congestion problem. selecting the compensating cells, the configuration parameters to be modified and the magnitude of the compensating action. As part of this new COC methodology, a study of the degradation produced by cell outages in the neighboring cells in a real network has been presented. Specifically, three new analysis methods have been proposed. A first offline method aims to analyze the degradation produced by the cell outage in the neighboring cells by correlating KPIs using historical records of cell outages. The two other methods are online approaches, which are executed immediately after the outage detection. These online methods are distinguished by the way that degradations in KPIs are determined. One of them is based on correlations and the other is based on delta detection. Both methods allow to determine the type of cell outage so as to effectively adapt the compensation algorithm. In addition, a method for estimating the lost traffic caused by a cell outage has been presented. Results have shown that cell outages may result in different issues that are reflected by different types of degradations in the neighboring cells. Such a set of different situations should be compensated in a different way. Then, an adaptive COC methodology comprising several COC algorithms has been proposed. Once the previous analysis method has determined the set of affected neighboring cells and the type of degradation caused by the cell outage, a different COC algorithm can be applied to each neighboring cell. In particular, three different cell outage situations are considered in this work: one for compensating coverage degradation, another one for alleviating load congestion degradation and one for solving mobility problems. The different COC algorithms solve the effects of the outage by modifying antenna tilt, HOM or HOH. A sensitivity analysis has been carried out to show how different types of cell outage situations should be compensated by modifying different control parameters. Results show that, for each cell outage problem, only 92 5.4. CONCLUSIONS one COC strategy achieves a successful compensation. Based on the previous observation, three COC algorithms implemented by FLC have been applied to different cell outage failures. In all cases, the COC algorithm manages to compensate the degradation produced by the cell outage without affecting other cells in the scenario. Finally, a more realistic scenario has been tested where the cell outage causes different types of degradation in the neighboring cells simultaneously. Results show that the proposed COC methodology successfully compensates the fault situation. 93 CHAPTER 5. CELL OUTAGE COMPENSATION 94 Chapter 6 Cell Degradation Compensation A novel CDC algorithm based on HOM modifications is presented in this chapter. Unlike the previous chapter, the fault considered here is weak coverage due to a power fault. In this faulty situation, the problematic cell is active during compensation, but its power is abnormally low. Section 6.1 presents the related work. Sections 6.2 and 6.3 outline the problem formulation and the system model. The proposed algorithm is presented in Section 6.4. Performance results are analyzed in Section 6.5.2. Finally, Section 6.6 summarizes the conclusions. 6.1 Related work This chapter presents a compensation algorithm that tries to mitigate the degradation caused by a fault different from a cell outage problem. Nonetheless, the related work includes all the studies about the compensation function. Thus, state-of-the-art COC algorithms presented in Section 5.1 can be considered as the related work in this chapter too. From the analysis presented in Section 5.1, it can be concluded that most compensation and coverage optimization algorithms have been applied to the same network failure, i.e., a cell outage. However, there are many other network failures that may cause an important degradation in network performance [46]. In this case, the faulty cell may carry traffic and it can therefore be considered for compensation. When this occurs, new parameters, such as HOM, can be considered to perform the compensation. This parameter has been extensively used for network performance optimization. In that context, the aim of adjusting HOMs has typically been the optimization of the HO process [91] or load balancing in case of congestion [89] in macrocellular scenarios. There are other works that investigate the HO management in heterogeneous scenarios, such as small cells or femtocells scenarios [92, 93, 94, 95]. However, the HOMs have not been previously considered for fault compensation. Even if both types of algorithms 95 CHAPTER 6. CELL DEGRADATION COMPENSATION (i.e., compensation and coverage optimization) tune the same parameters (e.g., HOM), in most cases, the changes made by each algorithm are different, since their objectives are different. 6.2 Problem formulation This section analyzes a compensation method for a degraded cell that is carrying traffic even when it is affected by a fault. The degradation considered in this thesis is weak coverage that occurs when the average signal level received by users in a cell is below the minimum required level [96]. This problem may be caused by issues such as a wrong parameter configuration or wiring problems, which lead to a reduction in the eNB transmission power. Such a power reduction causes a limitation of the cell coverage area in addition to a reduction of the received signal level. In this work, the analyzed fault is modeled as a reduction of the transmission power from the normal level (e.g., 46 dBm in macro cell [97]) to a lower level in the faulty cell. In particular, an offset representing the reduction of the transmitted power is defined. The transmission power, PowTx, of the faulty cell is defined as: PowTx =PowTxmax −offsetT x ,(6.1) where PowTxmax is the maximum transmission power and offsetT x is a configurable parameter to model different degradation levels. 6.3 System model The system model includes control parameters and system measurements. 6.3.1 Control parameters One of the most important effects caused by the considered problem is the limitation of the faulty cell coverage. For this reason, the most interesting parameters to take into account by the compensation algorithm are those whose modifications affect the degraded cell service area. In this work, the HOM is considered. This parameter determines the radio quality conditions for users to change their serving cell. By modifying this parameter, it is possible to control the service area of a cell. The HO algorithm considered in this work is based on the A3 event defined by the 3GPP [97], which determines the condition that must be fulfilled to execute the HO. This condition is given by the expression 96 6.3. SYSTEM MODEL (RSRPj)≥(RSRPi+HOM(i, j)) ,(6.2) which must be satisfied for a specific time period defined by the TTT parameter. In the formula, RSRPiand RSRPjare the RSRP received by the user from cells iand j, respectively, and HOM(i, j)is the HOM defined between a serving cell and each of its adjacent cells. HOM(i, j) is the parameter used in the proposed compensation algorithm. 6.3.2 System measurements To analyze the proposed compensation algorithm, a set of indicators reflecting the network performance is defined. These indicators allow to assess the effectiveness of the proposed algorithm and to determine the possible negative effects derived from the changes of configuration parameter settings. The considered system measurements are: •Retainability. This KPI represents the capacity of the network to maintain active connections under different radio link conditions. Thus, this indicator is related to the number of dropped calls in a cell. In this work, a dropped call occurs when a user abnormally loses its connection due to problems in the connection quality or coverage. The following expression indicates how to calculate this KPI: Retainability =Nsucc Ndrops +Nsucc ,(6.3) where Nsucc is the number of successfully finished connections and Ndrops is the number of dropped connections. •HO Success Rate (HOSR). This indicator shows the ratio of HOs that have been executed successfully, reflecting whether the HO parameters are correctly configured or there is a mobility problem in the network, e.g., a too late HO problem. It is essential to consider this KPI, since the proposed compensation algorithm is based on HOM modifications. These modifications may lead to a too late HO problem. This problem occurs when the RSRP received by a user from a certain neighboring cell is slightly higher than that received from the serving cell, but the expression 6.2 is not fulfilled. In this situation, the HO is not executed and maintaining the user connection may result in a dropped call if the RSRP from the serving cell is too low [8]. The too late HO problem appears when the HOM has high values which may occur due to the compensation algorithm modifications. Specifically, the HOSR can be calculated as the ratio between the number of HOs that have finished successfully and the total number of HO attempts. The latter includes all the successful HOs and the failed HOs. In this work, only the failed HOs due to a too 97 CHAPTER 6. CELL DEGRADATION COMPENSATION Then, a second set of simulations is carried out to analyze the algorithm performance in different situations. These tests include different levels of degradation in the faulty cell, scenarios with different user mobility conditions, different baseline configuration, different algorithm internal settings or several faulty cells simultaneously. In general, all simulations consist of the same phases. The first phase presents the normal operation of the network (i.e., no faults occur). The second stage presents the network performance when the weak coverage fault occurs in one or several cells. During this phase, the faulty cell suffers a reduction of the transmission power, but the compensation algorithm is not activated yet. Finally, in the third phase, the compensation algorithm is activated. Each simulation comprises 30 simulation loops in total (3 simulation loops of normal situation, 2 simulation loops of faulty situation and 25 simulation loops of compensation). Each simulation loop corresponds to an hour of network performance. The duration of each simulation loop is selected to guarantee that the indicators are statistically stable. However, these times can be reduced when the algorithm is applied to a real network. In this case, the limitation is the periodicity of updating the KPIs that are used as inputs of the algorithm. This periodicity can be less than an hour (e.g., 15 minutes) so that the total time to achieve a compensation situation can be reduced. The main figures of merit to compare the methods are SINR50 and Retainability. Unless stated otherwise, the values of interest are those obtained at the end of the compensation process. The following subsections present the results obtained in the different tests. The detailed configuration parameters, such as the considered faulty scenario and the specific algorithm and network settings, are described for each simulation case. 6.5.2 Results Preliminary study This first set of simulations analyzes the compensation of a weak coverage failure in a uniform scenario. The analysis includes three approaches based on HOM and antenna tilt modifications: HOM, TILT and HOM+TILT. The aim of changes is to reduce the service area of the problematic cell by forcing users located in the cell edge to change to a neighboring cell, to improve the service quality of the affected users. Specifically, the antenna tilt is modified from 5◦(i.e., default value) to 12◦for the faulty cell with a step of 1◦. In the case of the HOM, the parameter is modified from 3 dB (i.e., default value) with a step of 1 dB. The modification of the HOM is symmetric in each adjacency to maintain a hysteresis value and avoid the HO ping-pong. Fig. 6.6 and Fig. 6.7 show SINR50 and Retainability values obtained at the end of the compensation process for the faulty cell, the average value of the neighboring cells and the average value of the rest of cells. From the SINR values in Fig. 6.6, it is deduced that only the methods based on HOM modifications (i.e., HOM and HOM+TILT) achieve a significant improvement of signal quality, while the method based on tilt modifications (i.e., TILT) cannot improve signal quality in the faulty cell. Regarding Retainability, shown in Fig. 6.7, the methods based on tilt modifications (i.e., TILT and HOM+TILT) obtain the best results, although the results obtained 104 6.5. PERFORMANCE ANALYSIS with the HOM method are also acceptable. From these results, it can be concluded that, in this scenario, the proposed method (i.e., HOM) is competitive with the best of methods (i.e., that combining the modification of both parameters, HOM+TILT). Faulty cell Neighboring cells Rest of cells 0 0.5 1 1.5 2 2.5 3 3.5 4 SINR50 Faulty situation HOM TILT HOM+TILT Figure 6.6: SINR50 (dB). Faulty cell Neighboring cells Rest of cells 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Retainability Faulty situation HOM TILT HOM+TILT Figure 6.7: Retainability. Algorithm performance The following paragraphs describe the results of experiments carried out in the live scenario to assess the proposed algorithm (based on changing HOM). 105 CHAPTER 6. CELL DEGRADATION COMPENSATION A. Basic performance In this first test, the basic performance of the proposed algorithm is presented. As a first step, it is necessary to select the cell that is going to suffer the weak coverage fault. For this purpose, an analysis of the SINR50 and ThrUL50 values on a per-cell basis is carried out. Fig. 6.8 and Fig. 6.9 show SINR50 and ThrUL50 values for each cell in a normal situation. Although the scenario consists of 75 cells, only the value of 33 cells are considered for statistics. It can be seen that there is diversity in the values. 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 28 29 30 31 32 33 40 41 42 61 62 63 64 65 66 0 1 2 3 4 5 6 Number of cell Initial SINR50 (dB) Figure 6.8: 50th percentile of the SINR in a normal situation. In this work, cell 11 is selected as faulty cell. Since it is located in the center of the scenario, it allows to analyze the proposed compensation algorithm avoiding border effect. However, it can be seen that, for the selected faulty cell, the values of SINR50 and ThrUL50 of the neighboring cells are low. There are several cells that could be selected as faulty cells avoiding border effect (cells located in the center of the scenario) but the average value of SINR50 and ThrUL50 of the neighboring cells in these cases are very similar. As described in Section 6.4, the first stage of the compensation method is to select the neighboring cells in charge of the compensation and to obtain the algorithm settings. Compensating neighbors are obtained from the number of HOs executed between cell 11 and the remaining cells of the scenario in the normal situation. In this case the selected cells are: 2, 3, 10, 12, 13 and 15. In Fig. 6.5, it can be checked that these cells correspond to the first tier of neighboring cells of the faulty cell. Algorithm settings are calculated from the SINR50 statistics in the first simulation loops (i.e., normal situation), as described in Section 6.4. Table 6.3 summarizes the algorithm settings used in this test. 106 6.5. PERFORMANCE ANALYSIS 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 28 29 30 31 32 33 40 41 42 61 62 63 64 65 66 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 Number of cell Initial ThrUL50 (kbps) Figure 6.9: 50th percentile of ThrUL in a normal situation. Table 6.3: Algorithm settings Parameter Configuration Th10.97 Th20.98 Th32 Th43 Fig. 6.10 and 6.11 present the results of the algorithm by showing the time evolution of SINR50 and statistics. The curves correspond to the faulty cell KPI values and the average value of neighboring cells. In Fig. 6.10(a), it can be observed that, when the fault occurs (simulation loops 4 and 5), the SINR50 of the faulty cell suffers an important degradation. Once the compensation algorithm is activated and HOM modifications begin (simulation loops 6 and thereafter), SINR50 increases, reaching values similar to those of the normal situation. At the same time, the algorithm keeps the SINR50 of neighboring cells without degradation. A possible undesired effect might be a degradation in the uplink of the neighboring cells, since the changes made by the compensation algorithm cause an increase of the service area of the neighboring cells. Along the simulation loops, HOM changes cause that users from the faulty cell are re-assigned to a neighboring cell. These users are farther than other users of the neighboring cell, which translates into an increase of service area. Such an increase may affect the uplink performance, since these users may be interfered by the users from the faulty cell. Fig. 6.10(b) shows that, the proposed algorithm achieves a compensation situation without degrading the uplink of the neighboring cells. Fig. 6.10(b) shows how ThrUL50 in the faulty cell is increased due to a coverage area reduction by new HOM values, while ThrUL50 for the 107 CHAPTER 6. CELL DEGRADATION COMPENSATION 0 5 10 15 20 25 30 -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 Simulationloops (a) SINR50 (dB) 0 5 10 15 20 25 30 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 x104 Simulationloops (b) ThrUL50 (kbps) Faultycell Neighbors Normal Fault Compensation Normal Fault Compensation Faultycell Neighbors Figure 6.10: Basic performance: (a) SINR50 (dB) and (b) ThrUL50 (kbps). 0 5 10 15 20 25 30 0.99 0.991 0.992 0.993 0.994 0.995 0.996 0.997 0.998 0.999 1 Simulationloops (a) Retainability 0 5 10 15 20 25 30 0.8 0.82 0.84 0.86 0.88 0.9 0.92 0.94 0.96 0.98 1 Simulationloops (b) HOSR Faultycell Neighbors Faultycell Neighbors Normal Fault Compensation Normal Compensation Fault Figure 6.11: Basic performance: (a) Retainability and (b) HOSR. neighboring cells remains stable, despite the increase of their service areas. Fig. 6.11(a)-(b) present the time evolution of Retainability and HOSR, respectively. It can be seen that both Retainability and HOSR values for the neighboring cells suffer a slight degradation, but the final value remains in an acceptable level. For completeness, Fig. 6.11(b) includes error bars to show the variability of HOSR measurements across different simulations. Such a ratio is the only indicator that is not statistically stable due to the simulated scenario, where few HO attempts take place. To quantify this variability, a set of 20 simulations is carried out with the same setup. Error bars represent the 5th and 95th percentile of HOSR measurements in each iteration. It can be observed that, even if 108 6.5. PERFORMANCE ANALYSIS the minimum HOSR value is 0.82, the average value is higher. As remarked in Section 6.3, the fault considered in this work mainly affects users in the cell edge of the faulty cell. Although the proposed algorithm uses the 50th percentile as an input, it is also important to consider other percentiles to analyze the impact of the fault on other groups of users (e.g., users in the cell edge or near the base station). Thus, if a 5th percentile of SINR is considered, the analyzed information is related to the group of users that presents the worst quality performance in a cell. Usually, these users are located in the cell edge. Conversely, when a high percentile (e.g., 95th) is selected, the information provided is related to users with the best quality performance. These users are usually located near the base station. Finally, the 50th percentile (i.e., the percentile used as input of the algorithm) provides information about users with a performance near the average. Fig. 6.12 presents the 5th, 50th and 95th percentile of SINR when a reduction of 10dB in the base station transmission power is considered. Firstly, it can be seen that the fault affects users in all zones of the cell, as the three percentiles present degradation when the fault occurs. In this work, the improvement of the 50th percentile is the objective of the compensation algorithm. For this reason, this indicator achieves, after the compensation, similar values to that of the normal situation. The compensation algorithm does not have a significant effect in the 95th percentile because HOM changes mainly affect users in the cell edge. In fact, Fig. 6.12 shows that the values of the 5th percentile at the end of the compensation process are higher than that of the normal situation. Therefore, although the proposed algorithm uses the 50th percentile as an input, the cell edge of the faulty cell is also improved. In summary, the weak coverage fault causes a degradation in the SINR50 of the faulty cell, which was successfully overcome by the proposed method, without degrading neighboring cells. The analysis is completed with a comparison of different compensation approaches in the realistic scenario. The considered method for the comparison is based on tilt modifications since it is the method most frequently used for outage compensation. The initial antenna tilt angle of the selected cells (i.e., cell 11 as faulty cell and cells 2, 3, 10, 12, 13 and 15 as neighbors) are 2◦, 4◦, 2◦, 5◦, 4◦, 5◦and 4◦, respectively. The step used in the simulations is 1◦because bigger values may produce significant variations in the KPIs. The minimum angle is limited to 0◦in order to avoid negative values of the tilt angle. Several simulations have been executed in order to analyze different combinations of tilt changes for the faulty cell and the neighboring cells. The strategy that obtains the best results is based on downtilting neighboring cells (DN). Moreover, two additional compensation methods based on tilt modifications are tested. In particular, an approach based on uptilting neighboring cells (UN) and a method based on downtilting the faulty cell (DF). Fig. 6.13 shows the results obtained for the tilt compensation methods and the proposed algorithm (HOcomp). DN method allows to improve SINR50 of the faulty cell. With DN, the final achieved antenna tilt angle is 12◦for all compensating cells. Downtilting neighbors produce an interference reduction in the faulty cell and allow to increase the SINR50 even when it is accepting more users from the neighboring cells. The reason for that is that the experienced 109 CHAPTER 6. CELL DEGRADATION COMPENSATION 0 5 10 15 20 25 30 -10 -5 0 5 10 15 20 Simulationloops SINR(dB) Faultycell(5-tile) Neighbors(5-tile) Faultycell(50-tile) Neighbors(50-tile) Faultycell(95-tile) Neighbors(95-tile) Normal Fault Compensation Figure 6.12: 5th, 50th and 95th percentiles of SINR in a weak coverage fault situation. quality in the cell edge of the faulty cell is improved. However, the HOcomp method achieves a higher value of the SINR50, which is more similar to the normal situation. Specifically, the HOcomp method achieved a 33.1% of improvement compared to the DN method. In addition, the DN method presents a negative effect in the cell edge of the neighboring cells which is reflected by a reduction of the ThrUL50. On the other hand, the HOcomp method presents a lower value of the HOSR although it remains above 95%. Based on the obtained results, it can be concluded that the HOcomp and DN methods have a similar behavior (although the HOcomp method achieves a better result). Normal Fault HOcomp DN UN DF -1 0 1 2 3 4 SINR50 (dB) (a) Normal Fault HOcomp DN UN DF 0 0.5 1 1.5 2x104 ThrUL50 (kbps) (b) Normal Fault HOcomp DN UN DF 0.95 0.96 0.97 0.98 0.99 1 Retainability (c) Normal Fault HOcomp DN UN DF 0.8 0.85 0.9 0.95 1 HOSR (d) Faultycell Neighbors Figure 6.13: Comparison results: (a) SINR50 (dB), (b) ThrUL50 (kbps), (c) Retainability and (d) HOSR. 110 6.5. PERFORMANCE ANALYSIS DN presents an important disadvantage, downtilting several neighboring cells at the same time may generate coverage holes. Note that DN algorithm increases the service area of the faulty cell which is the opposite effect to that obtained with the proposed algorithm. Conversely, a reduction of the service area of the faulty cell can be obtained uptilting neighbors (i.e., UN method). This strategy is usually applied to outage problems. However, it is observed in Fig. 6.13 that this approach does not achieve a SINR50 improvement, since the interference in the faulty cell is also increased, as the interfering antennas point towards the service area of the faulty cell and the traffic load of interfering neighbors is higher. Another option to reduce the coverage area of the faulty cell is to increase its antenna tilt (i.e., DF method). Signal quality of the neighboring cells can be improved by downtilting the faulty cell antenna because the interference produced by the faulty cell can be reduced. However, the expected improvement is small, since the transmission power of the faulty cell has been reduced by the fault, so that the interference level produced in the neighboring cell is small too. In addition, when the faulty cell antenna tilt is increased, a deterioration of the cell edge of the faulty cell may be caused so that the SINR50 of the faulty cell cannot be increased (Fig. 6.13). For a more detailed analysis, Fig. 6.14 shows the 5th, 50th and 95th percentile of the SINR of the faulty cell and the neighboring cells when the DF method is applied. It can be seen that this approach does not achieve any improvement in the SINR of the faulty cell, so that it is not an alternative to compensate the fault. 1 2 3 4 5 6 7 8 9 -10 -5 0 5 10 15 20 Simulationloops SINR(dB) Faultycell(5-tile) Neighbors(5-tile) Faultycell(50-tile) Neighbors(50-tile) Faultycell(95-tile) Neighbors(95-tile) Normal Fault Compensation Figure 6.14: Comparison results: 5th, 50th and 95th percentile of the SINR of the faulty cell (dB). B. Impact of the magnitud of degradation In this test, the proposed algorithm is evaluated for different levels of degradation. For this purpose, several values for the offsetTx parameter are defined in order to implement different levels of degradation. Specifically, the simulations are carried out with values of 7 dB (Fault case 1), 10 dB (Fault case 2) and 13 dB (Fault case 3). Fig. 6.15 and 6.16 show the results obtained in 111 CHAPTER 6. CELL DEGRADATION COMPENSATION the simulations. Specifically, Fig. 6.15(a) shows the SINR50 temporary evolution for the different degradation levels. As expected, the higher level of degradation (i.e., Fault case 3) causes the most significant reduction in the SINR50 during the fault stage. However, once the compensation algorithm is activated, an improvement of this KPI is obtained for all the simulated levels of degradation. The final SINR50 value achieved by the algorithm is similar to that obtained in the normal situation, except for Fault case 3, which is a bit lower. The final value of the HOM achieved by the algorithm is different depending on the degradation level. In particular, the compensation situation for Fault case 1 is achieved for a minimum HOM(i, j)value of -5 dB. However, in the Fault case 3 a HOM(i, j)equal to -12 dB (i.e., the minimum allowed value) is needed to successfully compensate the fault. For the Fault case 2 the minimum HOM(i, j)value obtained in the compensation stage is -7 dB. The rest of the considered KPIs (i.e., ThrUL50, HOSR and Retainability) in the neighboring cells are not degraded much by the compensation actions. 0 5 10 15 20 25 30 -2 -1 0 1 2 3 4 5 Simulationloops (a) 0 5 10 15 20 25 30 0 0.5 1 1.5 2 2.5 x104 Simulationloops (b) Faultycell(7dB) Neighbors(7dB) Faultycell(10dB) Neighbors(10dB) Faultycell(13dB) Neighbors(13dB) Normal Fault Compensation Normal Fault Compensation SINR 50 (dB) ThrUL50 (kbps) Faultycell(7dB) Neighbors(7dB) Faultycell(10dB) Neighbors(10dB) Faultycell(13dB) Neighbors(13dB) Faultycell(7dB) Neighbors(7dB) Faultycell(10dB) Neighbors(10dB) Faultycell(13dB) Neighbors(13dB) Faultycell(7dB) Neighbors(7dB) Faultycell(10dB) Neighbors(10dB) Faultycell(13dB) Neighbors(13dB) Figure 6.15: Results for different degradation levels: (a) SINR50 (dB) and (b) ThrUL50 (kbps). C. Impact of initial network state As described before, the initial values of SINR50 and ThrUL50 for the selected neighboring cells are low. New simulations are performed to obtain differences in the initial SINR50 and ThrUL50 values. Specifically, the users distribution is modified to slightly improve the performance of the neighboring cells. For this purpose, the number of users generated is decreased compared to the default configuration. Fig. 6.17 below shows that the proposed algorithm achieves similar qualitative performance regardless the initial SINR50 value. 112 6.5. PERFORMANCE ANALYSIS 0 5 10 15 20 25 30 0.985 0.99 0.995 1 Simulationloops (a) Retainability 0 5 10 15 20 25 30 0.85 0.9 0.95 1 Simulationloops (b) HOSR Faultycell(7dB) Neighbors(7dB) Faultycell(10dB) Neighbors(10dB) Faultycell(13dB) Neighbors(13dB) Normal Fault Compensation Normal Fault Compensation Faultycell(7dB) Neighbors(7dB) Faultycell(10dB) Neighbors(10dB) Faultycell(13dB) Neighbors(13dB) Figure 6.16: Results for different degradation levels: (a) Retainability and (b) HOSR. 0 5 10 15 20 25 30 -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 Simulationloops (a) SINR50 (dB) 0 5 10 15 20 25 30 0.99 0.991 0.992 0.993 0.994 0.995 0.996 0.997 0.998 0.999 1 Simulationloops (b) Retainability Faultycell(defaultconfig.) Neighbors(defaultconfig.) Faultycell(newusersdistrib.) Neighbors(newusersdistrib.) Normal Fault Compensation Normal Fault Compensation Faultycell(defaultconfig.) Neighbors(defaultconfig.) Faultycell(newusersdistrib.) Neighbors(newusersdistrib.) Figure 6.17: Results for different users distribution: (a) SINR50 (dB) and (b) Retainability. D. Impact of HOM limits As described in Section 6.4, some rules are included in the algorithm to limit the HOM modifications to avoid that the performance of the neighboring cells gets worse as a consequence of the compensation actions. To illustrate the situation when no limit are applied to the proposed algorithm, new figures are presented. Fig. 6.18 and 6.19 show the network performance when consecutive HOM modifications are applied to a weak coverage problem with no limitations. The network performance when the proposed algorithm is activated is also included in the figures. Specifically, Fig. 6.18(a) shows that when no limits are applied to the HOM modifications the final SINR50 value of the faulty cell is higher than the one achieved by the method with limits. However, Fig. 6.19 shows the consequences of not having limits in the HOM modifica113 CHAPTER 6. CELL DEGRADATION COMPENSATION 120 Chapter 7 Conclusions This chapter summarizes the main contributions of this thesis. In addition, some future lines of work are suggested. Finally, a list of publications related to this thesis is presented. 7.1 Contributions This thesis is focused on two use cases defined by the 3GPP in the context of Self-Healing for mobile networks. Specifically, the work developed in this thesis includes new proposals for the detection and fault compensation functions. According to the structure of this report, the main contributions of this thesis can be organized in the following lines: a) Simulation tool A computationally-efficient tool for dynamic system-level LTE simulations has been developed in order to analyze the different algorithms proposed in this thesis. The definition, implementation and validation of this simulator have been part of this thesis. A physical layer abstraction is included to predict link-layer performance with a low computational cost. Thus, realistic OFDM channel realizations with multi-path fading propagation conditions have been generated. Additionally, the main RRM functions such as link adaptation, dynamic scheduling, admission control and mobility management are included in the simulator. The simulator is conceived for large simulated network time to evaluate the SON functions proposed in this work. b) Cell Outage Detection Most COD algorithms found in the literature are based on detecting cells in outage by monitoring KPIs or alarms from the cell in outage. Other works present COD algorithms 121 CHAPTER 7. CONCLUSIONS that are based on neighboring cell measurements. The main drawback of the first approaches is that they can detect outages only when there are available KPIs from the cell in outage, but there are many situations when the outage affects a whole site and the eNB is also affected. The second group of solutions need user traces to perform the detection, but trace collection is normally disabled in live networks to reduce processor load in the eNB. A novel COD algorithm has been proposed in this thesis. The proposed algorithm is based on the number of inHO measured on a per-cell basis by neighboring cells. Specifically, the proposed algorithm monitors situations where the number of inHO becomes zero as a potential symptom of cell outage. The main advantages of the algorithm are that: a) it is capable of detecting cell outages even if the base station is also affected, and b) it is based on performance counters that are available in the network management system. Due to the simplicity of the algorithm, it is possible to perform the detection immediately after collecting the KPIs. c) Cell Outage Compensation Most works related to COC in the literature assume that the main effect produced by the cell outage is a coverage hole. In densely populated area, where many sites are deployed, this is not typically the case. One of the main contributions of this thesis has been to overcome this limitation. •Analysis of cell outages. One of the main contributions of this thesis is considering that each cell outage situation may produce different effects in the neighboring cells. For this reason, it is essential to analyze the degradation caused in the neighboring cells when a cell outage occurs. In this sense, the analysis of massive amount of data can provide several benefits together with new avenues and challenges for improving the COC by analyzing real data sets of cell outages. In this thesis, the analysis of data collected in neighboring cells has led to the following contributions: –First, the idea of addressing the problem of each neighbor of a cell in outage as an independent case, as opposed to the state-of-the-art approach of not considering the different effects of the outage on neighbors. The proposed approach is based on the fact that cell outages can impact on each neighboring cell in a different way. For instance, in some cases, a cell outage could lead to mobility issues in neighboring cells due to a lack of a dominant cell, whereas, in other cases, the cell outage may produce a coverage hole. The analysis of effects is made by measuring the correlation (dependence) between KPIs in order to extract valuable information from a vast amount of data. Depending on whether the dataset includes historical (offline) or real-time (online) information of cell outages, different methods have been proposed to estimate the impact of outages on neighboring cells. To illustrate this approach, an analysis of cell outages in a mature LTE network has been carried out by looking at degraded metrics in neighboring cells. 122 7.1. CONTRIBUTIONS –Second, the idea that estimating the amount of traffic that is potentially lost as a consequence of cell outages should be considered by COC policies. This is important, since lost traffic can result in a potential loss of market share and revenues for the service provider. A method to estimate the lost traffic based on computing the traffic absorbed by neighboring cells has been proposed. •Adaptive COC algorithms. By taking advantage of the previous cell outage analysis, a new COC approach from the perspective of neighboring cells is discussed. Current COC solutions are typically focused on modifying a specific radio parameter (e.g., the antenna downtilt or the base station transmit power) in a predetermined number of neighboring cells to solve in most cases a problem of coverage hole. In this thesis, the shortcomings of this kind of solutions have been addressed and an improvement of the COC function has been presented. Such an improvement is achieved by adapting the COC function to each particular cell outage failure depending on the different effects that a cell outage may produce in the neighboring cells. Once a cell outage is detected, a detailed analysis of the effects produced in the neighboring cells is carried out. Based on the results of this analysis, it is possible to determine the set of cells that will take part in the compensation and the control parameter to be modified on a cell basis in order to adapt the compensation to the specific problem detected in each neighboring cell. Thus, the COC function is focused on mitigating the particular degradation caused by the cell outage. More specifically, three different COC fuzzy algorithms based on the modification of antenna tilt and HOM have been proposed. Antenna tilt modifications has been widely used for outage compensation in the literature. However, HO parameters have not previously been used with this purpose. Typically, the modification of HO parameters has been used for other SON functions, such as mobility robustness optimization or load balancing in case of congestion. d) Cell Degradation Compensation State-of-the-art compensation works aim to compensate a specific failure, namely the cell outage problem. In this thesis, the compensation of a network failure different to the cell outage is considered. Specifically, the analyzed degradation is a coverage deterioration in a cell due to a reduction of its transmission power in the downlink. The transmission power reduction may be caused by wiring problems or a wrong parameter configuration. In this situation, the faulty cell is still carrying traffic although its coverage area is reduced due to the fault. Thus, the faulty cell can be considered for the compensation process. This thesis has proposed to use the HOM as a new parameter for compensating cell degradation. In particular, the original contribution is related to two main aspects: •First, this work has considered a weak coverage fault, which is a different problem as the cell outage problem commonly addressed in SON literature. When a cell is in outage, it cannot carry traffic. Therefore, the main effect caused by a cell outage is the total loss of service in the problematic area. In such a situation, it is not possible 123 CHAPTER 7. CONCLUSIONS to consider the faulty cell as a part of the compensation algorithm. This thesis has proposed a compensation algorithm with the aim of mitigating the effects caused by a weak coverage fault considering modifications of the faulty cell parameters as part of the compensation method. Moreover, the degradation produced by the fault will be different to the one produced by a cell outage. •Second, this work has proposed a compensation algorithm based on HOM modifications, including the faulty cell and its neighbors. The use of this parameter with a compensation objective is an important contribution of this work. The HOM has been extensively used with optimization purposes, but it has not been previously considered for fault compensation. 7.2 Future work Possible lines of research that might continue the work in this thesis are the following: •One of the main research lines addressed in this thesis is the detection of failures in a network. In particular, this thesis is focused on the detection of the cell outage problem. A possible line of future work is the definition and implementation of cell detection algorithms that detect other kind of failures. Specifically, a cell detection algorithm to detect a weak coverage failure can be defined. •The COC use case has also been considered in this thesis. In this case, the network failures considered are the cell outage and the weak coverage. However, there are many other failures that can occur and need compensation actions (e.g., missing neighbors, external interference, high traffic, etc.). •The proposed method for cell outage analysis is based on the correlation of a set of KPIs. In this thesis, the most representative KPIs have been selected. However, depending on the kind of degradation, the most interesting KPIs may be different. In order to improve the analysis method, the set of considered KPIs can be extended. In addition, the selection of the most appropriate KPIs for each kind of degradation can be made automatically. The main drawback of this approach is that the complexity of the analysis increases. Recent development of Big Data and Data Mining techniques may help operators to cope with such a complexity. •Considering the previous possible lines, the ultimate goal may be to define a complete Self-Healing framework that allows to detect a large amount of different failures in a live network, to analyze the degradation produced by the fault in both the faulty cell and the neighboring cells, so as to adapt the compensation algorithms to mitigate the negative effects caused by the fault as quickly as possible. •A relevant issue in the field of SON is the coordination of use cases. In this context, the coordination between the proposed compensation algorithms, based on modifying HOMs and 124 7.3. PUBLICATIONS AND PROJECTS tilts, and optimization methods, such as mobility robustness optimization, load balancing or CCO algorithms, can be explored. •Finally, this thesis is focused on the development of SON functionalities. These features allow to cope with the complexity of network operation and to reduce costs. These two issues are foreseen to be the biggest challenges in 5th Generation (5G). A possible line of future work is the extension of the proposed methods to 5G networks by adapting the algorithms to the specific characteristics of these new networks (e.g., considering small cells). 7.3 Publications and projects The following subsections present the publications related to this thesis. 7.3.1 Journals Publication arising from this thesis [I] I. de-la-Bandera, R. Barco, P. Muñoz and I. Serrano, “Cell Outage Detection Based on Handover Statistics”, IEEE Communications Letters, vol. 19, no. 7, pp. 1189-1192, July 2015. [II] I. de-la-Bandera, R. Barco, P. Muñoz, A. Gómez-Andrades, and I. Serrano, “Fault Compensation Algorithm based on Handover Margins in LTE Networks”, EURASIP Journal on Wireless Communications and Networking, (2016) 2016:246, October 2016. [III] I. de-la-Bandera, P. Muñoz, I. Serrano and R. Barco, “Improving Cell Outage Management Through Data Analysis”, IEEE Wireless Communications, vol.PP, no.99, pp.2-8, February 2017. [IV] I. de-la-Bandera, P. Muñoz, I. Serrano and R. Barco, “Adaptive Cell Outage Compensation in Self-Organizing Networks”, IEEE Transactions on Vehicular Technology, Under review, 2016. [V] P. Muñoz, I. de-la-Bandera, F. Ruiz, S. Luna-Ramírez, R. Barco, M. Toril, P. Lázaro and J. Rodríguez, “Computationally-Efficient Design of a Dynamic System-Level LTE Simulator”, International Journal of Electronics and Telecommunications, 2011. Publication related to this thesis [VI] P. Muñoz, R. Barco, I. de-la-Bandera, E. J. Khatib, A. Gómez-Andrades and I. Serrano, “Root Cause Analysis based on Temporal Analysis of Metrics toward Self-Organizing 5G Networks”, IEEE Transactions on Vehicular Technology, Accepted, 2016. 125 CHAPTER 7. CONCLUSIONS [VII] A. Gómez-Andrades, P. Muñoz, E. J. Khatib, I. de-la-Bandera, I. Serrano and R. Barco, “Methodology for the Design and Evaluation of Self-Healing LTE Networks”, IEEE Transactions on Vehicular Technology, vol. 65, no. 8, pp. 6468-6486, Aug. 2016. [VIII] E. J. Khatib, R. Barco, P.Muñoz, I. de-la-Bandera and I. Serrano, “Self-healing in mobile networks with big data”, IEEE Communications Magazine, vol. 54, no. 1, pp. 114-120, January 2016. [IX] R. Acedo-Hernández, M. Toril, S. Luna-Ramírez, I. de-la-Bandera and N. Faour, “Analysis of the impact of PCI planning on downlink throughput performance in LTE”, Computer Networks, 76 - 1, pp. 42-54, 2015. [X] P. Muñoz, R. Barco and I. de-la-Bandera, “Load Balancing and Handover Joint Optimization in LTE Networks using Fuzzy Logic and Reinforcement Learning”, Computer Networks, vol. 76, pp. 112-125, 2015. [XI] P. Muñoz, R. Barco and I. de-la-Bandera, “On the Potential of Handover Parameter Optimization for Self-Organizing Networks”, IEEE Transactions on Vehicular Technology, vol. 62, no. 5, pp. 1895-1905, Jun 2013. [XII] P. Muñoz, R. Barco and I. de-la-Bandera, “Optimization of Load Balancing using Fuzzy Q-Learning for Next Generation Wireless Networks”, Expert Systems with Applications, vol. 40, no. 4, pp. 984-994, March 2013. [XIII] P. Muñoz, R. Barco, J. M. Ruiz-Avilés, I. de-la-Bandera and A. Aguilar, “Fuzzy Rule-based Reinforcement Learning for Load Balancing Techniques in Enterprise LTE Femtocells”, IEEE Transactions on Vehicular Technology, vol. 62, no. 5, pp. 1962-1973, Jun 2013. [XIV] J. M. Ruiz-Avilés, S. Luna-Ramírez, M. Toril, F. Ruiz, I. de-la-Bandera, P. Muñoz, R. Barco, P. Lázaro and V. Buenestado, “Design of a Computationally Efficient Dynamic System-Level Simulator for Enterprise LTE Femtocell Scenarios”, Journal of Electrical and Computer Engineering, 2012. 7.3.2 Patents Patents arising from this thesis [XV] I. de-la-Bandera, R. Barco, P. Muñoz and I. Serrano, “First network node, method therein, computer program and computer-redeable medium comprising the computer program for detecting outage of a radio cell”, WO 2016068761 A1, 2016. 126 7.3. PUBLICATIONS AND PROJECTS Patents related to this thesis [XVI] P. Muñoz, R. Barco, I. Serrano and A. Gómez-Andrades, “First network node, method therein, computer program and computer-readable medium comprising the computer program for determining whether a performance of a cell is degraded or not”, WO/2016/169616, 2016. [XVII] V. Buenestado, M. Toril, J. M. Ruiz-Avilés, I. de-la-Bandera and M. A. Regueira, “A new cell overshooting indicator for optimizing remote electrical tilt”, WO2015002676 A1, 2015. 7.3.3 Conferences and Workshops Conferences arising from this thesis [XVIII] I. de-la-Bandera, R. Barco, P. Muñoz and I. Serrano, “A Novel Cell Outage Detection Method for Self-Organizing Networks”, 13th MC & Scientific Meeting COST IC1004, Valencia (Spain) 2015. [XIX] I. de-la-Bandera, R. Barco, A. Gómez-Andrades, P. Muñoz and I. Serrano, “Compensación de Celdas Degradadas en Redes LTE”, XXIV Simposium nacional de la Unión Científica Internacional de Radio, Valencia (Spain), 2014. Conferences related to this thesis [XX] A. Gómez-Andrades, P. Muñoz, E. J. Khatib, I. de-la-Bandera, I. Serrano and R. Barco, “Simulador de fallos en una red LTE para sistemas de diagnosis”, XXIV Simposium nacional de la Unión Científica Internacional de Radio, Valencia (Spain), 2014. [XXI] P. Muñoz, R. Barco and I. de-la-Bandera, “Optimización Conjunta de Balance de Carga y Movilidad en Redes LTE”, XXIV Simposium nacional de la Unión Científica Internacional de Radio, Valencia (Spain), 2014. [XXII] P. Oliver-Balsalobre, M. Toril, I. de-la-Bandera, S. Luna-Ramírez and J. M. Ruiz-Avilés, “Simulador Dinámico Multi-Servicio para Análisis de la Calidad de Experiencia en Redes LTE”, XXIV Simposium nacional de la Unión Científica Internacional de Radio, Valencia (Spain), 2014. [XXIII] J. A. Fernández-Segovia, A. J. García-Pedrajas, I. de-la-Bandera, M. Toril and S. LunaRamírez, “Simulador estático de nivel de sistema del canal ascendente de datos en redes LTE”, XXIV Simposium nacional de la Unión Científica Internacional de Radio, Valencia (Spain), 2014 . [XXIV] R. Acedo-Hernández, M. Toril, S. Luna-Ramírez and I. de-la-Bandera, “Análisis de la planificación de señales de referencia en LTE con tráfico no uniforme”, XXVIII Simposium 127 CHAPTER 7. CONCLUSIONS nacional de la Unión Científica Internacional de Radio, Santiago de Compostela (Spain), 2013. [XXV] J. M. Ruiz-Avilés, I. de-la-Bandera, V. Buenestado and M. Toril, “Construcción de Contadores Sintéticos mediante Procesado de Eventos Complejo en redes LTE”, XXVIII Simposium nacional de la Unión Científica Internacional de Radio, Santiago de Compostela (Spain), 2013. [XXVI] R. Acedo-Hernández, M. Toril, S. Luna-Ramírez and I. de-la-Bandera, “Analysis of the impact of PCI planning on throughput performance in LTE”, 6th Management Committe And Scientific Meeting, Action Cost Ic1004, Málaga (Spain), 2013. [XXVII] P. Muñoz, I. de-la-Bandera, R. Barco, M. Toril, S. Luna-Ramírez and J. M. Ruiz-Avilés, “Sensitivity Analysis and Self-Optimization of LTE Intra-Frequency Handover”, 6th Management Committe And Scientific Meeting, Action Cost Ic1004, Málaga (Spain), 2013. [XXVIII] J. Rodríguez-Membrive, I. de-la-Bandera, P. Muñoz and R. Barco, “Load Balancing in a Realistic Urban Scenario for LTE Networks”, IEEE 73RD Vehicular Technology Conference, VTC, Budapest (Hungary), 2011. [XXIX] P. Muñoz, R. Barco, I. de-la-Bandera, M. Toril and S. Luna-Ramírez, “Optimization of a Fuzzy Logic Controller for Handover-Based Load Balancing”, IEEE 73RD Vehicular Technology Conference, VTC, Budapest (Hungary), 2011. [XXX] I. de-la-Bandera, P. Muñoz, R. Barco, M. Toril and S. Luna-Ramírez, “Auto-Ajuste del Margen de Handover en Redes LTE,”, XXVI Simposium Nacional de la Unión Científica Internacional de Radio, Leganés (Spain), 2011 [XXXI] P. Muñoz, I. de-la-Bandera, R. Barco, M. Toril and S. Luna-Ramírez, “Optimización del Balance de Carga en Redes LTE Mediante el Algoritmo de Q-Learning Difuso,”, XXVI Simposium Nacional de la Unión Científica Internacional de Radio, Leganés (Spain), 2011. [XXXII] J. Rodríguez-Membrive, I. de-la-Bandera, P. Muñoz and R. Barco, “Balance de carga en escenario urbano mediante controlador difuso para redes LTE”, XXVI Simposium Nacional de la Unión Científica Internacional de Radio, Leganés (Spain), 2011. [XXXIII] J. M. Ruiz-Avilés, S. Luna-Ramírez, M. Toril, F. Ruiz-Vega and I. de-la-Bandera, “Simulación Eficiente de Red de Femtoceldas LTE en Entornos de Oficina”, XXVI Simposium Nacional de la Unión Científica Internacional de Radio, Leganés (Spain), 2011. [XXXIV] J. M. Ruiz-Avilés, S. Luna-Ramírez, M. Toril, F. Ruiz-Vega, I. de-la-Bandera and P. Muñoz, “Analysis of Load Sharing Techniques in Enterprise LTE Femtocells”, Wireless Advanced (WIAD), London(UK), 2011. 128 7.3. PUBLICATIONS AND PROJECTS [XXXV] I. de-la-Bandera, S. Luna-Ramírez, R. Barco, , M. Toril, F. Ruiz.Vega and M. FernándezNavarro, “Controlador Difuso para Auto-Ajuste de Parámetros en Entorno de Red Móvil Heterogénea,”, XXV Simposium Nacional de la Unión Científica Internacional de Radio, Bilbao (Spain), 2010. [XXXVI] P. Muñoz, I. de-la-Bandera, R. Barco, F. Ruiz-Vega, M. Toril and S. Luna-Ramírez, “Diseño del Nivel de Enlace para un Simulador LTE”, XXV Simposium Nacional de la Unión Científica Internacional de Radio, Bilbao (Spain), 2010. [XXXVII] I. de-la-Bandera, S. Luna-Ramírez, R. Barco, M. Toril, F. Ruiz.Vega and M. FernándezNavarro, “Inter-System Cell Reselection Parameter Auto-Tuning in a Joint-RRM Scenario”, IEEE Fifth International Conference on Broadband and Biomedical Communications (IB2COM), Málaga (Spain), 2010. [XXXVIII] P. Muñoz, I. de-la-Bandera, R. Barco, F. Ruiz-Vega, M. Toril and S. Luna-Ramírez, “Estimation of Link-Layer Quality Parameters in a System-Level LTE Simulator”, IEEE Fifth International Conference on Broadband and Biomedical Communications (IB2COM), Málaga (Spain), 2010. 7.3.4 Book chapters [XXXIX] P. Muñoz, I. de-la-Bandera, F. Ruiz-Vega, S. Luna-Ramírez, J. Rodríguez-Membrive, R. Barco, M. Toril, P. Lázaro and M. Fernández-Navarro, “Developing a ComputationallyEfficient Dynamic System-Level LTE Simulator”, 4G Wireless Communication Networks: Design, Planning and Applications, River Publishers, 2013 7.3.5 Related projects This thesis was partially funded by the following projects: •P08-TIC-4052 grant from the Junta de Andalucía. •8.06/5.59.3721, contract with Optimi-Ericsson, with support from the Junta de Andalucía (Agencia IDEA, Consejería de Ciencia, Innovación y Empresa) and ERFD. •8.06/5.59.3722, contract with Optimi-Ericsson, with support from the Junta de Andalucía (Agencia IDEA, Consejería de Ciencia, Innovación y Empresa) and ERFD. •P12-TIC-2905 grant from the Junta de Andalucía. 129