Imran, AliAsghar, Ahmad2018-12-182018-12-182019https://hdl.handle.net/11244/316814Mobile cellular network operators spend nearly a quarter of their revenue on network management and maintenance. Remarkably, a significant proportion of that budget is spent on resolving outages that degrade or disrupt cellular services. Historically, operators have mainly relied on human expertise to identify, diagnose and resolve such outages while also compensating for them in the short-term. However, with ambitious quality of experience expectations from 5th generation and beyond mobile cellular networks spurring research towards technologies such as ultra-dense heterogeneous networks and millimeter wave spectrum utilization, discovering and compensating coverage lapses in future networks will be a major challenge. Numerous studies have explored heuristic, analytical and machine learning-based solutions to autonomously detect, diagnose and compensate cell outages in legacy mobile cellular networks, a branch of research known as self-healing. This dissertation focuses on self-healing techniques for future mobile cellular networks, with special focus on outage detection and avoidance components of self-healing. Network outages can be classified into two primary types: 1) full and 2) partial. Full outages result from failed soft or hard components of network entities while partial outages are generally a consequence of parametric misconfiguration. To this end, chapter 2 of this dissertation is dedicated to a detailed survey of research on detecting, diagnosing and compensating full outages as well as a detailed analysis of studies on proactive outage avoidance schemes and their challenges. A key observation from the analysis of the state-of-the-art outage detection techniques is their dependence on full network coverage data, susceptibility to noise or randomness in the data and inability to characterize outages in both spacial domain and temporal domain. To overcome these limitations, chapters 3 and 4 present two unique and novel outage detection techniques. Chapter 3 presents an outage detection technique based on entropy field decomposition which combines information field theory and entropy spectrum pathways theory and is robust to noise variance. Chapter 4 presents a deep learning neural network algorithm which is robust to data sparsity and compares it with entropy field decomposition and other state-of-the-art machine learning-based outage detection algorithms including support vector machines, K-means clustering, independent component analysis and deep auto-encoders. Based on the insights obtained regarding the impact of partial outages, chapter 5 presents a complete framework for 5th generation and beyond mobile cellular networks that is designed to avoid partial outages caused by parametric misconfiguration. The power of the proposed framework is demonstrated by leveraging it to design a solution that tackles one of the most common problems associated with ultra-dense heterogeneous networks, namely imbalanced load among small and macro cells, and poor resource utilization as a consequence. The optimization problem is formulated as a function of two hard parameters namely antenna tilt and transmit power, and a soft parameter, cell individual offset, that affect the coverage, capacity and load directly. The resulting solution is a combination of the otherwise conflicting coverage and capacity optimization and load balancing self-organizing network functions.Mobile Cellular NetworksInformation and Communication TechnologyArtificial IntelligenceData Analytics