Optimization of Mobility Parameters using Fuzzy Logic and Reinforcement Learning in Self-Organizing Networks

Abstract

In this thesis, several optimization techniques for next-generation wireless networks are proposed to solve different problems in the field of Self-Organizing Networks and heterogeneous networks. The common basis of these problems is that network parameters are automatically tuned to deal with the specific problem. As the set of network parameters is extremely large, this work mainly focuses on parameters involved in mobility management. In addition, the proposed self-tuning schemes are based on Fuzzy Logic Controllers (FLC), whose potential lies in the capability to express the knowledge in a similar way to the human perception and reasoning. In addition, in those cases in which a mathematical approach has been required to optimize the behavior of the FLC, the selected solution has been Reinforcement Learning, since this methodology is especially appropriate for learning from interaction, which becomes essential in complex systems such as wireless networks.

Taking this into account, firstly, a new Mobility Load Balancing (MLB) scheme is proposed to solve persistent congestion problems in next-generation wireless networks, in particular, due to an uneven spatial traffic distribution, which typically leads to an inefficient usage of resources. A key feature of the proposed algorithm is that not only the parameters are optimized, but also the parameter tuning strategy. Secondly, a novel MLB algorithm for enterprise femtocells scenarios is proposed. Such scenarios are characterized by the lack of a thorough deployment of these low-cost nodes, meaning that a more efficient use of radio resources can be achieved by applying effective MLB schemes. As in the previous problem, the optimization of the self-tuning process is also studied in this case. Thirdly, a new self-tuning algorithm for Mobility Robustness Optimization (MRO) is proposed. This study includes the impact of context factors such as the system load and user speed, as well as a proposal for coordination between the designed MLB and MRO functions. Fourthly, a novel self-tuning algorithm for Traffic Steering (TS) in heterogeneous networks is proposed. The main features of the proposed algorithm are the flexibility to support different operator policies and the adaptation capability to network variations. Finally, with the aim of validating the proposed techniques, a dynamic system-level simulator for Long-Term Evolution (LTE) networks has been designed.