Abstract:
The increasing complexity of cellular network management and inhomogeneous Traffic patterns demand an enhanced level of automation in most of the network deployment and operational phases, it can not only simplify the complex network management tasks but also improve the user quality of experience by efficient resource utilization and minimizing the network response time to the network and environmental changes. In this thesis, we study the self-organized coverage and capacity optimization of cellular mobile networks using antenna tilt adaptations. We propose to use machine learning for this problem in order to empower the individual cells to learn from their interaction with the local environments. This helps the cells to get experienced with the passage of time and improve the overall network performance. We model this optimization task as a multi-agent learning problem using Fuzzy Q-Learning, which is a combination of Fuzzy Logic and Reinforcement Learning-based Q-Learning. Fuzzy logic simples the modeling of continuous domain variables and Q-learning provides a simple yet efficient learning mechanism. We study different structural and behavioral aspect of this multi-agent learning environment in this thesis and propose several enhancements for the basic FQL algorithm for this particular optimization tasks. Especially, we look into the effect of parallel antenna tilt updates by multiple agents (noise) to overcome the effect of noise environment on the learning convergence, the effect of selfish and We Develop this Work to get performance evolution in SINR, Data rate, Throughput, Spectrum efficiency and Delay Transmission.
