Antenna Position Optimization via Machine Learning

Sammanfattning

As wireless communication systems continue to evolve toward higher capacity and greater flexibility, traditional fixed antenna deployments have become increasingly inadequate for coping with dynamic user distributions and complex propagation environments. To address this challenge, movable antenna systems have been proposed, introducing an additional spatial degree of freedom for system optimization, i.e., the position of each antenna element is movable. This paper mainly studies the optimization of movable antenna arrays in a multi-user MIMO system based on realistic ray tracing channels, aiming to improve system and rate performance through machine learning-based algorithms. Given the high-dimensional, non-convex, and derivative-free nature of the optimization problem, this work adopts two intelligent optimization techniques: Particle Swarm Optimization (PSO) and Genetic Algorithm (GA). A simulation framework combining the field-response channel modeling and deterministic ray tracing was established in MATLAB. Based on this framework, the two-dimensional positions of antenna elements were optimized to maximize the overall system throughput. Simulation results demonstrate that, compared to fixed sparse or half-wavelength arrays, both PSO and GA significantly improve the achievable sum rate under various user distributions, array sizes, and multipath conditions. This study confirms the effectiveness of machine learning-based algorithms in addressing complex wireless system optimization problems and provides a foundation for future research directions, such as integrating real-time prediction or reinforcement learning techniques. The findings of this thesis contribute theoretical insights and practical guidelines for the future deployment of movable antenna systems in next-generation wireless networks.