Stochastic Methods for Beam Weight Optimization for Large Antenna Arrays

Abstract

Traditional methods for beam weight optimization suffer from poor convergence and long run times for large antenna arrays and are limited to ideal antenna element models. To overcome these limitations, more efficient optimization algorithms are needed. Previous research has demonstrated that stochastic approaches such as genetic algorithms and particle swarm optimization are promising alternatives for solving this problem. This thesis investigates the tailoring of these stochastic optimization algorithms for beam weight optimization using high-dimensional, realistic, simulated data. We propose the use of an absolute radiation pattern scale as opposed to a relative one, which allows us to simplify the objective function and promote a more efficient energy usage. We decrease runtimes by restricting the optimization to a low-rank search space, implementing warm starts, and leveraging GPU acceleration. We find that these tailored methods show robust performance and converge significantly faster compared to current methods. The resulting beams also consistently outperform those currently in use. We demonstrate the effectiveness of our approach on antenna arrays of up to 1,152 subarrays, where antennas of 288 subarrays or more are considered large.