An Efficient Learning Algorithm for Phase Shift Optimization in RIS-Aided MISO Wireless Systems

Abstract

Due to dynamic change of the wireless environment, devising the local phase shift matrix for reconfigurable intelligent surface (RIS) can be a thought-provoking task. Therefore, in this paper, we come up with a deep reinforcement learning (DRL)-based beamforming optimization algorithm for RIS-aided multiple-input single-output communication systems. To be precise, we exploit a DRL-based structure based on twin-delayed-deep-deterministic-policy-gradient (TD3) algorithm to automatically adjust the phase shift of each unit at RIS to maximize the downlink received signal-to-noise ratio (SNR). Simulation results show that the raised TD3 algorithm performs a higher received SNR than the conventional DRL algorithm with a reduced running time.