Multi-Agent DRL Approach to Two-Timescale Transmission for RIS-Aided MU-MISO Systems

Abstract

Reconfigurable intelligent surfaces (RISs) have become one of the key enabling technologies of the sixth generation (6G) wireless communications. In this letter, we investigate the joint precoding optimization at the base station (BS) and RIS for RIS-aided communication systems by leveraging the two-timescale paradigm. To balance between hardware cost and signal quality, we partition a column-wise controllable RIS into sub-surfaces with one-bit resolution. Then, we propose a scalable multi-agent deep reinforcement learning (MADRL) framework to maximize the system spectral efficiency (SE). To further reduce the computational complexity of BS precoding, we train a deep learning model to replace the numerical optimization methods. Simulation results verify the effectiveness and generalizability of the developed MADRL framework.