Deep Reinforcement Learning-Based Downlink Beamforming and Phase Optimization for RIS-Aided Communication System

Reconfigurable intelligent surface (RIS) has been envisioned as a critical technology for the future wireless communication systems. However, the inter-cell interference and the unbearable signal processing complexity seriously constrain the wide application of RIS. In this letter, we investigate a deep reinforcement learning (DRL)-based scheme of downlink beamforming and phase optimization for a multi-cell RIS-aided communication system. With the transmit power constraint, a maximization problem of average ergodic rate is formulated. We design a deep deterministic policy gradient (DDPG)-based scheme, which enables base station (BS) and RIS to intelligently control the beamforming and phase shift, respectively. Further, a DDPG-based distributed training decentralized execution algorithm is proposed to acquire the feature difference among multi-cells and maximize the average ergodic rate. Simulation results show that the proposed algorithm outperforms the other baseline algorithms, which proves the agents can effectively learn and sense the change of multi-cell interference.