Statistical CSI-based Beamforming for RIS-Aided Multiuser MISO Systems via Deep Reinforcement Learning

This letter presents a novel joint beamforming algorithm for reconfigurable intelligent surfaces (RIS) in multiuser multiple-input single-output (MISO) wireless communications. At first, by utilizing statistical channel state information (CSI) instead of instantaneous CSI, we significantly reduce channel estimation overhead. Then, the optimization of beamforming weights is accomplished using the proximal policy optimization (PPO) algorithm, a well-established actor-critic-based reinforcement learning (RL) approach. The impact of system parameters on user sum rate is also analyzed through simulations. The results show the PPO algorithm outperforms the existing methods by combining beamforming techniques with statistical CSI.