DRL-based Joint Beamforming Design for RIS-assisted mmWave MU-MISO system

Reconfigurable intelligence surface (RIS) is an emerging technology based on digital metamaterials, which can improve the transmission performance of wireless communication systems by controlling the incident signal beam by changing the electromagnetic properties of its reflecting elements. In this paper, a RIS-assisted millimeter-wave (mmWave) multi-user multiple input single output (MU-MISO) communication system is first designed, then the active hybrid precoding matrix at the base station (BS) and the shift matrix at the RIS are jointly optimized to maximize the user sum-rate. Finally, the deep deterministic policy gradient with prioritized experience replay (PER-DDPG) algorithm is proposed to solve the nonconvex problem arising from the joint optimization. Experimental results show that the proposed joint beamforming algorithm based on PER-DDPG has a higher average reward, faster convergence, and more stable training process than the traditional alternating iteration algorithm and the classical deep reinforcement learning (DRL) algorithm.