摘要
Robustness of reconfigurable intelligent surface (RIS) has been a concern due to potential hardware impairments (HWI) and imperfect channel state information (CSI) measurements caused by the numerous passive elements on board. Recent studies observe that the impairments not only introduce mis-alignment in phase adjustments but also affect the amplitude of reflected signals, which further complicates the issue. To address this issue, we introduce a novel deep reinforcement learning (DRL)-based discrete optimization framework aimed at mitigating various HWI and CSI imperfections in RIS-assisted millimeter-wave (mmWave) multi-input-single-output (MU-MISO) systems. Employing proximal policy optimization (PPO), our method discretely addresses HWI and CSI challenges without continuous relaxation. Simulation results demonstrate the superiority of our approach over the traditional optimal beamforming baseline in minimizing the worst-case mean squared error (MSE) of the signal received by the users. The code has been made open-source on GitHub, serving as a valuable reference for further research and application in RIS-assisted communication systems.