Reconfigurable Intelligent Surface-Enhanced Broadband OFDM Communication Based on Deep Reinforcement Learning

This paper investigates the downlink OFDM transmission assisted by reconfigurable intelligent surface (RIS). With single antenna implemented at both the base station (BS) and each user, we focus on the design of the phase shifts for the RIS, as well as power allocation on each subcarrier to improve the spectrum efficiency. To reduce the computation delay, we propose a deep reinforcement learning (DRL) based algorithm to optimize the RIS phase shift parameters, while allocating power on each subcarrier via water filling. Numerical results reveal that the proposed DRL-based framework can achieve a performance almost the same with that of successive convex approximation (SCA), while the computation delay can be greatly reduced.