DRL-Based RIS Phase Shift Design for OFDM Communication Systems
Peng Chen, Xiao Li, Michail Matthaiou, Shi Jin
Abstract
This letter investigates a downlink orthogonal frequency division multiplexing (OFDM) transmission system aided by a reconfigurable intelligent surface (RIS). To reduce the system overhead and cost, we consider a 1-bit resolution and column-wise controllable RIS, and aim to design the reflection phase shifts of the elements on the RIS to improve the spectral efficiency. By leveraging a deep Q-network (DQN) framework, a deep reinforcement learning (DRL) based optimization algorithm is proposed in order to design the reflection phase shifts. Simulations illustrate that the proposed DRL-based algorithm can achieve significant performance gains in the spectral efficiency, while greatly reducing the calculation delay.