A Deep Evolution Policy-Based Approach for RIS-Enhanced Communication System

This paper investigates the design of active and passive beamforming in a reconfigurable intelligent surface (RIS)-aided multi-user multiple-input single-output (MU-MISO) system with the objective of maximizing the sum rate. We propose a deep evolution policy (DEP)-based algorithm to derive the optimal beamforming strategy by generating multiple agents, each utilizing distinct deep neural networks (DNNs). Additionally, a random subspace selection (RSS) strategy is incorporated to effectively balance exploitation and exploration. The proposed DEP-based algorithm operates without the need for alternating iterations, gradient descent, or backpropagation, enabling simultaneous optimization of both active and passive beamforming. Simulation results indicate that the proposed algorithm can bring significant performance enhancements.