Deep Reinforcement Learning Applied to ORIS-Aided Downlink MU-MIMO VLC Systems

Abstract

Optical Reconfigurable Intelligent Surfaces (ORIS) has recently gained attention as a promising technology for enhancing the performance of multiuser multiple-input multiple-output (MIMO) visible light communication (VLC) systems. By integrating ORIS with advanced optimization techniques, such as Deep Reinforcement Learning (DRL), significant improvements can be achieved regarding spectral efficiency, energy efficiency (EE), and overall system capacity. This paper proposes a novel approach that employs DRL to optimize both the transmit BF matrix at the transmitter and the phase shift configuration at the ORIS. The proposed method demonstrates significant improvements over conventional optimization algorithms, achieving a performance increase of up to 25% in efficiency during extensive simulations conducted over 800 episodes. This enhanced solution provides greater adaptability for complex indoor MU-MIMO VLC environments, showcasing its effectiveness compared to traditional approaches.