Dynamic Channel Optimization of Reconfigurable Intelligent Surface in Communications

Reconfigurable Intelligent Surface (RIS) is emerging as a key enabler for post-5G and beyond networks. To address performance degradation in dynamic channels, this paper proposes a prediction-decision-control closed-loop framework combining deep reinforcement learning and group decomposition for real-time RIS phase adjustment. Validated via simulations and a millimeter-wave prototype, the method outperforms greedy and convex optimization approaches in spectral efficiency, retains notable capacity gains in non-line-of-sight (NLoS) scenarios, cuts down control latency effectively, and boosts energy efficiency significantly compared with static and semi-static strategies. The study also explores physical layer security and green communications, offering insights into future RIS material design. These results advance the development of efficient, robust, and scalable intelligent wireless networks.