Intelligent Reflecting Surface Configurations for Smart Radio Using Deep Reinforcement Learning

Intelligent reflecting surface (IRS) is envisioned to change the paradigm of wireless communications from “adapting channels” “changing channels”. However, current IRS configuration schemes, consisting sub-channel estimation and passive beamforming in sequence, conform conventional model-based design philosophies are difficult be realized practically complex radio environment. To create smart environment, we propose a model-free control that independent channel state information (CSI) requires minimum interaction between communication system. We firstly model as Markov decision process (MDP) apply deep reinforcement learning (DRL) perform real-time coarse phase IRS. Then, extremum seeking (ESC) fine Finally, by updating frame structure, integrate DRL ESC improve its adaptivity different dynamics. Numerical results show superiority our proposed joint scheme verify effectiveness without CSI.