DRL Approach for Spectral-Energy Trade-off in RIS-assisted Full-duplex Multi-user MIMO Systems

Reconfigurable intelligent surface (RIS) is a break-through technology that enhances both energy efficiency (EE) and spectrum efficiency (SE) by artificial reconfiguration of the electromagnetic waves utilizing the reflective property of the metasurface elements. This work studies the optimization of the SE-EE trade-off using the deep reinforcement learning (DRL) algorithm in a RIS-assisted full-duplex multi-user multiple-input multiple-output (MIMO) communication system. We use partial channel state information to control the overhead signaling requirement and demand for energy supply to the system. We consider resource efficiency (RE), in which the RIS’s phase-shift design and power allocation at the nodes (i.e., node in BS in downlink (DL) and user in uplink (UL)) are jointly optimized, with the goal of investigating the SE-EE trade-off of the considered system using an appropriate performance metric. We adopt a DRL-based approach for the proposed system to tackle the challenges involved in optimization due to time-varying channels and exploitation in real-time applications. Additionally, simulation outcomes exemplify the efficiency and swift conver-gence rate of the proposed algorithm and demonstrate how different system characteristics, including co-channel interference (CCI), residual self-interference (RSI), and the number of RIS reflecting elements, affect the system’s performance.