Deep Reinforcement Learning for Active RIS-Assisted Full-Duplex Integrated Sensing and Communication Systems

Compared to traditional passive reconfigurable intelligent surfaces (RISs), active RISs can actively amplify incident signals, enhancing signal strength, reducing path-loss, and extending coverage. In this paper, we investigate an active RIS-assisted full-duplex integrated sensing and communication (ISAC) system. By jointly designing the power allocation factor for communication user equipment (UE) and the reflecting coefficients of the active RIS, the long-term average sensing performance is maximized while satisfying the signal-to-interference-plus-noise ratio (SINR) constraints for all UEs. In this paper, we explore a deep reinforcement learning (DRL) algorithm to address the complex non-convex optimization problem, aiming to maximize long-term gains in system sensing performance. Simulation results indicate that the soft actor-critic (SAC) algorithm achieves superior performance compared to two benchmark algorithms, and that active RIS demonstrates enhanced capabilities over traditional RIS.