DRL-Based Resource Optimization for Full-Duplex RIS-aided Multi-user NOMA ISAC Systems

Integrated sensing and communication (ISAC) has emerged as a cornerstone technology for beyond-5G and 6G networks, addressing the dual requirements of high-capacity communication and precise environmental sensing. This paper introduces an innovative reconfigurable intelligent surface (RIS)-assisted non-orthogonal multiple access (NOMA) ISAC system with full-duplex (FD) capabilities. The proposed framework leverages RIS to reconfigure the wireless environment dynamically, NOMA for enhanced spectral efficiency, and FD operation to simultaneously manage uplink and downlink transmissions. A non-convex optimization problem is developed to maximize the minimum beam pattern gain aimed at communication users and radar targets. To address this, a deep reinforcement learning (DRL) algorithm [1] based on proximal policy optimization (PPO) is proposed, effectively addressing the high-dimensional action space and complex constraints. Additionally, a heuristic algorithm is employed as a baseline for performance comparison. Simulation results highlight the advantages of the proposed RIS-NOMA-FD ISAC system, achieving significant improvements in spectral efficiency and radar sensing accuracy compared to half-duplex (HD) and conventional RIS-ISAC systems.