Joint Optimization of Power Control and Receive Beamforming for Security Enhancement in SGF–NOMA Systems: A Distributed MADRL Approach

To enhance the physical layer security in semi-grant-free non-orthogonal multiple access systems, this work develops a strategy for simultaneous transmission by grant-free devices (GFDs). The proposed strategy utilizes the interference of active GFDs as coverts to safeguard the legitimate transmision of the grant-based device from the potential eavesdropping of inactive GFDs. By utilizing the centralized teacher with decentralized student framework, we develop a joint power control and receive beamforming scheme employing multi-agent deep reinforcement learning, which requires only the partial observation and implements in a distributed manner. Numerical results show that the proposed strategy surpasses both non-orthogonal multiple access and orthogonal multiple access benchmarks and achieves the secrecy performance close to its global observation counterpart.