MADDPG-Driven Uplink Secrecy Optimization for Multi-User MIMO-NOMA Networks

Non-orthogonal multiple access (NOMA) networks integrated with multiple-input multiple-output (MIMO) systems can significantly improve spectral efficiency and user connectivity. However, securing uplink data transmission in a MIMO-NOMA system remains challenging, especially when an eavesdropper has direct access to the uplink data. This paper proposes a user-centric artificial noise (AN)-aided uplink secrecy design for a MIMO-NOMA system, where multiple users communicate with a base station in the presence of a passive eavesdropper. Unlike prior works that either focus only on downlink NOMA or overlook security integration, this study combines AN, user pairing based on channel strength, and imperfect successive interference cancellation to enhance uplink secrecy. Singular value decomposition is utilized to nullify the impact of AN at the base station. After formulating the secrecy optimization problem incorporating AN, we propose a power allocation solution based on Multi-Agent Deep Deterministic Policy Gradient (MADDPG). By employing centralized training with decentralized execution, the model dynamically allocates power to users aiming to maximize the sum secrecy rate. Simulations, conducted under the challenging worst-case security scenario, demonstrate that the proposed design consistently achieves positive secrecy rates and significantly outperforms the benchmark methods.