DRL-based joint optimization for secure and efficient communication in UAV-RIS assisted 6G networks

Unmanned Aerial Vehicles (UAVs) and Reconfigurable Intelligent Surfaces (RIS) are a promising Sixth Generation (6G) wireless technology which provides enhanced coverage and connectivity in urban environments where non-line-of-sight (NLoS) propagation is common. Despite these potentials, the implementation of UAV-RIS-assisted systems is hindered by a number of unsolved issues, such as inefficient UAV flight paths, computationally demanding optimization processes, and the lack of security measures. These problems have a negative impact on key performance metrics, including secrecy rate, throughput, and energy efficiency. To deal with these issues, this paper proposes a single framework that simultaneously maximize energy use, enhances secrecy rate, and responds to changing network conditions in real time. The proposed framework uses the Block Coordinate Descent (BCD) and Alternating Direction Method of Multipliers (ADMM) to decouple interdependent variables, allowing the UAV trajectories, RIS phase configurations, and power allocation to be optimized efficiently. An Artificial Noise(AN)-assisted beamforming scheme is also included, which intentionally degrades the channel quality of the potential attackers without affecting the quality of service (QoS) of the legitimate users, thus improving the physical-layer security. Moreover, a Deep Deterministic Policy Gradient (DDPG)-based feedback loop control mechanism is implemented to constantly adjust the UAV trajectories, RIS configuration, and power distribution in real time. The overall framework is simulated in MATLAB and then compared with baseline schemes to demonstrate its effectiveness. Simulation results show that the proposed approach outperforms the baseline schemes by achieving 5.63% average improvement in secrecy rate, 29.48% in energy efficiency, and 36.58% in minimum achievable rate compared to both Deep Reinforcement Learning with Proximal Policy Optimization (DRL-PPO) and Multi-Agent Twin Delayed Deep Deterministic Policy Gradient (MATD3).