Deep Reinforcement Learning for Secure Communication in RIS-Assisted UAV Systems

Unmanned Aerial Vehicles (UAVs) are integral to modern wireless communication systems, providing flexible and efficient connectivity. However, ensuring secure communication between UAVs and ground users remains a significant challenge, particularly in the presence of malicious eavesdroppers or jammers. To enhance security, Reconfigurable Intelligent Surfaces (RIS) can be deployed to dynamically reconfigure the wireless environment, improving signal strength and mitigating interference. This study focuses on the joint optimization of the UAV's 3D trajectory and RIS phase shifts to maximize the secrecy rate of the UAV-to-user link. Given the non-convex nature of this problem, traditional optimization methods, such as Successive Convex Approximation (SCA), struggle to provide timely solutions. To address this, we leverage the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm to learn near-optimal policies. The simulation results demonstrate that, in scenarios where the UAV and ground users are located in complex urban environments and experience Non-Line-of-Sight (N-LOS) conditions, the TD3-based approach significantly outperforms conventional methods, enhancing the secrecy rate of the RIS-assisted UAV system.