Deep Reinforcement Learning-Based Secrecy Rate Optimization for Simultaneously Transmitting and Reflecting Reconfigurable Intelligent Surface-Assisted Unmanned Aerial Vehicle-Integrated Sensing and Communication Systems

This study investigates security issues in a scenario involving a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)-assisted unmanned aerial vehicle (UAV) with integrated sensing and communication (ISAC) functionality (UAV-ISAC). In this scenario, both legitimate users and eavesdropping users are present, which makes security a crucial concern. Our research goal is to extend the system's coverage and improve its flexibility through the introduction of STAR-RIS, while ensuring secure transmission rates. To achieve this, we propose a secure transmission scheme through jointly optimizing the UAV-ISAC trajectory, transmit beamforming, and the phase and amplitude adjustments of the STAR-RIS reflective elements. The approach seeks to maximize the average secrecy rate while satisfying communication and sensing performance standards and transmission security constraints. As the considered problem involves coupled variables and is non-convex, it is difficult to solve using traditional optimization methods. To address this issue, we adopt a multi-agent deep reinforcement learning (MADRL) approach, which allows agents to interact with the environment to learn optimal strategies, effectively dealing with complex environments. The simulation results demonstrate that the proposed scheme significantly enhances the system's average secrecy rate while satisfying communication, sensing, and security constraints.