Secure Transmission Design of RIS Enabled UAV Communication Networks Exploiting Deep Reinforcement Learning

Reconfigurable intelligent surface (RIS) enables performance enhancement of communication networks due to its significant reflective gain, which could also be used to strengthen secure transmission. In this paper, we consider a RIS enabled unmanned aerial vehicle (UAV) communication network, in which the UAV serves as a base station for a terrestrial legitimate user accompanied by a terrestrial eavesdropper. Specifically, security performance of the considered network is enhanced under both instantaneous and statistical eavesdropper channel state information (CSI). To maximize the average secrecy rate and minimize the secrecy outage duration, the trajectory and beamformer of UAV and the phase shift matrix of RIS are jointly optimized. For the instantaneous CSI scenario, we propose a successive convex approximation (SCA) based algorithm as a baseline solution, where the collective optimization problem is decomposed into three subproblems and solved separately. After that, benefiting from the state-of-the-art proximal policy optimization (PPO) algorithm, a deep reinforcement learning (DRL) based algorithm is developed to solve the average secrecy rate maximization problem from a holistic perspective. For the statistical CSI case, the close-form expression of secrecy outage probability (SOP) is derived to expound the secrecy outage duration. Since the secrecy outage duration minimization problem is hard to be solved via the SCA based algorithm, the developed DRL based algorithm is utilized again. Simulation results demonstrate that compared with benchmark algorithms, both the security performance and complexity of the proposed DRL based algorithm are competitive.