Multi-agent deep reinforcement learning for RIS-assisted secure UAV communication

Abstract

Abstract Unmanned aerial vehicles (UAVs) serve as aerial access points or base stations, providing network services for IoT devices in urban environments with dense infrastructure. However, the line-of-sight nature of air-ground communication links renders signal transmission vulnerable to interference from ground-based eavesdroppers. To address this challenge, we propose a UAV-ground secure communication system assisted by a reconfigurable intelligent surface (RIS). The RIS is deployed on building facades to enhance the transmission environment and improve the received signal quality for legitimate users. Our goal is to maximize the sum secrecy rate (SSR) of the communication system by jointly optimizing the UAV flight trajectory, active and passive beamforming, and transmission power. Additionally, the impact of system energy consumption is considered in the optimization process to improve secrecy energy efficiency (SEE). To achieve this, we develop a multi-agent twin delayed deep deterministic policy gradient algorithm with energy penalty (MATD3-EP), where multiple agents dynamically solve subproblems and share rewards to achieve global optimization. Simulation results demonstrate that the proposed algorithm significantly outperforms benchmark strategies in maximizing both average SSR and SEE.