Secrecy Rate Maximization for STAR-RIS-UAV-IoT with Imperfect CSI via DRL

Abstract

The simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) can effectively reshape wireless propagation paths, not only alleviating the severe performance degradation caused by blockages in the direct link but also significantly extending coverage. In addition, the high mobility and deployment flexibility of unmanned aerial vehicle (UAV) enable STAR-RIS-equipped UAV to dynamically establish high-quality communication links. In modern intelligent Internet of Things (loT) environments, where wireless connectivity is ubiquitous, ensuring secure data transmission is of critical importance. This paper investigates the maximization of the downlink secrecy rate in an STAR-RIS-UAV-loT system under imperfect channel state information (CSI). Due to the non-convexity and coupling among the optimization variables, solving such optimization problems is highly challenging. To overcome these challenges, we propose an improved deep deterministic policy gradient (DDPG) algorithm incorporating convolutional layers and the prioritized experience replay (PER) mechanism, denoted as DDPG-CP, which jointly optimizes the transmit beamforming matrix, the STAR-RIS phase shift matrices, and the UAV's horizontal position. Simulation results demonstrate that, in the presence of eavesdroppers, the proposed DDPG-CP algorithm can continuously improve the legitimate devices' communication performance while effectively suppressing the eavesdroppers' interception capability, thereby substantially enhancing the overall secrecy rate.