Robust secure transmission for STAR-RIS assisted ISAC against collaborative eavesdropper with imperfect CSI

Abstract

Integrated sensing and communication (ISAC), especially when assisted by simultaneously transmitting and reflecting reconfigurable intelligent surfaces (STAR-RIS), is a key technology for improving spectrum efficiency in future sixth-generation (6 G) networks. However, the inherent open and broadcast nature of wireless channels makes such systems vulnerable to eavesdropping by malicious users. Existing secure transmission schemes often overlook the severe threats posed by cooperative eavesdroppers (Eve), and many studies commonly assume that the channel state information (CSI) of Eve is perfectly known, which is unrealistic in practice. This paper investigates the problem of physical layer security (PLS) transmission in a downlink ISAC system assisted by a STAR-RIS. We consider a challenging scenario in which the dual regions of the STAR-RIS serve a sensing target (ST) and communication users, respectively, while facing the worst-case security threat caused by cooperative Eves on both sides. In addition, due to the practical difficulty in obtaining perfect CSI of the Eves, we formulate a secrecy rate maximization problem under imperfect CSI, while considering both the quality of service (QoS) requirement of communication users and the sensing performance constraint. To solve this highly non-convex optimization problem, a deep reinforcement learning (DRL) method based on the soft actor–critic (SAC) algorithm is developed, which can dynamically search for the optimal strategy for secure transmission. Numerical results illustrate that the proposed scheme keeps a good balance between sensing, communication, and security performance. Moreover, it exhibits strong robustness against cooperative Eves and uncertain CSI.