Optimizing Hybrid RIS-Aided ISAC Systems in V2X Networks: A Deep Reinforcement Learning Method for Anti-Eavesdropping Techniques

Abstract

Physical layer security (PLS) technique is expected to play a crucial part in the vehicle-to-everything (V2X) networks, by offering secure transmission to protect confidential information from potential eavesdropper. Considering a hybrid active-passive reconfigurable intelligent surfaces (RISs)-enhanced integrated sensing and communication (ISAC) system, this paper proposes a novel secure scheme for transmitting confidential information and performing radar sensing, where vehicle-to-vehicle (V2V) links share the spectrum resource preoccupied by vehicle-to-infrastructure (V2I) links. We aim to optimize the sum secrecy rate of V2I links by jointly designing the transmit beamforming of RSU, the radio spectrum reuse scheme of V2X links, and active and passive reflection beamforming of hybrid RIS. With above optimization, the proposed approach can enhance secure communication performance of V2I links while guaranteeing the communication quality of V2V links and target sensing capacity of RSU. Since the system model is dynamic, and it is difficult to handle the nonconvex problem, an efficient hierarchical twin delayed deep deterministic policy gradient (HTD3) method is developed to learn the secure beamforming and spectrum sharing strategies against potential eavesdropping. The proposed method decomposes the spectrum allocation into the deep Q-network procedure and designs the secure beamforming variables by employing the TD3 algorithm. Numerical results exhibit that given a sufficient power budget of hybrid RIS, our HTD3-based method enhances both the secure communication performance of V2I links and radar detection capability of RSU compared with the existing learning methods.