Secure Beamforming for Active RIS-Aided ISAC Systems Based on Improved TD3 Algorithm

Abstract

This paper focuses on physical-layer security against passive eavesdroppers in 6G integrated sensing and communication (ISAC) networks. It proposes an active reconfigurable intelligent surface (RIS)-aided secure-ISAC joint optimization framework, which utilizes the active RIS to overcome the multiplicative-fading bottleneck of the passive RIS and avoid the high-power leakage issue of conventional artificial-noise (AN) schemes. A optimization problem is formulated to maximize secrecy rate under constraints including radar signal-tointerference-plus-noise ratio (SINR), communication quality of service, transmit-power budget, hardware-impairment distortion, and RIS-radiation stealth. To solve the high-dimensional, nonconvex partially observable Markov decision process, an adversarial model-predictive-control based twin delayed deep deterministic policy gradient algorithm (A-MPC-TD3) is developed, combining generative adversarial network (GAN) for worstcase perturbations and lightweight MPC for real-time feasibility. Numerical results show it outperforms passive RIS and AN under CSI uncertainty.