DRL-Based Secrecy Rate Optimization for RIS-Assisted Secure ISAC Systems

This correspondance studies the physical layer security in a reconfigurable intelligent surface (RIS) assisted integrated sensing and communication (ISAC) system which serves multiple users and tracks a target simultaneously. Specifically, we consider the radar target, as a potential eavesdropper, who eavesdrops on legitimate users information. Artificial noise (AN) is utilized to disrupt eavesdropper reception. We aim at maximizing the achievable secrecy rate of all the legitimate users by jointly designing the transmit beamforming, the AN signals and the phase-shift of RIS. Since the problem is multivariable coupling and non-convex optimization, we adopt deep reinforcement learning (DRL) algorithm to find the optimal learning strategy through agent and environment interactive learning. Numerous results verify that the DRL algorithm can achieve substantial improvement in secrecy rate compared with benchmark approaches.