Time-modulated intelligent reflecting surfaces for integrated sensing, communication and security with generative artificial intelligence design

Abstract

We propose a physical layer security approach for integrated sensing and communication (ISAC) systems, where targets may act as eavesdroppers. A time-modulated intelligent reflecting surface (TM-IRS) preserves data integrity at legitimate communication users (CUs) while scrambling transmissions in other directions. The TM-IRS design leverages a generative flow network (GFlowNet) to learn a stochastic policy for sampling high-performing configurations from a large discrete parameter space. We formulate the achievable sum rate for legitimate CUs and beam pattern gain toward the target, constructing rewards that jointly capture communication and sensing performance. The TM-IRS design is modeled as a deterministic Markov decision process, where each terminal state represents a complete configuration. GFlowNets, parameterized by deep neural networks, sample configurations with probability proportional to reward. Experimental results demonstrate effective sensing, communication, and security integration, with significantly higher sampling efficiency than exhaustive search and improved robustness over existing benchmarks.