Intelligent Cooperative Covert ISAC via Fluid Antennas: A GAN-Optimized Framework

Integrated sensing and communications (ISAC) is a novel technology that utilizes communication signals to perform sensing and communication simultaneously. However, the dual functionality of ISAC signals introduces significant security challenges, including information leakage to unauthorized users and heightened vulnerability to detection by malicious adversaries. To address these issues, this paper leverages fluid antenna system (FAS) to establish covert and secure communications in a cooperative millimeter-wave (mmWave) orthogonal frequency-division multiplexing (OFDM) near-field ISAC system. In this system, multiple base stations (BSs) equipped with extremely-large FAS broadcast confidential information to users securely while simultaneously sensing multiple radar targets. Specifically, signal broadcasting can lead to information leakage, compromising information security. Furthermore, the multi-functionality of the ISAC signal simplifies detection by a malicious warden, Willie, thereby degrading communication covertness. To tackle these dual security threats, we formulate a joint optimization problem for the analog beamforming, digital precoder, FAS port activation matrix, and radar sensing signal covariance matrix. The objective is to maximize the minimum secrecy rate and the detection error probability (DEP) at Willie, subject to the constraint on the minimum sensing signal-to-interference-plus-noise ratio (SINR). This problem is highly challenging due to three main factors: (i) it is a multi-objective optimization problem with conflicting objectives; (ii) the lack of an analytical expression for the DEP makes a tractable problem formulation difficult; and (iii) the problem is non-convex, stemming from the combinatorial nature of the activation matrix design, coupled variables, and the non-convex objective functions. To overcome these challenges, we propose a novel joint design framework based on a generative adversarial network (GAN). This framework employs a discriminator network to emulate the optimal detector at Willie and a generator network to optimize the system parameters. Adversarial training is then used to optimize the system design against this optimal detector. Numerical simulations demonstrate the potential of FAS in enhancing both communication security and sensing performance, and confirm the superior performance of the proposed GAN-based joint optimization framework.