Stealth Signals: Multi-Discriminator GANs for Covert Communications Against Diverse Wardens

Covert wireless communications are critical for concealing the existence of any transmission from adversarial wardens, particularly in complex environments with multiple heterogeneous detectors. This paper proposes a novel adversarial AI framework leveraging a multi-discriminator Generative Adversarial Network (MD-GAN) to design signals that evade detection by diverse wardens, while ensuring reliable decoding by the intended receiver. The transmitter is modeled as a generator that produces noise-like signals, while every warden is modeled as an individual discriminator, suggesting varied channel conditions and detection techniques. Unlike traditional methods like spread spectrum or single-discriminator GANs, our approach addresses multi-warden scenarios with moving receiver and wardens, which enhances robustness in a realistic threat model in urban, tactical, and future 6G settings. The proposed GAN achieves lower detection probabilities across up to five wardens and better bit error rate (BER) performance at Bob, outperforming noise-injection and single-discriminator baselines, according to extensive simulations. The framework's scalability and adaptability make it a promising basis for next-generation covert and secure wireless systems. Future directions include real-time deployment and integration with 6G-enabling technologies such as intelligent reflecting surfaces (IRS) and dynamic spectrum access.