Abstract In large-scale Internet of Things (IoT) networks, detecting covert communications—hidden transmissions that evade monitoring—is essential to prevent misuse of wireless infrastructure. However, challenges such as fading, noise, jamming interference, and unpredictable traffic complicate reliable detection by a monitoring node (Willie). This paper introduces a hybrid analytical-machine learning (ML) framework for robust detection of covert signals under jamming and Rayleigh fading conditions. An analytical energy detection model is first derived to compute false alarm and missed detection probabilities, establishing a baseline and informing Monte Carlo simulations for dataset generation. The real and imaginary components of simulated complex baseband signals are extracted as features, enabling supervised training of Decision Tree (DT) and Random Forest (RF) classifiers without prior knowledge of transmit powers. Evaluations demonstrate that both ML models outperform the analytical benchmark, with RF achieving a 26.8% reduction in total detection error at a distance of d = 1 km. Further assessments across varying transmitter and jammer power levels confirm the framework’s robustness in interference-limited environments. By integrating theoretical modeling for data credibility with data-driven ML for adaptive classification, this approach provides a scalable, power-agnostic solution for securing real-world wireless networks against covert threats.
Esmaili et al. (Wed,) studied this question.