Deep Learning-Enabled Deceptive Jammer Detection for Low Probability of Intercept Communications
Hamda Bouzabia, Tri Nhu, Georges Kaddoum
Abstract
The detection of deceptive jamming in low probability of intercept/low probability of detection (LPI/LPD) tactical communications has been receiving increasing attention, especially in the context of radar communications. In this article, we propose a novel deep-learning-based algorithm, named two-stage deceptive jammer detection (TDJD), for pseudonoise (PN) code-integrated linear frequency modulation chirp jamming signals in radar communications. Specifically, the proposed algorithm uses the time–frequency distribution (TFD) and in-phase and quadrature (I/Q) representations of the received signals to obtain more discriminating features for the detection of unknown deceptive jammers under the realistic assumption that information about the deceptive jammer, such as the PN-code signal, power, and range, is not available. In particular, the received radar signals are converted into two kinds of TFD representations using the Choi–William distribution (CWD) and I/Q representations. Next, a two-stage approach based on deep learning-enabled outlier detection and automatic modulation classification techniques is applied to the CWD and I/Q representations to extract high-level features for jammer detection. Simulation results show that the TDJD algorithm outperforms other relevant deep learning algorithms in terms of the probability of detection and the probability of false alarm. In addition, the TDJD algorithm provides accurate detection at a low signal-to-jammer-plus-noise ratio.