Frequency Hopping Signal Detection in Low Signal-to-Noise Ratio Regimes

The detection of unauthorized frequency hopping (FH) signals has several applications in securing the radio frequency spectrum and achieving spectrum awareness in both tactical and cyber-physical systems. However, the blind detection of adversary FH signals is a challenging task, particularly in low signal-to-noise ratio (SNR) regimes, due to the adoption of dynamic hopping patterns. In this study, we propose a cyclo-stationary signal features-based blind FH signal detection scheme to address this challenge. Our proposed scheme consists of two steps: (i) feature extraction, where cyclic features are extracted from the spectral correlation function of the signals, and (ii) feature classification, where the extracted features are associated with ON/OFF detection states using a trained support vector machine (SVM) classifier. We leverage both binary and one-class SVM classifiers to enable adversary FH signal detection with and without pre-existing signal labels. Extensive simulations are conducted to verify the efficacy of the proposed FH signal detection scheme in low SNR regimes. Simulation results also provide insights into the interplay of various system parameters, such as the numbers of cyclic features and emission bandwidth, on the detection performance of the proposed SVM classifiers.