Open-Set Jamming Pattern Recognition via Generated Unknown Jamming Data
Guoqiang Wang, Yulong Gao
Abstract
Jamming pattern recognition (JPR) is extensively investigated as a crucial aspect of anti-jamming in wireless communication. However, with the emergence of unknown malicious jamming without training data, there is a growing demand for JPR methods that can effectively recognize both known and unknown jamming patterns. For this issue, we propose a generative open-set JPR framework, Jamming Classifier Generative Adversarial Network (JCGAN). JCGAN converts the open-set JPR problem into a closed-set JPR problem by generating valid fake unknown jamming data. Furthermore, by introducing the triplet loss, JCGAN effectively suppresses the influence of the communication signal overlapping with jamming in both time and frequency domains on JPR. Simulation studies validate these benefits as well as the efficacy of JCGAN in open-set JPR.