One-Step Network for Open-Set Jamming Detection and Parameter Estimation

Jamming Detection and Parameter Estimation (JDPE) is critical for anti-jamming in wireless communication, but traditional two-step methods struggle with unknown deceptive jamming lacking training data. We propose One-step Network for Open-set JDPE (ONOJ) which using instance-level contrastive learning. In open-set scenarios, jamming of different classes have subtle feature differences due to similar generation principles, while same-class ones can vary significantly due to parameter disparities. To address this problem, we optimize the network backbone and region proposal modules, then introduce a novel aggregation method with unique prototype selection and a dynamic margin contrastive loss designed for jamming characteristics. These enhance open-set performance by balancing intra-class compactness and inter-class separability. Simulation results indicate that ONOJ demonstrates excellent adaptability and better efficiency in the real-world open wireless communication environment against unknown jamming.