Abstract
Jamming pattern recognition (JPR) is critical in modern communication systems, including civil, and industrial applications, where complex and unpredictable jamming patterns (JPs) frequently disrupt normal operations. Accurately identifying known JPs while detecting unknown JPs is essential for maintaining robust communication. This paper tackles the jamming pattern open set recognition (JP-OSR) challenge by proposing a novel framework that integrates three key components: Spiking Wavelet Transformer (SWT), Hyperspherical Maximum Class Separation (MCS), and an energy-based model with an adaptive threshold. The framework begins by extracting high-resolution time-frequency features through the SWT, effectively capturing the intricate temporal and spectral patterns of jamming signals. These features are then aligned in a hyperspherical space using MCS to maximize class separation and improve inter-class distinctions. To address the variability in feature distributions across different Jamming-to-Signal Ratios (JSR) and JPs, the proposed energy-based model dynamically adjusts the threshold, enabling accurate recognition of both known and unknown jamming types. Simulation experiments have validated the effectiveness and superiority of the proposed method for JP-OSR.