A two-stage physics-informed framework for open-set rogue emitter detection in edge-enabled WSNs
TL;DR
A physics-driven two-stage open-set radio frequency (RF) signal detection framework, termed OSR-SignalDet, is proposed, featuring a task decoupling paradigm, dedicated RF perception modules, and a prototype-based open-set metric learning head.
Abstract
With the integration of Wireless Sensor Networks (WSNs) and broadband communication systems, Specific Emitter Identification (SEI) has emerged as a core enabler for physical layer security against rogue node intrusion. Existing end to-end spectrogram-based detection architectures face three critical limitations in real-world scenarios: fingerprint erasure caused by 8-bit quantization, task conflict between signal localization and fine-grained feature extraction, and insufficient robustness against unknown zero-day spoofing attacks. To address these challenges, a physics-driven two-stage open-set radio frequency (RF) signal detection framework, termed OSR-SignalDet, is proposed, featuring a task decoupling paradigm, dedicated RF perception modules, and a prototype-based open-set metric learning head. Experiments conducted on the CommRad real-measurement dataset demonstrate that OSR-SignalDet achieves 95.6% closed-set authentication accuracy at -5 dB signal-to-noise ratio (SNR), a 98.4% interception rate for unknown attacks, and retains 88.7% authentication accuracy under severe Rayleigh multipath fading channels, while maintaining a total parameter count of merely 3.8M, fully satisfying the deployment constraints of WSN edge nodes.