A Physics-Informed Multi-Correlator System for GNSS Abnormal Detection

Abstract

Global Navigation Satellite System (GNSS) signals are vulnerable to spoofing and meaconing attacks that threaten safety-critical applications including aviation, maritime navigation, and autonomous vehicles. Existing signal-level detection methods predominantly rely on labelled examples of real attack patterns for training, creating a dependency on prior knowledge of faulty signal behaviour that may not be available in practice. This paper proposes a physics-informed multi-correlator system for GNSS abnormal detection that does not require comprehensive prior knowledge of real attack signatures. At its core, a Physics-Informed Neural Network (PINN) learns the clean correlation physics from real practical recordings, whose genuine receiver noise and multipath make this a more demanding and more credible test than simulation-only studies, augmented with three physics-derived synthetic distortion archetypes and a minority subset of real attack epochs; the large majority of real Meaconing and Untargeted Spoofing attack observations is reserved strictly for validation. Encoding the triangular physical law into the network training loss distinguishes benign imperfections from genuine attack-induced distortions and suppresses false alarms. Per-channel scores are computed across up to 13 simultaneous channels spanning GPS L1 and Galileo E1B and fused through a physics-weighted consensus, with a Cumulative Sum sequential detector accumulating evidence before raising an alarm. The system achieves an area under the ROC curve of 0.9997 with the multi-channel consensus, compared with 0.9816 for a single-channel baseline, confirming effective abnormal detection driven primarily by physics-derived synthetic archetypes rather than comprehensive real attack examples.