Deep temporal semi-supervised one-class classification for GNSS radio frequency interference detection

We present a deep learning approach for near real-time detection of Global Navigation Satellite System (GNSS) radio frequency interference (RFI) based on a large amount of aircraft data collected onboard from the Global Positioning System (GPS) and Attitude and Heading Reference System (AHRS). Our approach enables detection of GNSS RFI in the absence of total GPS failure, i.e. while the receiver is still able to estimate a position, which means RFI sources with low power or at larger distance can be detected. We demonstrate how deep one-class classification can be used to detect GNSS RFI. Furthermore, thanks to a unique dataset from the Swiss Air Force and Swiss Air-Rescue (Rega), preprocessed by Swiss Air Navigation Services Ltd. (Skyguide), we demonstrate application of deep learning for GNSS RFI detection on real-world large scale aircraft data containing flight recordings impacted by real jamming. The approach we present is highly general and can be used as a foundation for solving various automated decision-making problems based on different types of Communications, Navigation and Surveillance (CNS) and Air Traffic Management (ATM) streaming data. The experimental results indicate that our system successfully detects GNSS RFI with 83 ⋅ 5% accuracy. Extensive empirical studies demonstrate that the proposed method outperforms strong machine learning and rule-based baselines.