Towards Simple Machine Learning Baselines for GNSS RFI Detection

TL;DR

This work questions the default move toward complex deep learning models for GNSS RFI detection, arguing that simple, well-engineered baselines can achieve equal or superior performance on real-world data. Using a large, real-world HRRF dataset from Swiss operators, the authors design an interpretable baseline based on engineered features from C/No and motion signals, and compare logistic regression and LightGBM against several deep learning architectures. The results show the simple linear baseline achieving strong ROC AUC, with the non-linear baseline outperforming many deep models, challenging the assumption that increased model complexity yields better performance in GNSS RFI detection. The study underscores the value of pragmatic, transparent benchmarks and real-world data to drive meaningful progress in critical avionics applications.

Abstract

Machine learning research in GNSS radio frequency interference (RFI) detection often lacks a clear empirical justification for the choice of deep learning architectures over simpler machine learning approaches. In this work, we argue for a change in research direction-from developing ever more complex deep learning models to carefully assessing their real-world effectiveness in comparison to interpretable and lightweight machine learning baselines. Our findings reveal that state-of-the-art deep learning models frequently fail to outperform simple, well-engineered machine learning methods in the context of GNSS RFI detection. Leveraging a unique large-scale dataset collected by the Swiss Air Force and Swiss Air-Rescue (Rega), and preprocessed by Swiss Air Navigation Services Ltd. (Skyguide), we demonstrate that a simple baseline model achieves 91\% accuracy in detecting GNSS RFI, outperforming more complex deep learning counterparts. These results highlight the effectiveness of pragmatic solutions and offer valuable insights to guide future research in this critical application domain.

Figures (7)

• Figure 1: RFI on a Stationary GNSS Receiver.
• Figure 2: Helicopter Maneuver Affecting GNSS Satellite's C/No.
• Figure 3: Roll and Pitch Angles Affecting the Carrier to Noise Ratio Shown in Figure \ref{['fig:2']}.
• Figure 4: Situation where a potential RFI might be present. The mean value of the normalized C/No decreases by up to 10 dB-Hz. The standard deviation remains similar compared to situations without RFI.
• Figure 5: EC145 of the HEMS operator REGA. The GPS antenna is installed on the top of the fin in front of the strobe light (Courtesy REGA).