Empirical Evaluation of GNSS Timing Spoofing and Detection Using Machine Learning

This paper presents an empirical analysis of GNSS timing spoofing detection using data from Jammertest 2024. While positioning spoofing has been extensively studied, timing spoofing presents unique threats to critical infrastructure including power grids, telecommunications, and financial systems. Our analysis reveals that position parameters become compromised at lower spoofing power levels (20-25 dBm) compared to timing parameters (30 dBm), creating potential for early warning systems. Traditional carrier-to-noise ratio methods proved unreliable, while validity flags in NMEA messages showed strong but incomplete detection capability. To address detection gaps, we implemented an unsupervised Isolation Forest algorithm achieving perfect recall (100%) and high specificity (99.96%) without requiring extensive labeled examples. A combined approach leveraging both validity flags and machine learning is able to offer robust protection with minimal computational overhead. Our findings demonstrate practical robustness measures for timing-dependent critical infrastructure and how it can be implemented.