Detection and Mitigation of Jamming, Meaconing, and Spoofing based on Machine Learning and Multi-Sensor Data

Philipp Bohlig, Jorge Morán García, Ramadevi Lalgudi, Jan Fischer

Abstract:	Global navigation satellite systems (GNSS) are an integral part to the functioning of various global applications including autonomous driving, aviation, and more. However, GNSS signals are often non-authenticated and weak, making them easy to override with stronger, false signals. While Galileo Open Service Navigation Message Authentication (OSNMA) tackles one element of these weaknesses, several GNSS interference types remain a challenge to be solved. Any localization application and especially if safety-critical highly depends on GNSS and thus, needs to incorporate concepts of detecting and mitigating GNSS interference, i.e., Jamming, Meaconing, and Spoofing, to provide safe and robust position, navigation, and timing (PNT) solutions. To provide robustness against GNSS interference while maintaining a high accuracy, we propose to include a machine learning based approach which utilizes not only GNSS measurements but also additional features from unaffected sensors like an IMU. Converse to various approaches which consider only one or two features, the common processing of the features listed in the following paragraph in a recurrent neural network (RNN) reduces the risk of false alarms and missed detection, enhancing the system robustness against GNSS interference and maintaining high accuracy and availability. The recurrent approach is used to include features of multiple consecutive epochs for the detection of an interference state of a specific epoch. Including the average Carrier-to-noise ratio (C/N0) of a frequency and the corresponding Gain amplifier together as reported by the GNSS receiver, which implements Automatic Gain Control (AGC), abrupt changes due to interference can be detected and distinguished from a normal C/N0 drop based on signal occlusion. Cross-band carrier-phase divergence is another phenomenon observed during e.g., spoofing. The ratio between carrier phases from signals transmitted by the same satellite in different bands changes substantially more than under nominal conditions. Finally, the feature set is complemented by raw IMU measurements, which serve as a robust cross-checking source given their immunity to GNSS interference. The features are combined in a normalized label vector, where each sample is labeled by the affected carrier (e.g., GPS L1 or GAL E6), resulting in a multi-label classification problem. In this work, we compare the results of an individual labeling based on information of the type of interference (Jamming, Meaconing, Spoofing) and a common labeling as interference. The proposed deep learning framework for multi-label GNSS interference detection consists of a convolutional neural network (CNN) encoder combined with a long short-term memory (LSTM) network. It is shown that the AI-based detector effectively classifies the interference per GNSS constellation and signal, allowing to selectively remove only the compromised signals rather than discarding the entire solution. This targeted approach enables smarter decision-making at the GNSS Kalman filter level. One key aspect of our work is that the network will be trained and tested with real jammed and spoofed signals, leveraging on the multi-sensor data recorded at Jammertest 2024. The well labeled collection of datasets is made publicly available together with the network itself.
Published in:	Proceedings of the 38th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2025) September 8 - 12, 2025 Hilton Baltimore Inner Harbor Baltimore, Maryland
Pages:	765 - 777
Cite this article:	Bohlig, Philipp, García, Jorge Morán, Lalgudi, Ramadevi, Fischer, Jan, "Detection and Mitigation of Jamming, Meaconing, and Spoofing based on Machine Learning and Multi-Sensor Data," Proceedings of the 38th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2025), Baltimore, Maryland, September 2025, pp. 765-777. https://doi.org/10.33012/2025.20368
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In