GNSS interference detection and classification with real-field Jammertest 2024 data and neural networks

Intentional interferences such as spoofing and jamming pose active threats to Global Navigation Satellite Systems (GNSS). These disruptions can compromise GNSS functionality and lead to critical failures or reduced safety in numerous location-based services. Finding efficient solutions to cope with such interferences typically requires access to realistic interference data, yet the transmission of wireless spoofing and jamming signals is forbidden in many countries. The annual Jammertest campaigns are among the very few legally authorized exceptions in Europe and are by far the largest campaigns in the world for acquiring wireless GNSS interference data. These campaigns have already been running for four years, but public access to the Jammertest datasets is still very limited, due to various intellectual property constraints, and there are still very few papers in the current literature proposing interference detection based on Jammertest datasets. In particular, we propose neural network-based algorithms to detect interferences and to classify the jamming, spoofing, and clean signals from Jammertest 2024 data. Our paper shows, for the first time in the literature, the achievable interference detection and classification accuracies with realistic Jammertest 2024 jamming and spoofing, when the testing datasets are completely different from the training datasets, namely in a cross-testing manner. We are deeply analyzing and comparing seven neural network algorithms across single and multiple GNSS bands, and we reach three-class classification accuracies of up to 84.6%.