|Abstract:||Convolutional neural networks (ConvNets) have been used to provide superhuman accuracy on image classification tasks. These same techniques can be applied to classify radio frequency signals and by extension detect and classify global navigation satellite system (GNSS) spoofing signals. This experiment demonstrates the application of current ConvNet-based machine learning design principles combined with novel truncated singular value decomposition (SVD) to provide detection and classification of GNSS signals and spoofers. It can run in real-time on a desktop class computer and can use software defined radios for collection or an appropriate in-phase/quadrature-phase source. This provides a low-cost and low-complexity method of detecting GNSS spoofing. It also generalizes to detection of other types of radio frequency interference (RFI). This experiment utilizes the TEXBAT GPS, and OAKBAT GPS and Galileo datasets to demonstrate supervised model classification performance. It also provides results from adapting the same type of classification model to an autoencoder-based unsupervised anomaly detection system that can be conducted in real-time on a system without training data.|
Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022)
September 19 - 23, 2022
Hyatt Regency Denver
|Pages:||1137 - 1150|
|Cite this article:||
Maynard, Logan L., "GNSS Spoofing Detection Using Machine Learning and Truncated Singular Value Decomposition," Proceedings of the 35th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2022), Denver, Colorado, September 2022, pp. 1137-1150.
ION Members/Non-Members: 1 Download Credit