SigFeaDet: Signal Features-based UAV GPS Spoofing Detection using Machine Learning

The GPS is the most common used satellite-based navigation and positioning system. It is an indispensable component for a UAV as it provides accurate location data that is critical for navigation and mission success. However, GPS spoofing attacks are becoming a growing threat to GPS-dependent systems. Most existing GPS spoofing detection methods for UAVs are proposed based on simulation data or depend on multiple UAVs. In this paper, we propose a signal feature-based GPS spoofing detection approach, SigFeaDet, for a UAV using machine learning techniques. Our basic idea is to use the anomalies of GPS signal features due to the discrepancy between normal and spoofing signals. The Carrier-to-Noise Density Ratio (CN0) and Doppler frequency are significant signal features for GPS satellite positioning, which can be applied to identify spoofing signals. Various machine learning algorithms are applied to train with normal and spoofing data to choose the best classifier. The open GPS dataset TEXBAT is processed to obtain signal data and we also perform flight experiments to collect GPS data to augment normal GPS signal dataset. The detection rate is more than 94.8% and the equal error rate (EER) is about 5%, and the detection period is only 0.4 second.