Machine Learning-Based Detection of Spoofing Attacks in GNSS: A Study Using TEXBAT Dataset

Open service signals from Global Navigation Satellite Systems are vulnerable to interference. Spoofing, being intentional interference, is a serious threat since the receiver is oblivious to altered signals and would subsequently generate incorrect positioning solutions. Spoofing is traditionally detected through statistical tests on received signal power and correlation distortion. However, as the spoofing strategies are evolving, the testing techniques should learn and adapt as well. Therefore, this research proposes a scheme to detect spoofing via machine learning using k nearest neighbor (kNN) algorithm with optimal k selection. The proposed method is tested using two recorded spoofing scenarios from Texas Spoofing Test Battery (TEXBAT) dataset compiled for evaluating civil GPS signal authentication techniques. Carrier-to-noise ratio (C/N
0
) and delay-lock loop power extracted after tracking are used as test metrics for kNN. Training identifies distinguishing features of real and spoofed signals, whereas channel-wise testing validates that the system can effectively differentiate between real and spoofed signals with an accuracy of 94. 5%. Comparison with the prevalent Support Vector Machine (SVM) classification shows that an optimum value of k in the kNN model can provide similar accuracy while utilizing fewer features. This highlights the effectiveness of proposed scheme for spoofing detection.