Exploiting Correlation Among GPS Signals to Detect GPS Spoofing in Power Grids

Global positioning systems (GPS) are the major source to provide time synchronization information for phasor measurement units (PMUs) used in modern wide-area measurement systems (WAMS). GPS signals are typically sent from satellites thousands of kilometers away in space without any encryption. A cybercriminal may capture these signals to carry out GPS spoofing attacks on WAMS by maliciously desynchronizing the PMUs. Although modern power grids rely on accurate time synchronization, most of the existing works on GPS spoofing focus on location spoofing attacks. To solve this issue, we propose a light-weight GPS time spoofing detection technique based on the autocorrelation among consecutive GPS signals. The statistical runs test is used to generate training data for the classifier. The power spectral density values of multiple received GPS signals are combined to form a window. Then the runs test is used to obtain a p-value to quantify the correlation between the present and previous window. These p-values are used with supervised learning techniques to classify GPS signals as spoofed or safe. Using an experimental setup with actual GPS equipment, data are generated to train and test the proposed technique. The results show that the proposed technique can effectively detect GPS time spoofing attacks with a probability of detection above 95% while maintaining negligible probability of false alarm and probability of missed detection.