GCI-GANomaly: A Novel GPS Spoofing Detection Scheme Based on Grayscale Constellation Image

This paper addresses the spoofing detection issue for the Global Positioning System (GPS) based on radio frequency fingerprinting (RFF). We first introduce a new RFF feature in the post-despreading domain of GPS signal processing to better capture the hardware characteristics of GPS satellites. We model the new RFF feature as grayscale constellation images (GCIs) and provide in-depth analysis to show the hardware characteristics that can be captured by GCIs. Using this feature, we develop a deep-learning based spoofing detection framework named GCI-GANomaly, which applies a generative adversarial network (GAN) with a cosine latent anomaly scoring strategy for robust detection. We evaluate the detection accuracy and false positive rate (FPR) of the proposed method based on the open-source Texas Spoofing Test Battery (TEXBAT) dataset. The results showed that GCI-GANomaly improves the detection accuracy of traditional signal quality monitoring (SQM)-based methods by up to 30.8% and reduces the average FPR of existing RFF-based methods by 4.2% with much less training data while achieving slightly better average detection accuracy. We further evaluate the robustness (against spoofing power and time) of GCI-GANomaly based on a specific GPS signal dataset that we collected from the real-world constellation. The results showed that GCI-GANomaly achieves robust detection performance under varying spoofing power and shows acceptable stability as time elapses.