Noise-Like Features-Assisted GNSS Spoofing Detection Based on Convolutional Autoencoder

The global navigation satellite system (GNSS) is susceptible to spoofing, limiting its uses for national security. As a solution to this problem, this article modeled the GNSS spoofing detection (GNSS-SD) problem as a one-class novelty detection (OND) problem and developed a framework with high accuracy and rapid response for GNSS-SD in the postcorrelation domain. Specifically, based on the radio frequency fingerprints (RFFs) of transmitters, this framework utilized three steps to detect spoofing: noise-like feature construction, anomaly score calculation, and adaptive threshold decision-making. To verify the effectiveness of the proposed framework, we performed three comparative experiments using the University of Texas dataset. Compared with traditional signal quality monitoring (SQM) methods, the proposed framework achieved an accuracy of over 99.7%, representing a 14% improvement in complex spoofing scenarios. Besides, it won an accuracy exceeding 4% of the artificial intelligence methods in the existing literature under the same GNSS spoofing scenarios. In addition, in dynamic scenarios with multipath effects, the 93.2% accuracy demonstrated the generality of the proposed framework. Furthermore, the experiments revealed that 96.7% of spoofing can be successfully detected by the framework within a 30-s timeframe.