Deep Learning for GNSS Spoofing Detection: A Performance Analysis

The global navigation satellite systems (GNSS) are still dominant in the field of navigation and time keeping due to their affordability, worldwide coverage as well as their amazing accuracy. However, with open signal design and the natural low signal levels, they are susceptible to a range of both intentional and unintentional interference. Signal spoofing is a subversive and insidious type of intrusion, in which an adversary sends a victim receiver fake navigation information. With a false signal injected into the GNSS receiver, the attacker is able to deceive the receiver, and, as a result, poses a great risk due to the high efficiency and comfort with which it can be hidden. Modern anti-spoofing techniques to detect such spoofing are effective in many cases, but face some significant drawbacks: they have high false-positive probabilities, high computational complexity, and they require tuning to the continually varying properties of the received signal. This work includes a detailed analysis and comparative evaluation of various detection algorithms, and specifically neural-network designs, such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs) and Long-Short-Term Memory (LSTM) units. Finally, this study highlights the potential of machine-learning-based solutions to improve detection and reduce false positives to their lowest point possible and to successfully combat the entire repertoire of threats posed by spoofing attacks.