Satellite Navigation Anti-Spoofing Using Deep Learning on a Receiver Network

The Global Navigation Satellite Systems (GNSS) use both open-access and secure signals for terrestrial receivers to navigate, but spoofers can replicate open-access signals and cause a receiver to unknowingly lose tracking of the true signal. Under such attacks, anti-spoofing techniques are required to classify true GNSS signals from spoofed signals. Since most spoofers lack access to the codes for the secure signals, they are unable to replicate them. As such, a receiver network can use codeless processing on the secure signals to verify that the open-access signals are authentic. In this paper, deep learning is applied to classify true GNSS signals from spoofed signals using a network of receivers that does not have access to the proprietary codes used in the Galileo Public Regulated Service (PRS). Specifically, the pairwise cross-correlation of different receivers is classified to detect a spoofer who cannot replicate the PRS codes. In addition, the cyclic profiles, computed from the cross-correlation cyclostationary content of the received signals, are compared. The classification is then tested on Rayleigh faded Galileo signals. Comparing the performance of long short-term memory (LSTM) and convolutional neural networks (CNN) for a noisy GNSS signal, CNN-based neural networks are shown to classify with the highest accuracy when cyclic profiles are used as the input.