Spoofing Detection of Satellite Transmitters by Exploiting Convolutional Neural Networks

Detecting spoofing and reply attacks is an age-old problem. This is primarily due to the inherent broadcast nature of radio signals that make the verification of the sender challenging. Although there has been a lot of research done in this field, the satellite scenario did not receive too much attention while many services are provided with no security, e.g., Global Positioning System. This work introduces a brand new solution to the detection of illegitimate terrestrial transmitters trying to mimic the behaviour of a satellite. Our solution exploits the intrinsic differences between the signal propagation on a satellite link and that one of the terrestrial link. Indeed the satellite channel is very different from a terrestrial one, and our solution exploits those differences to detect the presence of a terrestrial transmitters willing to spoof a satellite one. It is worth noting that the solution proposed in this thesis is independent of the transmitter and the receiver, but only it only focuses on the signal modifications that the two channels (the satellite and the terrestrial ones) introduce on the over-the-air signal. We apply state-of-the-art neural network analysis to distinguish between signals that travelled through a satellite and a terrestrial link. Our solution shows that the terrestrial link uniquely affects the transmitted signals and make them clearly distinguishable from the one received from a satellite transducer. We tested our solution by combining real satellite data with state-of-the-art terrestrial channel models and we report our results using different metrics such as accuracy, precision, and F1-score.