GNSS Spoofing Attack Detection By IMU Measurements Through A Neural Network

Spoofing attacks against global navigation satellite system (GNSS) receivers are a serious threat to secure navigation, also in autonomous driving. Cars typically include, beyond the GNSS receiver, also an inertial measurement unit (IMU), whose data can be used to detect GNSS spoofing attacks. We consider a specific spoofing attack, with the spoofed trajectory that gradually diverges from the true trajectory, and we propose a spoofing detection method based on machine learning. First, a feature vector is designed, collecting the difference of two estimates of the device velocity, obtained from the GNSS receiver and the IMU. Then, a neural network (NN) is trained over a set of true and spoofed trajectories to detect the attack. We compare the proposed solution with an approximated Neyman-Pearson test and a literature reference direct comparison method, confirming the low error probabilities of our novel solution.