Multidomain Joint Spoofing Detection Based on a Semi-Supervised Detection Network for GNSS-Based Train Positioning

Abstract

As the positioning technology advances, the global navigation satellite system (GNSS) has been highly concerned in railway systems. Meanwhile, the attack surface to GNSS-enabled train positioning has expanded, and thus, the concerns regarding the positioning security have increased. To deal with potential threats from a GNSS spoofing attack, research on active detection is conducted. A framework of multi-domain-feature-based spoofing detection is established, under which the semi-supervised GANomaly network is designed to generate the data-driven model. Different blocks, including the multiscale group dilated convolution, the attention gate, and double-dilated temporal convolutional network residual block, are involved to improve the reconstruction ability of the vanilla GANomaly network. A dynamic threshold mechanism is adopted in identifying the existence of the spoofing attack. The experimental results with the open-source Texas Spoofing Test Battery dataset under the ds7 static scenario and a railway train dataset from the spoofing injection test demonstrate the certain advantages of the proposed solution over the reference single-feature-based methods and existing networks, which highlight the potential for achieving the resilient train positioning under the observing conditions with respect to the adopted datasets.