GNSS Spoofing Detection Under Missing Sensor Data: A Robust Learning Framework

To address the security challenges posed by global navigation satellite system (GNSS) spoofing in real-world autonomous vehicle (AV) operations, this study proposes a robust learning framework for spoofing detection, specifically optimized for missing sensor data scenarios. The framework consists of two core components: the motion state estimation module and the GNSS spoofing detection module. The motion state estimation module estimates the AV’s displacement and turning angle between consecutive timestamps through sensor data fusion. To enhance the model’s robustness in scenarios with missing data, a self-supervised learning (SSL) framework based on iTransformer is designed within this module, employing a masked data reconstruction strategy for pre-training, which improves the model’s capability to handle missing data. The GNSS spoofing detection module performs dead reckoning based on the estimated displacement and turning angle. Efficient spoofing detection is achieved through consistency verification with the reported GNSS coordinates within a sliding time window. By appropriately adjusting the size of the sliding time window, a balance can be achieved between controlling cumulative errors and maintaining sensitivity to spoofing. Based on a real-world driving dataset, the experimental results demonstrate that the framework outperforms baseline models in handling both complete and missing sensor data, and can efficiently detect both abrupt spoofing and slowly-varying spoofing.