GNSS spoofing detection using temporal representation learning of Doppler residuals
Abstract
As a receiver-side threat to global navigation satellite systems (GNSS), spoofing can mislead a receiver to output incorrect position, velocity, and timing solutions. This paper proposes a GNSS spoofing detection method based on temporal representations of Doppler residuals. The proposed method first establishes a seven-dimensional Doppler observation model with receiver position, velocity, and clock drift as state variables. Doppler residuals are then constructed epoch by epoch to reduce the effects of satellite geometry, receiver motion, and common clock-drift components on anomaly detection. Sliding windows are used to form residual sequence samples, and a CNN-Transformer encoder is adopted to extract local residual fluctuations and long-term temporal features. Considering that normal samples are easier to obtain while spoofing samples are difficult to fully cover in practice, a one-class anomaly detection strategy is introduced. The model learns a feature center using only normal residual samples, and the anomaly score is calculated by the distance between each test embedding and the normal center. Experimental results on the public TEXBAT dataset show that the proposed method achieves detection rates of 89.97% and 92.09% in the DS2 and DS5 non-frequency-locked spoofing scenarios, respectively. Compared with Doppler-RAIM, the proposed method improves the average detection rate by 5.25% and reduces the average time to alarm by 0.7 s. These results show that the proposed method can effectively exploit temporal features of Doppler residuals for receiver-side GNSS spoofing detection without external sensors.