Incremental learning-enhanced LSTM for spoofing detection in GNSS signals

With the widespread application of Global Navigation Satellite Systems (GNSS), spoofing interference has emerged as a major threat to system security due to its high level of concealment. Traditional static learning models face performance degradation issues in long temporal sequences, making it challenging to meet detection needs in dynamic scenarios. To address this, this paper proposes a Long Short-Term Memory (LSTM) model based on an incremental learning mechanism that dynamically integrates multidimensional features, combined with a hierarchical learning rate strategy and ring-buffer mechanism, for detecting spoofing interference in streaming GNSS signals. Furthermore, to improve robustness under non-stationary and dynamic conditions, the proposed framework introduces a residual standardization based on the Median Absolute Deviation (MAD), a Hampel dual-threshold decision mechanism, and a Huber loss function to suppress the influence of outliers. Experimental validation shows that the proposed method continuously optimizes model parameters through incremental learning, effectively overcoming the extrapolation failure of traditional static LSTM models in ultra-long sequence scenarios. The detection response time is reduced to the millisecond level (7.93 ms), with an accuracy exceeding 98%. Additionally, validation using both simulated Doppler signals and the TEXBAT dataset demonstrates that the method not only achieves high accuracy but also exhibits strong robustness in complex dynamic spoofing scenarios, providing a lightweight and efficient solution for GNSS anti-interference technology.