ML Hybrid Model-Based Anomaly Detection for GNSS Signal Quality Monitoring

Dayeon Lee, Euiho Kim*

Signal Quality Monitoring (SQM) is a critical technology for ensuring the integrity of Global Navigation Satellite System (GNSS) signals by detecting distortions, such as Evil Waveforms (EWFs), to prevent significant positioning errors. In aviation augmentation systems like LAAS and WAAS, where high reliability is mandatory, SQM serves as an essential safety component. The conventional SQM2b technique primarily relies on heuristic methods that monitor ratios between specific correlation points. However, these static approaches are vulnerable to environmental noise fluctuations due to their fixed-threshold settings and exhibit limited discriminative power against complex distortion models. To address these limitations, this study proposes a hybrid detection framework that integrates unsupervised feature extraction using an Autoencoder (AE) with a supervised classification model using XGBoost. By encoding the overall geometric profile of the correlation function into latent variables, the proposed method effectively captures subtle waveform variations that are often overlooked by the localized metrics of SQM2b. Experimental evaluations confirm that the proposed model provides more sophisticated distortion discrimination than conventional techniques. Furthermore, by leveraging multi-dimensional features derived from the correlation function, the framework demonstrates robust detection performance that is more resilient to environmental changes than fixed-threshold methods.

Keywords: signal quality monitoring, autoencoder, XGBoost, machine learning, GNSS

Speaker 이다연 홍익대학교