XAI GNSS-A Comprehensive Study on Signal Quality Assessment of GNSS Disruptions Using Explainable AI Technique.

Area of Science:

• Navigation and Positioning Systems
• Signal Processing
• Artificial Intelligence

Background:

• Global Navigation Satellite Systems (GNSS) are vulnerable to jamming and spoofing attacks, degrading signal quality.
• Effective post-processing of GNSS signals requires careful analysis under interference and multipath conditions.
• Identifying signal disruptions is crucial for maintaining GNSS receiver performance and reliability.

Purpose of the Study:

• To identify influential time and spectral domain features for analyzing GNSS signals affected by various disruptions.
• To evaluate the effectiveness of explainable AI (XAI) models in feature selection for GNSS signal analysis.
• To compare XAI-based feature selection with traditional methods for improved classification accuracy in GNSS signal prediction.

Main Methods:

• Examination of GNSS signal records under various disruption scenarios (pure, CWI, MCWI, MP, spoofing, pulse, chirp).
• Application of Machine Learning (ML) techniques to assess feature importance.
• Utilizing SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) for feature analysis and model interpretability.

Main Results:

• XAI models identified key time and spectral domain features crucial for classifying GNSS signal disruptions.
• Using selected important features with SHAP and LIME improved classification accuracy compared to traditional feature selection.
• XAI models provided clear explanations for ML model predictions based on individual feature contributions.

Conclusions:

• XAI models, particularly SHAP and LIME, offer superior feature selection for GNSS signal analysis, enhancing classification accuracy.
• These models effectively reveal the decision-making process of black-box ML models in identifying signal disruptions.
• The application of XAI facilitates fault detection and resilience diagnosis in GNSS post-processing for ground stations.