COSMIC-2 RFI Prediction Model Based on CNN-BiLSTM-Attention for Interference Detection and Location

Area of Science:

• Satellite Systems Engineering
• Signal Processing
• Machine Learning Applications

Background:

• Expanding Global Navigation Satellite System (GNSS) applications necessitate robust interference detection for stability and safety.
• Radio-frequency interference (RFI) degrades GNSS signal quality and can cause system failure.
• Low Earth Orbit (LEO) satellites offer unique advantages for GNSS interference monitoring.

Purpose of the Study:

• To develop a method for predicting RFI measurements using signal-to-noise ratio (SNR) correlation variations in GNSS signals.
• To enable detection and localization of terrestrial GNSS interference signals using LEO satellite data.
• To assess the performance of a novel deep learning model against traditional methods.

Main Methods:

• Utilized signal-to-noise ratio (SNR) and radio-frequency interference (RFI) data from COSMIC-2 satellites.
• Developed a CNN-BiLSTM-Attention deep learning model to process multi-channel GNSS SNR time series.
• The model predicts the maximum RFI measurement based on correlated SNR variations across different GNSS signal channels.

Main Results:

• The proposed CNN-BiLSTM-Attention model achieved a Root Mean Square Error (RMSE) of 1.0185 and Mean Absolute Error (MAE) of 1.8567 in RFI prediction.
• The model demonstrated a high correlation coefficient (R²) of 0.9693, indicating superior RFI detection accuracy.
• The model exhibits effective rough localization capabilities for civil terrestrial GNSS interference signals.

Conclusions:

• The developed deep learning model significantly outperforms traditional methods in RFI prediction accuracy.
• The model's ability to leverage SNR correlation variations makes it suitable for future GNSS-Refractivity Observation (GNSS-RO) missions.
• This approach enhances the reliability and security of GNSS applications against interference.