Robust design of a machine learning-based GNSS NLOS detector with multi-frequency features.

Area of Science:

• * Navigation Systems
• * Machine Learning
• * Signal Processing

Background:

• * Global Navigation Satellite System (GNSS) non-line-of-sight (NLOS) signals pose significant risks to safe navigation due to positioning errors.
• * Urban environments present complex signal conditions challenging accurate GNSS signal classification.
• * Machine learning (ML) offers potential for classifying GNSS line-of-sight (LOS)/NLOS signals, but generalization remains a challenge.

Purpose of the Study:

• * To develop robust ML algorithms for accurate GNSS LOS/NLOS signal detection.
• * To enhance the generalization capability of ML algorithms across various scenarios and receiver configurations.
• * To improve GNSS positioning safety by mitigating errors caused by NLOS signals.

Main Methods:

• * Feature pre-normalization using open-sky models to improve algorithm generalization.
• * Design of a branched (parallel) ML process to utilize multi-frequency GNSS measurements.
• * Application of logistic regression for binary LOS/NLOS decision and probability estimation.

Main Results:

• * Proposed method achieves 90% detection accuracy in evaluated validation scenarios.
• * Branched logistic regression with pre-normalized multi-frequency features outperforms existing state-of-the-art algorithms.
• * The approach effectively handles intermittent GNSS features across different frequencies.

Conclusions:

• * The developed ML approach significantly improves GNSS NLOS signal detection accuracy and reliability.
• * Pre-normalization and multi-frequency processing enhance algorithm generalization and performance.
• * This method offers a pathway to more dependable navigation systems in challenging environments.