Deep Learning Soft-Decision GNSS Multipath Detection and Mitigation.

Area of Science:

• Signal Processing
• Machine Learning
• Geomatics Engineering

Background:

• Multipath effects in Global Navigation Satellite Signal (GNSS) signals degrade positioning accuracy.
• Existing methods for multipath detection often lack robustness across various signal conditions.
• Convolutional Neural Networks (CNNs) show promise for complex signal analysis.

Purpose of the Study:

• To develop and validate a CNN-based technique for detecting multipath interference in GNSS signals.
• To evaluate the performance enhancement achieved by using frequency-domain data pre-processing.
• To propose navigation solution strategies based on CNN outputs for mitigating multipath effects.

Main Methods:

• A CNN was designed and trained using a synthetic dataset of noisy correlator outputs across various Doppler frequencies and code delays.
• The dataset simulated multipath-disturbed signals based on an established multipath model.
• A two-dimensional Discrete Fourier Transform was applied for frequency-domain pre-processing of the correlator outputs.

Main Results:

• The CNN demonstrated effective detection of multipath effects across a range of Carrier-to-Noise density (C/N0) values.
• Pre-processing the data in the frequency domain significantly improved the CNN's detection accuracy compared to time-domain processing.
• Two distinct strategies, hard and soft decision, were proposed for navigation equation solving using CNN outputs.

Conclusions:

• The proposed CNN technique, particularly with frequency-domain pre-processing, offers a robust method for multipath detection in GNSS.
• The developed approach can be integrated into navigation systems to either reject or mitigate the impact of multipath signals.
• This research contributes to enhancing the reliability and accuracy of GNSS positioning in challenging environments.