Multipath Detection in GNSS Using Differential Corrections and Machine Learning-Based Classification

This paper proposes a novel framework for detecting and mitigating multipath interference in Global Navigation Satellite Systems (GNSS), a common issue caused by reflections of signal off surfaces such as buildings and the ground. To address this challenge, the study integrates Differential GNSS (DGNSS) techniques with advanced machine learning models, such as Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Random Forest-to automatically detect and exclude satellites affected by multipath. The methodology involves synchronized GNSS data collection from a stationary base station and a mobile rover using high-precision u-blox ZED-F9P receivers with polarized antennas, DGNSS corrections via single and double differencing, and feature vector construction from both corrected and raw observation data. Signal quality labels ("Clean" or "Noise") are derived through skyplot analysis and environmental modeling. Three classification approaches are explored: direct classification using DGNSS-derived vectors (Approach 1), image-based classification (Approach 2), and classification using combined feature vectors (Approach 3). Experimental evaluation using 2,312 labeled samples shows that ensemble learning significantly outperforms single-model classifiers for multipath detection. Random Forest achieves the highest performance across all approaches, reaching up to 99.75% accuracy, while CNN outperforms traditional methods, reaching up to 86.77% accuracy in image-based classification in Approach 2. The findings demonstrate the effectiveness of the framework in identifying and excluding compromised satellite signals, which has the potential to enhance the accuracy of GNSS positioning, with potential for real-world application in complex urban environments.