Unsupervised Clustering of GNSS Receiver Measurements for Satellite Selection

This article suggests an approach for selecting satellites in the Global Navigation Satellite Systems (GNSS) receiver positioning process using Self Organizing Map (SOM) and K-means, which are unsupervised Machine Learning (ML) algorithms. The algorithm processes common GNSS receiver observables, namely, carrier phase, pseudorange, and signal carrier-to-noise ratio. Its objective is to create clusters of consistent measurements to select a subset of visible satellites. As relying on all available satellites and utilizing them in the positioning process may not necessarily enhance positioning accuracy, it is essential to select an optimal subset of visible satellites, while considering the potential errors in each satellite signal. Given that the Geometric Dilution Of Precision (GDOP) approach for satellite selection only takes into account the geometric arrangement of satellites, it is not the ideal method to consider low-elevation satellites, as it fails to take into account the effects of the atmosphere. The effectiveness of the suggested method is demonstrated by utilizing actual GNSS receiver data that were collected at one of Wide Area Reference Stations (WRS), namely Kansas City, of Wide Area Augmentation System (WAAS).