On the Use of Machine Learning Algorithms to Improve GNSS Products

This paper presents relevant results on the investigation of possible uses of machine learning based techniques for the processing of data in the field of Global Navigation Satellite Systems (GNSSs). The work was performed under funding of the European Space Agency and addressed different kind of data present in the entire chain of the positioning process, as well as different kind of machine learning approaches. This paper presents the results obtained for two promising GNSS applications: the prediction of ionospheric maps for the correction of the related error on the pseudorange measurement; and the forecast of fast corrections normally present in the EGNOS messages, in case the latter are missing. Results show how, based on the historical data and the time correlation of the values, machine learning methods outperformed simple regression algorithms, improving the positioning performance at GNSS user level. The work results also confirmed the validity of this approach for the automatic detection of outliers due to ionospheric scintillation phenomena.