How Machine Learning Techniques can help to improve GNSS products and services
Last Updated: 19/12/2022 12:41 Created at: 16/12/2022 12:28
Final Presentation of NAVISP Project EL1 035 BIS now available:
On Thursday, December 15, 2022, Altec S.p.A presented, together with its consortium partner Politecnico di Torino the results of the NAVISP project “Machine Learning Techniques to Model GNSS”.
Modern GNSS navigation message performance depends on underlying complicated models but at user level it relies on a relatively simple Keplerian parameterization, plus potential correction parameters. The estimation accuracy of these parameters from real measurement data, or from simulated data in case of performance prediction, is critical for the final performance in the positioning domain.
The main objective of the activity was therefore to assess the usability of machine learning (ML) techniques and to improve the GNSS navigation message performance. This was achieved trying to reduce the impact of the different error sources, which could be mitigated by detecting missing parameters and predicting their values , or by filtering out data affected by the errors. To this end, the GNSS Machine Learning Demonstrator (GMLD) was developed, aiming at investigating and demonstrating the use of ML in various areas of GNSS domain. The results of the activity cover:
- the demonstration of usability of ML to improve data availability and/or quality in some segments of a GNSS system;
- the implementation of selected ML techniques into tools to simulate GNSS system segments capabilities and behaviour;
- the implementation of a framework to conduct additional investigation on the usage of ML techniques in other use cases of the GNSS domain.
Four different use cases have been selected to cover a variety of ML applications in the GNSS domain as well as different data structures with the goal of predicting missing information or the presence of possible error sources. The use cases are orbit prediction improvement using machine learning approaches, prediction of ionosphere daily maps, estimation of SBAS correction parameters in missed messages, and disturbance-based outlier detection classifications.
Thanks to the support from NAVISP, the project was able to define, implement and validate all proposed GNSS applications identified to investigate ML in the GNSS domain using the GMLD software, that has reached the TRL 3 target. The hereby obtained results have been analysed in depth to recognise pro and cons with respect to standard approaches. In addition, applications have been integrated in the GMLD software demonstrator allowing GNSS experts to run them in a simple and effective way. Moreover, the GMLD software has been designed and implemented providing the general data management and ML capabilities as part of the framework therefore it can be easily reused to execute further investigation and implement new applications. Currently the software is running at ALTEC deployed as standalone system and at the GNSS Science Support Centre (GSSC) deployed as DataLab containers.
The project was carried out in the scope of NAVISP Element 1, which is dedicated to technology innovation of the European industry in the wide PNT sector.
More detailed information can be found in the slides of the Final Presentation.