Lan, T.; Dodinoiu, A.; Geffert, A.; Becker, U.:
Machine-Learning-Based Position Error Estimation for Satellite-Based Localization Systems.
ettc2020 - European Test and Telemetry Conference, Juni 2020.
Satellite-based localization systems are nowadays widely deployed in transportation, especially with the progress of global navigation satellite systems (GNSS). However, GNSS signals are easily degraded by the local environment. This compromises the accuracy of the position solution and makes it challeng-ing to implement satellite-based localization systems in the road and railway domain. With the help of Bayes filters (e.g., Kalman and particle filter), the localization accuracy can be improved. However, these filters are constrained by assumptions, and they require accurate modeling of errors for optimal estima-tion. Under these circumstances, another modeling method is researched in this paper. As machine learning has become more sophisticated over the years, neural networks are now suitable for learning the relation between the position errors and the abundant information from the GNSS receiver without prior knowledge. Therefore, the dilution of precision, the elevation angle and the carrier-to-noise ratio are appropriate indicators for signal degradation. In this paper, it is shown how neural networks are trained to estimate the position error of satellite-based localization systems. For modeling the temporal correlation in position error measurements, the long short-term memory (LSTM) network is applied. Finally, it can be demonstrated that the neural networks are able to learn the trend in position errors.