Filter De-Noising Method Using Long Short-Term Memory

Abstract

GNSS brings more signals and more satellites to improve positioning services. This paper introduces data fusion from multiple Global Navigation Satellite System (GNSS) constellations. In fact, some failures in satellite's signals negatively impact the quality of positioning. For this purpose, this paper presents the robust Extended Kalman Filter (robust-EKF) to eliminate the outliers and de-noising method based on the Long Short-Term Memory (LSTM). The algorithms are tested using GPS, Galileo and GLONASS data corresponding to base station ABMF in Guadeloupe. Robust combination of GPS, Galileo, and GLONASS data improve the position accuracy from 41.0% to 95.0% compared to the use of independent systems and by about 84.0% compared to the non-robust combination of GPS, Galileo, and GLONASS data. In particular, the position precision improves significantly using the method LSTM-EKF by about 74.0% compared to the robust-EKF.