Robust-Extended Kalman Filter and Long Short-Term Memory Combination to Enhance the Quality of Single Point Positioning

Abstract

In the recent years, multi-constellation and multi-frequency have improved the positioning precision in GNSS applications and significantly expanded the range of applications to new areas and services. However, the use of multiple signals presents advantages as well as disadvantages, since they may contain poor quality signals that negatively impact the position precision. The objective of this study is to improve the Single Point Positioning (SPP) accuracy using multi-GNSS data fusion. We propose the use of robust-Extended Kalman Filter (referred to as robust-EKF hereafter) to eliminate outliers. The robust-EKF used in the present work combines the Extended Kalman Filter with the Iterative ReWeighted Least Squares (IRWLS) and the Receiver Autonomous Integrity Monitoring (RAIM). The weight matrix in IRWLS is defined by the MM Estimation method which is a robust statistics approach for more efficient statistical data analysis with high breaking point. The RAIM algorithm is used to check the accuracy of the protection zone of the user. We apply the robust-EKF method along with the robust combination of GPS, Galileo and GLONASS data from ABMF base station, which significantly improves the position accuracy by about 84% compared to the non-robust data combination. ABMF station is a GNSS reception station managed by Météo-France in Guadeloupe. Thereafter, ABMF will refer to the acronym used to designate this station. Although robust-EKF demonstrates improvement in the position accuracy, its outputs might contain errors that are difficult to estimate. Therefore, an algorithm that can predetermine the error produced by robust-EKF is needed. For this purpose, the long short-term memory (LSTM) method is proposed as an adapted Deep Learning-Based approach. In this paper, LSTM is considered as a de-noising filter and the new method is proposed as a hybrid combination of robust-EKF and LSTM which is denoted rEKF-LSTM. The position precision greatly improves by about 95% compared to the non-robust combination of data from ABMF base station. In order to assess the rEKF-LSTM method, data from other base stations are tested. The position precision is enhanced by about 87%, 77% and 93% using the rEKF-LSTM compared to the non-robust combination of data from three other base stations AJAC, GRAC and LMMF in France, respectively.

A New Multi-Scale Sliding Window LSTM Framework (MSSW-LSTM): A Case Study for GNSS Time-Series Prediction

“…In navigation and positioning, Tan et al used LSTM as a de-noising filter and proposed the rEKF-LSTM method to significantly improve single-point positioning accuracy [29]. Jiang et al proposed an LSTM-RNN algorithm to filter MEMS gyroscope outputs, and the results indicated that the method was effective for improving MEMS INS precision [30].…”

Static Attitude Determination Using Convolutional Neural Networks

“…To solve the problem, it is possible to use a combination of artificial neural networks (ANNs) with Kalman filters (KFs) [ 9 ] or even recurrent neural networks (RNNs) in order to estimate the new attitude [ 10 ]. These approaches require some previous attitude information in order to adjust the predictions, that is, they depend on time information.…”

Observability Analysis and Navigation Filter Optimization of High-Orbit Satellite Navigation System Based on GNSS

“…In addition, many scholars have proposed a large number of corresponding improved filter algorithms for the nonlinearity and uncertainty of the actual navigation system as well as the non-gaussian and correlation of noise, etc. [26][27][28][29][30]. Reference [31] proposed an adaptive filter method based on the state observability.…”