Machine Learning Based Overbound Modeling of Multipath Error for Safety Critical Urban Environment

Abstract

In the urban environment, multipath and non-line of-sight are the critical source of measurement errors and signal power loss. In urban canyons, whilst the user can still acquire the required number of satellites to obtain a position, thanks to multi-constellation GNSS, such signals may be subject to gross multipath errors and lead to a potentially unsafe position. In this paper, machine learning techniques are used to model the multipath error distributions. The set of features which have been assessed are commonly used parameters such as the elevation, S/N, and user speed. For modeling and evaluation of the model validity, a large number of hours of experimental data has been collected by driving a sensor-equipped vehicle in the urban area in Toulouse. Considering the processing of data from single-frequency type GNSS receiver, the multipath error component is extracted from measurement using appropriate techniques (measurement differential, clock bias estimation, etc.). Quantile of multipath error are modeled using neural network-based regression technique with the features. Modeling results using the proposed method are validated by an integrity assessment of the experimental data.