Abstract

A Machine Learning (ML)-based framework for navigating with global navigation satellite system (GNSS) signals in urban environments is developed. This framework aims to incorporate a pureL5^TM navigation system to obviate the requirement for a dual frequency RF front-end. To this end, first, this paper quantifies the performance of a pureL5^TM receiver in static and dynamic heavy multipath signal environments. Then, a deep neural network (DNN)-based methodology to leverage ML to mitigate multipath is presented. The performance of the proposed framework is analyzed. Experimental results for a dynamic receiver navigating in a deep urban environment show that the proposed framework reduces the 95% horizontal confidence level from 44.0 m to 18.8 m. It is also shown that the proposed framework is able to reduce the standard deviation of the pseudorange error from 11.22 m to 5.34 m.