Novel 3D mapping-aided algorithm for real-time, robust GNSS positioning in urban environments

Position-aware portable, low-cost, GNSS electronic devices accumulate to more than 98% of the GNSS installed base and have found fertile soil especially in highly populated areas with complex road networks, i.e. large urban centres. Their positioning performance in such challenging propagation environments is hindered due to complete or partial blockage of line-of-sight GNSS signals, which results to multipath interference and NLOS propagation phenomena. One possible way to improve the accuracy of mass-market GNSS devices under such conditions is to exploit 3D mapping data to re-fine the estimated position of the receiver. The shadow-matching class of algorithms follow such paradigm and rely on three main steps to deliver accuracy improvements. First, a set of candidate positions is defined within a region-of-interest (ROI) that contains the truth receiver position, based on covariance information provided by the least-square adjustment process. Second, the “geometric” visibility status (i.e. LOS or NLOS) of each available satellite is evaluated at each candidate position with the aid of pre-computed building-boundaries. Last, each candidate position is assigned a score by matching the visibility status predicted based on signal strength-based trained model and the “geometric” visibility status estimated in the previous step, considering all available satellites. In this work, I propose several modifications and improvements to shadow-matching algorithm that enable single digit (i.e. < 10m) positioning accuracy performance in real-time, even in deep-urban environments. Initially, I provide a novel methodology to define the ROI around the true receiver position, with a high-success rate, utilising the pre-fix residuals and a machine learning (ML) model. Moreover, a new, computationally efficient, methodology is presented to estimate the “geometric” visibility of a satellite in one or more candidate positions and finally, a machine learning scoring scheme is established that allows selection of the candidate position closest to the true one.