GNSS Shadow Matching: Improving Urban Positioning Accuracy Using a 3D City Model with Optimized Visibility Prediction Scoring

The poor performance of global navigation satellite systems (GNSS) user equipment in urban canyons is a well-known problem, especially in the cross-street direction. A new approach, shadow matching, has recently be proposed to improve the cross-street accuracy using GNSS, assisted by knowledge derived from 3D models of the buildings close to the user of navigation devices. In this work, four contributions have been made. Firstly, a new scoring scheme, a key element of the algorithm to weight candidate user locations, is proposed. The new scheme takes account of the effects of satellite signal diffraction and reflection by weighting the scores based on diffraction modelling and signal-to-noise ratio (SNR). Furthermore, an algorithm similar to k-nearest neighbours (k-NN) is developed to interpolate the position solution over an extensive grid. The process of generating this grid of building boundaries is also optimized. Finally, instead of just testing at two locations as in the earlier work, realworld GNSS data has been collected at 22 different locations in this work, providing a more comprehensive and statistical performance analysis of the new shadowmatching algorithm. In the experimental verification, the new scoring scheme improves the cross street accuracy with an average bias of 1.61 m, with a 9.4% reduction compared to the original SS22 scoring scheme. Similarly, the cross street RMS is 2.86 m, a reduction of 15.3%. Using the new scoring scheme, the success rate for determining the correct side of a street is 89.3%, 3.6% better than using the previous scoring scheme; the success rate of distinguishing the footpath from a traffic lane is 63.6% of the time, 6.8% better than using the previous scoring scheme.