Evaluating Machine Learning & Antenna Placement for Enhanced GNSS Accuracy for CAVs

Abstract

Localization accuracy obtainable from global navigation satellites systems in built up areas like urban canyons and multi-storey car parks is severely impaired due to multipath and non-line-of-sight signal propagation. In this paper, a simple classifier was used in discriminating between multipath and line-of-sight GNSS signals. By using the carrier to noise ratio which characterizes the received signal strength of the GNSS signals, and the rate of change of the epochs of the satellite vehicles in view, a prediction accuracy of 98% was attained from the classifier. Also investigated in this paper is the effect of antenna placement on localization accuracy. Our measurement campaign using a Nissan Leaf hatch back model showed that the centre longitudinal line of the roof generated the least localization errors for an urbanized route.