Doppler and Pseudorange Measurements as Prediction Features for Multi-Constellation GNSS LoS/NLoS Signal Classification

Precise positioning and timing are key challenges for safety & security critical applications in autonomous and trusted vehicles (ATVs). A global navigation satellite system (GNSS) receiver is one of the sensors used in ATVs to provide a satellite-based positioning, navigation and timing solution. GNSS robustness in built-up areas like urban canyons and in sheltered locations like multi-storey car parks is severely impaired due to non-line-of-sight and multi-path signal propagation. In this paper, an optimised decision tree classifier is used for the discrimination of GNSS signals into line-of-sight and non-line-of-sight signals. In addition to using the carrier-to-noise ratio and the satellite elevation angle as prediction features, adding the pseudoranges and the Doppler measurements enhances the classifier performance by 12 %. As a result, an overall prediction accuracy of 97.8 % is obtained. For the studied data-set, the pseudorange measurements have a higher prediction importance than the Doppler measurements. Also evaluated in this study, is the use of a GPS L1 signal prediction regression model on other GNSS constellations. Analysis of the investigated scenario shows that the use of the current model is not adequate for non-GPS constellations.