Supervised Machine Learning Assisted Hybrid Positioning Based on GNSS and 5G

Abstract

Global Navigation Satellite System (GNSS) and New Radio (NR) signals based positioning are both specified for User Equipment (UE) positioning in a 3rd Generation Partnership Project (3GPP) network. They may also be fused to determine UE position by hybrid methods. In an urban scenario, the UE often suffers from non-line of sight (NLoS) propagation conditions from the satellite or the base station (BS). Identifying the links that have a line of sight (LoS) condition between the transmitter and the receiver is of paramount importance to enhance the accuracy of position estimates from GNSS, NR signals, or hybrids thereof. To address the issues with NLoS links, we propose a novel positioning solution fusing measurements made using GNSS and fifth-generation (5G) signals in the frequency range 1 (FR1) in an urban environment. We apply a supervised machine learning (ML) approach to classify LoS and NLoS for both GNSS and 5G signals based on the feature set. An extended Kalman filter (EKF) fuses observable measurements with LoS from both GNSS and 5G to estimate the UE position. We obtain positioning errors below 30 cm indoors, and below 2 m for 90% of all positioning fixes. Moreover, we observe that using our proposed fusion approach outperforms positioning using either NR signals or GNSS signals alone. We demonstrate that it is advantageous to deploy a transmission and reception point (TRP) at the areas where GNSS-based positioning shows degradations, as the results show that a single TRP for hybrid positioning already halves the positioning error compared to using only the LoS GNSS signals.