Unsupervised Machine Learning Approach for Multipath Classification of NavIC Signals

Abstract

Indian NavIC signals are received by various users equipped with the NavIC receivers in both L5 (1176.45 MHz) and S band (2492.028 MHz) frequencies. A NavIC receivers basically measures pseudo-ranges, carrier-ranges and Doppler observables for PVT. Pseudo-range measurements are absolute in nature, robust and therefore more widely used for the positioning purposes. However, pseudo-range measurements are more vulnerable to multipath. This deteriorates the measurement precision as well as robustness of the system to perform positioning in the challenging environments such as foliage, tall building or in the presence of the reflectors in the vicinity of the receiver. However, if receiver has the knowledge of Line of Sight (LOS), Multipath and Non Line of Sight (NLOS) signals, degradation in position accuracy due to the multipath affected signals can be addressed. In this regard, machine learning algorithms can be very effective in detection and classification of these signals. However, supervised machine learning algorithms require labelled data from the independent sources such as ray tracing, building models, 3D mapping aided with positioning etc.. To overcome this limitation, in this paper, we propose to use unsupervised learning algorithms and classify the signals purely based on the unlabeled data. Results are presented using NavIC data collected at Dehradun, India. This study may be useful for detection and removal of multipath affected signals for more robust positioning applications.