The positioning accuracy of Global Navigation Satellite System (GNSS) is becoming more important as the demand for location based service(LBS) is expanding to autonomous vehicles and security services. However, GNSS positioning is still vulnerable to multipath in urban areas. Thus, discriminating non-line-of-sight (NLOS)/ line-ofsight signal (LOS) condition and compensating for the multipath error are critical to mitigate the multipath error and improve the positioning accuracy. The conventional GNSS receivers, however, utilize LOS and NLOS signals without any distinction and compensation. In this paper, we propose a positioning technique that detects NLOS signal using a deep neural network (DNN) and compensates for multipath error using a differential range correction network based on multi-layer perceptron (MLP). With the field measurements in urban areas, the NLOS detector shows about 94% accuracy and the differential range network achieves to reduce distance root mean square (DRMS) up to one third of the conventional algorithm error.

Keywords: GPS, LOS, multi-layer perceptron, multipath error