Performance evaluation of GNSS-based railway applications

Global Navigation Satellite Systems (GNSS) can be used for various land applications, especially for safety-related applications such as train localization and car localization for the purpose of control. Although a significant amount of integrity assessment has been done in the field of air traffic control, integrity or reliability studies related to train and car applications have generally not been done to date. The most difficult factor in land applications of GNSS is multipath errors. It is known that GNSS signals may be reflected by buildings, walls, vehicles, and the ground. These reflected signals can interfere with reception of the signals coming directly from the satellites, a phenomenon known as multipath. To use GNSS-based position for these types of safety applications, not only position information but also reliability information have to be implemented. Therefore we need to evaluate the real multipath errors for the future reliability evaluation. In this study, we analyzed raw GNSS data obtained in a real-railway condition to demonstrate the accuracy of GNSS. These results will be very important for proposing a better positioning algorithm and for railway reliability estimation. GNSS data were derived from a large number of test runs of more than 2,000 km along several railway lines in the area of the West Japan Railway Company in Japan. First, we evaluated the accuracy to establish the GNSS position errors under real railway lines based on precise reference positions. The precise reference positions were prepared using local RTK-GNSS fixed positions, GNSS velocity, IMU, and a speed sensor equipped on the train. We analyzed the temporal multipath errors for each satellite to show the error distribution. Furthermore, to improve the accuracy of these data, several methods including signal strength check, pseudo-range check by Doppler frequency, and data screening by using the true distance between two antennae, were proposed. As a result, it was confirmed that large errors and jumps were greatly reduced. For example, the percentage of large horizontal error over 5 m was reduced approximately 66% by using our proposed multipath detection techniques.