Estimation of Multipath in Deep Urban Area Using Nonlinear Model-based Dynamic Map

Yongjun Lee, Yongrae Jo, Yunho Cha, Minhyoung Cho, Byungwoon Park*

Precise positioning in urban areas has been a major challenge in GNSS field for a long time. Multipath, which is the dominant error source in urban areas where satellite signals are easily blocked and prone to reflection, cannot be removed by the differential technique since it is a user specific error such as receiver noise. Their types and effects are highly sensitively dependent on the user’s reception environment, and it is very difficult to model using methods traditionally used in the GNSS field due to their nonlinear characteristics. As various attempts have recently made to utilize artificial intelligence technique in GNSS fields, studies have been conducted to used machine learning technique in estimating multipath errors. In our previous work, we introduced a model that estimates multipath using relative position information of users and satellites in urban areas using support vector regression (SVR) and showed that the horizontal error was reduced to 34% by applying the model to static test in deep urban areas. In this study, we proposed a technique for generating a multipath map in urban area using multiple SVR-based multipath prediction models and applying the map according to the user’s approximate position. To verify the applicability of models in urban areas, a dynamic test was performed in Teheran-ro, Seoul.

Keywords: multipath, machine learning