Analysis And Mitigation Of NLOS Errors In Gnss Applications In Urban Canyons

With the development of Io T and telecommunication techniques,Global Navigation Satellite System(GNSS) based smart devices have been widely adopted and deployed in various industries.Navigation services and other location-based services have been integrated into many ordinary consumers’ daily life to provide convenience.The users’ requirements for the availability and stability of the GNSS have then also been increasing along with their dependency on the aforementioned services and relevant technologies.Meanwhile,as the development of cities,countless tall buildings with complicated surfaces have been built up to provide residential,commercial and industrial areas.As a result,dense buildings around urban streets would inevitably lead to the urban canyon effect,which would cause the GNSS location’s accuracy decline.Generally speaking,a consumer level GNSS handset is able to determine the user’s location with an error within meters or decimeters in widely open areas,but it would cause a location error of tens of meters or even over a hundred of meters in several cases in dense urban areas.This kind of location error is mainly caused by the obstruction,reflection,and refraction of satellite signals by common obstacles in the city,such as surfaces of most buildings and other objects including trees and vehicles.And it seriously affects the service quality of various location based services.To address the existing big location error problem for GNSS-based handsets in complicated urban areas,this thesis therefore proposes two possible attempts to mitigate the error by analyzing the causes and the characteristics of the problem.The proposed two methods are a 3D ray tracing method,and an artificial neural network based method.The former method which is based on three-dimensional city models,can be utilized to analyze and mitigate location errors by taking positions of satellites,approximate locations of the user and surfaces of nearby buildings into consideration with the help of ray-tracing algorithm,in order to improve the final accuracy.The latter is based on artificial neural networks.It tries to learn and fit the pattern of the displacement of devices’ reported locations and the corresponding ground truths by creating and training artificial neural networks for the specified regions,which helps to modify the location errors and further improve the accuracy.Simulation results demonstrate that both methods can enhance the user devices’ positioning accuracy to some extent.On one hand,the 3D model-based solution presents positioning improvements mostly on the vertical direction.The improvement on the horizontal direction is generally from zero to three meters.But in several adverse cases,the accuracy even decreases.On the other hand,the solution with the deep learning method demonstrates significantly better results over the unmodified positioning method.Specially,the achieved positioning error by the proposed method is 0.046 m,3.53 m,and 23.48 m for the 50 th percentile,75 th percentile,and 90 th percentile,respectively,while the corresponding one for the unmodified positioning method is 21.34 m,39.34 m,and 69.28 m,respectively.