Using Semi-parametric Models To Improve The Accuracy Of Baseline Solutions With Multipath Errors

Most of the relative positioning errors of high-precision global navigation satellite system?GNSS?are generally reduced or eliminated through the double-differencing method.However,due to differences in the observation environments between the reference station and the rover,the effects of multipath cannot be reduced by double-differencing and may amplify its impact,even for relative positioning with short baselines.Multipath errors may cause obvious decreases in the precision of ambiguity floating-point solutions and deteriorate the results.To solve these problems,a semi-parametric model is proposed to reduce multipath errors.Compared with the ordinary least squares method,the experimental results show that the effects of multipath effects in static baseline solution and single epoch baseline solution can be effectively reduced by the semi-parametric regression model.The main contents of this paper include:1)Data collection.The principle of multipath error generation is analyzed,and the measurement experiment is designed according to it.Then GNSS short baseline observation data including multipath errors are collected.2)Static solution.The least-squares method and a semi-parametric model are used to resolve observation data with multipath errors.The superiority of the semi-parametric model is proved by comparing the residual distribution and the accuracy of the solution results.As the key of the semi-parametric model,the regularization matrix R is experimented by the natural spline method and the time series method.At the same time,the regular parameter?is determined by three methods,including the L-curve method,the generalized cross-validation?GCV?method and minimum mean square error?MSE?method.The experimental results show that,in the static baseline solution,the multipath error can be effectively reduced by the semi-parametric model,and the coordinate precision can be controlled to the millimeter level.3)Single-epoch solution.In order to avoid the rank defect of the normal equation,the P code pseudorange equations,widelane carrier phase observation equations and₁carrier phase observation equations are formed.Considering that a single epoch solves a baseline with multipath errors,resolving the ambiguity is difficult.Therefore,part of the search method is adopted.On this basis,the regularization combination required by the model is determined by three different schemes.By comparing the solution results of the semi-parametric model and the least squares model,it can be proved that the semi-parametric model can control the accuracy of most epochs to 1.5cm and below,and is more suitable for processing multipath errors.