An Investigation into Robust Estimation Applied to Correlated GPS Networks

Abstract

High precision positioning can be achieved using GPS satellites. However, there exist many error sources. For example, receiver locations have to be chosen on the basis of task requirements rather than by optimality of the environment with respect of GPS signal propagation. Hence the effect of unfavorable error sources on precise GPS positioning is a typical problem. Least squares estimation (LSE) yields results of low accuracy in the presence of outliers in GPS measurements. Different weight models have therefore been suggested in order to account for such errors. In this study the authors use robust estimators which clearly identify outlier observations. They perform significantly better than LSE. Furthermore, GPS observations are fundamentally correlated, hence the robust estimators are applied to heterogeneous and correlated observations. The mean success rate (MSR) statistic is employed as a practical tool for measuring the abilities of different methods. Many different correlated GPS networks based on IGS sites were used and the robust methods were applied to simulated corrupted samples and the degree of corruption is varied. Performance of the estimators was further demonstrated using real GPS data.