Mitigation of GPS Periodic Multipath Using Nonlinear Regression

Motivated by the idea of imposing machine learning approaches to improve fidelity of Global Positioning System (GPS) measurements, this work proposes a nonlinear regression method to tackle multipath mitigation problem for GPS fixed ground stations. Posing multipath error corresponding to each visible satellite as a function of the satellite's repeatable geometry with respect to a fixed receiver on sidereal daily basis, the multipath estimator is trained using historical data of a few reference days and is then used to correct multipath-corrupted measurements on the successive days. The well-known Support Vector Regression (SVR) is employed to train the estimator of multipath of each satellite. With error analysis on real recorded data, we show that our proposed method achieve state-of-the-art performance in code multipath mitigation with 79% reduction on average in terms of standard deviation of multipath error. The improvement on precision of positioning solution of multipath-corrected data is of 25-35%.