Random forest-based multipath parameter estimation

Abstract

Multipath is recognized as one of the major error sources for GNSS urban navigation. This study proposes a random forest (RF)-based multipath parameter estimator that uses random forest regression for parameter estimation, thereby mitigating multipath effect by removing the estimated reflected signal components. The proposed estimator is evaluated and compared with the multipath estimation delay-lock loop (MEDLL) for one-multipath and three-multipath cases, respectively. Simulation results demonstrate that the RF-based estimator is less affected by the front-end bandwidth of received signals, compared with MEDLL. The proposed RF-based estimator shows better performance than MEDLL for signals with front-end bandwidths of lower than 6 MHz. In 20 sets of tests on signals with a front-end bandwidth of 10 MHz in the three-multipath case, the RF-based estimator obtains smaller standard deviations than MEDLL. In experiments using real data with a front-end bandwidth of 2 MHz, the RF-based estimator reduces the 2D and 3D positioning errors by 8.5% and 8.7% over 180 epochs, respectively, against the conventional delayed-locked loop (DLL).