Non-line of Sight Error Mitigation in Bayesian Differential Localization Systems

This paper evaluates and compares the accuracies of the three commonly used Bayesian filters namely extended Kalman Filter (EKF), H-infinity filter and unscented Kalman filter (UKF) in differential time of arrival (TDoA) localization topology in ultra-wide band (UWB) systems. Furthermore, two non-line of sight (NLOS) error suppression mechanisms are introduced and the accuracies of these methods when applied on the mentioned filters are evaluated. The results are evaluated in several trials both in LOS and NLOS conditions using a mobile node carried by a robot. In practical trials, it was witnessed that the H-infinity filter is more robust against non-ideal conditions with excessive noise or inaccurate model of the system whereas the UKF filter has the highest localization accuracy in nearly ideal cases such as LOS. In NLOS conditions, the proposed methods are able to suppress the error up to 50% with the prerequisite that the NLOS condition is correctly determined.