Robust ToA-Based Localization in a Mixed LOS/NLOS Environment Using Hybrid Mapping Technique

Abstract

A two-stage hybrid method based on the machine learning approach is proposed for source localization using time of arrival (ToA) measurements in a mixed line of sight (LOS) and non-line of sight (NLOS) environment. The first stage applies an artificial neural network (NN) to detect the NLOS measurements that are outliers and the second stage passes the identified LOS measurements to an inverse weighted self-organizing network (IWSON) for determining the source location. The NN NLOS detector is able to take care of a variable number of NLOS measurements while the IWSON handles naturally a variable number of inputs and yields a solution without explicitly solving the nonlinear estimation problem. Simulations validate the good performance of the system with a different number of NLOS measurements. It provides a solution in reaching the Cramer-Rao‘ lower bound (CRLB) accuracy under a harsh multipath noisy environment, except over the small error region where it can act as an initialization for the iterative MLE to refine accuracy if necessary.