A neurodynamic optimization approach to TDOA-based IoT localization in NLOS environments

Wenxin Xiong,Christian Schindelhauer,Hing Cheung So,Junli Liang,Zhi Wang

2020-01-01

Abstract:Time-difference-of-arrival (TDOA) localization using unreliable data collected by the Internet of Things (IoT) sensors in non-line-of-sight (NLOS) environments is considered herein. We exploit a hardware realizable neurodynamic optimization model, with the use of a Welsch $M$-estimator based loss function to enhance the robustness to outliers. Since the outlying nature of NLOS bias is covered when generating the TDOAs from IoT-sensor-garthered timestamps, we propose to retrieve the underlying time-of-arrival components during the problem formulation by jointly estimating the source position and onset time. A projection-type neural network (PNN) based on the redefined augmented Lagrangian and projection theorem is then applied to solve the derived nonconvex constrained optimization problem, whose local stability of equilibrium is examined and discrete implementation complexity is analyzed. Simulation investigations show that our neurodynamic TDOA localization solution is superior to several existing schemes in terms of localization accuracy, especially when the NLOS paths and errors tend to exhibit sparsity and severeness, respectively.