Hybrid RF Mapping and Kalman Filtered Spring Relaxation for Sensor Network Localization

Abstract

An accurate and low-cost hybrid solution to the problem of autonomous self-localization in wireless sensor networks (WSN) is presented. The solution is designed to perform robustly under challenging radio propagation conditions in mind, while requiring low deployment efforts, and utilizing only low-cost hardware and light-weight distributed algorithms for location computation. Our solution harnesses the strengths of two approaches for environments with complex propagation characteristics: RF mapping to provide an initial estimate of each sensor's position based on a coarse-grain RF map acquired with minimal efforts; and a cooperative light-weight spring relaxation technique for each sensor to refine its estimate using Kalman filtered inter-node distance measurements. Using Kalman filtering to pre-process noisy distance measurements inherent in complex propagation environments, is found to have significant positive impacts on the subsequent accuracy and the convergence of our spring relaxation algorithm. Through extensive simulations using realistic settings and real data set, we show that our approach is a practical localization solution which can achieve sub-meter accuracy and fast convergence under harsh propagation conditions, with no specialized hardware or significant efforts required to deploy.