Gaussian message passing-based cooperative localization on factor graph in wireless networks

Abstract

Location information has become attractive for a variety of applications in wireless networks. Cooperative localization was proposed to improve the performance in harsh environment where conventional localization methods failed due to the insufficient number of anchors. In this paper, distributed cooperative localization is studied based on message passing on factor graph. The joint a posteriori distribution of nodes׳ positions is represented by factor graph. Due to the nonlinearity between the positions and range measurement, the expressions of messages cannot be obtained in closed form by directly applying sum-product algorithm. Most existing methods resort to particle-based representation, which leads to both high computational complexity and large communication overhead. We propose to replace the factor node of the likelihood function by a linear Gaussian model. Accordingly, we are able to derive Gaussian messages on the revised factor graph, which only requires to update the mean vectors and the covariance matrices of multivariate Gaussian distributions. Simulation results show that the proposed method significantly outperforms the distributed maximum likelihood estimator and the extended Kalman filter, and performs very close to or even better than SPAWN estimator with much lower communication overhead and computational complexity in both static and mobile wireless networks.