Mixture density networks for WSN localization

Abstract

A new algorithm for determining soft range information in network localization is proposed. It applies a variant of neural networks called mixture density networks. When used in particle-based Bayesian localization procedures, it has a similar low computational complexity and provides comparable localization accuracy as existing methods. This property enables the proposed algorithm to be implemented on low-power wireless sensor network (WSN) nodes that are equipped with commercial ultra-wideband transceivers. The proposed algorithm is validated in indoor network localization experiments.