Variational inference cooperative network localization with narrowband radios

Abstract

Distributed cooperative network location estimation in motionless and mobile narrowband wireless sensor networks (WSNs) is studied, where agents (to be localized) are several hops away from the anchors (with a priori known location). This work proposes a reduced-communication cooperative and distributed particle filtering (CoopPF) approach based on variational inference (VI) Gaussian mixture modeling (GMM), where network nodes exchange information locally, i.e. only with neighboring terminals; each node transmits the parameters of the estimated Gaussian mixture, instead of the whole posterior density, offering tremendous reduction in communication overhead, as required in narrowband applications (e.g. underwater communications). The proposed VI approach jointly estimates the number of required Gaussians and their parameters, in sharp contrast to standard expectation-maximization techniques, where the number of components must be estimated first with other techniques (e.g. clustering). Accuracy comparable to state-of-the-art PF cooperative localization is demonstrated, with an order of magnitude reduction in communication overhead.