Bayesian Localization in Sensor Networks: Distributed Algorithm and Fundamental Limits

Abstract

Abstract-Self-localization in ad-hoc sensor networks is becoming a crucial issue for several location-aware applications. This technology implies the combination of absolute anchor locations with relative inter-node information exchanged on a peer-topeer basis. In this paper we investigate a distributed algorithm and fundamental performance bounds for Bayesian cooperative localization in stochastic networks. Nodes are assumed to be randomly deployed within a finite space according to a prior distribution. Bayesian inference is performed through an iterative local message passing procedure based on belief propagation and particle-filtering message representation. The algorithm performance is analyzed for a simplified scenario in which unknown node positions are randomly scattered along a line segment and anchors are fixed. Global Cramer-Rao bounds are derived and compared to the performance of the distributed algorithm.

Cooperative Bayesian Estimation of Vehicular Traffic in Large-Scale Networks

“…With this perspective, graph theory is a promising tool for describing the statistical relations existing among traffic observations at different, and typically nonuniformly distributed, locations or time instants. Unlike other contexts such as image processing, wireless sensor networks, wireless positioning, and decision making, in which Markov random fields (MRFs) and BNs [27]- [30] are now mature tools, the approach is relatively new in traffic estimation. Traffic can be view as an MRF defined over an undirected graph, whose nodes are associated with the traffic variable observed at different locations/instants, while edges define the relations existing among the variables according to the fluid-dynamic theory [31].…”

“…2) The message that node b sends to its neighbor at iteration k + 1 is computed following the approach in [27]. Particles are propagated from node b to node using the distribution ψ(s t,b , s t, ) to obtain the new set {s …”

Cooperative estimation for under-determined linear systems

“…Examples of distributed statistical inference in wireless communications are spectrum analysis in cognitive radio [1], (time and/or frequency) synchronization [2], and localization [3]. Let us consider N networked agents modelled as an undirected graph labeled in V = {1, .…”

Distributed estimation of macroscopic channel parameters in dense cooperative wireless networks

“…Knowledge of these long-term propagation features is instrumental in cooperative and cognitive networks to optimize robust decentralized group-selection and resource-sharing strategies [5]. RSS-based cooperative [7]- [9] and non-cooperative [10] localization systems could also benefit from fast-converging distributed estimation of these channel features. For distributed estimation, in this work we propose to employ the consensus approach [11]- [14], based on successive refinements of local estimates at nodes with information exchange between neighbors.…”