|Stochastic Systems Group|
Pat Kreidl - SSG, MIT
We consider the Bayesian detection problem in the context of ad-hoc wireless sensor networks, assuming a decentralized model that is most appropriate for applications where communication resource is scarce (e.g., voluminous local data exceeds stipulated link capacities, reliable single-hop connections are sparse) but per-node computation resource is in relative abundance. Under certain assumptions, the problem is known to admit a set of necessary optimality conditions, leading to a system of nonlinear equations for which an algorithmic solution relies on an iterative use of the computation required to solve a single-sensor Bayesian detection problem. We identify additional assumptions that allow this iterative algorithm to exhibit a message-passing property, greatly facilitating its distributed implementation: the computations and communications can be sequenced in a manner where each sensor node need only be aware of information that is either (i) locally available at initialization or (ii) provided by messages from only its immediate neighbors in the network. We also discuss known convergence properties, practical examples for which all required modeling assumptions are satisfied, and directions for future research.
Problems with this site should be emailed to firstname.lastname@example.org