Information-theoretic node scheduling in collaborative sensor networks

In this contribution, we propose an efficient collaborative online filter in a wireless sensor network under communication constraints. The observed system is assumed to evolve according to a non linear state-space model. The distributed intelligent sensors collaborate to estimate the marginal probability of the hidden continuous state, within a Bayesian framework. The communication protocol relies on the activation of only few nodes. The selection of the leader node, through time, and its spatially collaborating nodes is based on a trade-off between data relevance (measured by an information criteria), message approximation error and communication constraints.