Stochastic Systems Group
		Home Research Group Members Programs   Demos Calendar Publications Mission Statement Alumni

We consider the estimation of a correlated time varying random-field with a wireless sensor network; in particular, we focus on estimating the field at locations where there may be no sensors. We obtain reduced-order models describing the field at the desired locations by cut-point sets derived from a graph approach and estimate the field by Kalman filtering. We explore tradeoffs in the selection of the cut-point sets: a smaller cut-point set uses few sensors, leading to a Kalman filter of low dimension with low communication cost; a larger cut-point set, using higher dimensional filters, may provide higher estimation accuracy. We use Pareto optimality to identify the set of "best" solutions from which the cut-point set and associated Kalman filter should be selected. Experimental results illustrate the approach and explore the tradeoffs in a sensor network estimating a temperature field.

Problems with this site should be emailed to jonesb@mit.edu