Distributed localization and clustering using data correlation and the Occam´s razor principle

We present a distributed algorithm for computing a combined solution to three problems in sensor networks: localization, clustering, and sensor suspension. Assuming that initially only a rough approximation of the sensor positions is known, we show how one can use sensor measurements to refine the set of possible sensor locations, to group the sensors into clusters with linearly correlated measurements, and to decide which sensors may suspend transmission without jeopardizing the consistency of the collected data. Our algorithm applies the “Occam´s razor principle” by computing a “simplest” explanation for the data gathered from the network. We also present centralized algorithms, as well as efficient heuristics.