The marginal utility of cooperation in sensor networks

Abstract

We present arguments that a small number of sensors within the network provide most of the utility. That is, cooperation of more than a small number of nodes has little benefit. We present two scenarios. In the first scenario, all sensors provide identical utility, and their utilities are aggregated sequentially. The second scenario is sensor fusion with signal strength decreasing with distance. In that scenario the source is at the origin and the sensors are distributed, either uniformly or according to a planar standard normal distribution. We also vary the total number of sensors distributed in both scenarios to observe the utility/density trade off. Localization using the Fisher information as the utility metric is used to demonstrate that few sensors are sufficient to derive most of the utility out of the sensor network. Simulation results back up an order statistics analysis of the behavior. The implication is that while co-operation is useful for some objectives such as combating fading and uncertainty of individual sensors, it is inefficient as a mean to increase the utility of a sensor network if the best sensorpsilas utility is significantly short of the desired utility.