Abstract
We study distributed detection in a sensor network where the sensors cooperate by exchanging information to reach a common understanding about the environment. We address two main issues: (1) distributed fusion: how to achieve a global decision without transmitting the information (measurements or local decisions) from all the sensors to a common central location like in parallel architectures; and (2) connectivity graph: what should be the connectivity pattern among the sensors, in other words, with which sensors should each sensor communicate. This is a nontrivial question since it corresponds to designing the structure of a graph to achieve a given goal. For the first issue, we propose an iterative algorithm that fuses the data globally without the need for collecting them at one central location. For the second issue, we present a design methodology based on "small world" network engines that leads to connectivity patterns that provide fast convergence to the distributed detection algorithm. Results show that introducing 10% to 30% randomness in the connectivity graph leads to significant improvements over both regular patterns and totally random networks.