Error exponents for decentralized detection in tree networks

Abstract

Consider a set of sensors, one of them designated as the fusion center. We are given two hypotheses H0 and H1, with associated probability spaces. In this chapter, we consider only simple hypothesis testing, i.e., the probability measures under both hypotheses are known to the network. The goal of the network is to make a decision on the true hypothesis based on information provided by observations made at each sensor node. This is commonly known as the decentralized detection problem. Decentralized detection in sensor networks has attracted a lot of interest in recent years, because of new technologies (especially, the availability of low-cost sensing devices) and numerous potential applications. The decentralized detection problem was first formulated and studied by [3], which considers a "parallel configuration" whereby each sensor makes an observation and sends a quantized version of that observation to a fusion center. The goal is to make a decision on the two possible hypotheses, based on the messages received at the fusion center. © 2008 Springer-Verlag US.