Decentralized detection in relay sensor networks

In this paper, we consider the binary decentralized detection problem in large scale sensor networks. We propose a broadcast-based decentralized detection scheme: after making local decisions based on their observations, sensors broadcast their decisions to a number of nearby relay nodes. The relay nodes then summarize what they receive from the sensors and report to the fusion center. We analyze the asymptotic system performance of this scheme, characterized by the error exponent, in the following two scenarios: (i) sensors nodes are regularly placed on the grid points (ii) sensors and relay nodes are distributed in a fixed area and their communication suffers from the channel fading. The asymptotic performance of the parallel sensor network with perfect communication channels between sensors and fusion center is used as a benchmark for comparison. We prove that the error exponent of our scheme approaches that of the parallel sensor network with perfect fusion channel asymptotically when the number of relay nodes for each sensor goes to infinity.