On the use of sample detectors in large distributed detection systems

This paper studies large distributed detection systems that are constrained to use local sample detectors in which sensors produce a binary output by comparing the measurement sample against a threshold. We focus on the problem of choosing between two system design options: spend resources to find optimum local thresholds or deploy additional sensors to compensate for non-optimum thresholds. To provide guidance for this decision, we determine sufficient conditions on the noise distribution to ensure that the overdeployment ratio required by the second option decreases to 1 as the signal-to-noise ratio decreases. We further illustrate that this conclusion is not reached by many noise distributions, including the uniform distribution.