Optimizing systems of threshold detection sensors

Abstract

When implementing a system of sensors, one of the biggest challenges is to establish a threshold at which a signal is generated. All signals that exceed this detection threshold are then investigated to determine whether the signal was due to an "event of interest," or whether the signal is due simply to noise. Below the threshold all signals are ignored. We develop a mathematical model for setting individual sensor thresholds to obtain optimal probability of detecting a significant event, given a limit on the total number of false positives allowed in any given time period. A large number of false signals can consume an excessive amount of resources and could undermine confidence in the system's credibility. One motivation for this problem is that it allows decision makers to explicitly optimize system detection performance while ensuring it meets organizational resource constraints. Our simulations demonstrate the methodology's performance for various sizes of sensor networks, from ten up to thousands of sensors. Such systems apply to a wide variety of homeland security and national defense problems, from biosurveillance to more classical military sensor applications.