Year
2010
Abstract
Almost all of the \"architectures\" for radiation detection systems in Department of Energy (DOE) and other USG programs rely on some version of layered detector deployment. Efficacy analyses of layered (or more generally extended) detection systems in many contexts often assume statistical independence among detection events and thus predict monotonically increasing system performance with the addition of detection layers. We show this to be a false conclusion for the ROC curves typical of current technology gross gamma detectors, and more generally show that statistical independence is often an unwarranted assumption for systems in which there is ambiguity about the objects to be detected. In such systems, a model of correlation among detection events allows optimization of system algorithms for interpretation of detector signals. These algorithms are framed as optimal discriminant functions in joint signal space, and may be applied to gross counting or spectroscopic detector systems, provided that signal information can be shared “downstream” across layers. We have shown how system algorithms derived from this model dramatically improve detection probabilities compared to the standard serial detection operating paradigm for these systems. These results would not surprise anyone who has confronted the problem of correlated errors (or failure rates) in the analogous contexts, but seem to be largely unappreciated in the literature analyzing the radiation detection problem – independence is widely assumed and experimental studies often fail to measure correlation. This situation, if not rectified, could lead to several unfortunate results, including [1] overconfidence in system efficacy, [2] overinvestment in layers of similar technology, and [3] underinvestment in diversity among detection assets.