Practical Considerations in Censored Distributed Detection Systems

Censored distributed detection schemes represent an interesting alternative to quantization when communication between the sensors and the Data Fusion Center (DFC) is to be minimized. The idea of censoring is that each sensor sends only "informative" observations, and leaves those deemed "uninformative" un-transmitted. The problem, naturally, is to decide what is "informative" and what is not. In [1] this was (at least in principle) solved under both Bayes and Neyman-Pearson critenra; this paper is a more practically-oriented treatment. We shall search for simplification in design, specifically in the most interesting case that the DFC threshold is high and the communication constraint is severe. We shall also examine the problem lem from a distance-measure viewpoint. We shall compare censoring to the more classical binary-transmission framework work and observe its considerable decrease in communication needs. Finally, we shall explore the use of feedback to achieve optimal probability of error at a fraction of the bandwidth.