A two-level distributed multiple hypothesis decision system

A two-level distributed decision system consisting of a number of local decision makers (LDMs) connected to a global decision maker (GDM) is considered. The LDMs share a common M-hypothesis testing problem, have their own observations independent of each other, and employ likelihood ratios for their decision making. Each LDM transmits its inference to the GDM where the final decision is derived. The local inferences consist of the ranking of the candidate hypotheses and a degree of confidence based on likelihood ratios. Using a maximum distance criterion, the optimum confidence-based subpartitioning of local decision space is studied. It is shown that the presented system can greatly outperform one in which LDM provides a single-hypothesis hard decision and can perform nearly as well as the optimum centralized system