A Decentralized Sequential Test with Data Fusion

In this paper, we generalize the sequential probability ratio test (SPRT) of Wald to a distributed system. In particular, we consider a network of two detectors which are connected in parallel. Each local detector performs an SPRT based on its own observations and communicates its local decision to the global decision maker. The global decision maker combines the local decisions according to a predetermined fusion rule, and decides either to terminate the overall test and accept one of the two hypotheses, or to continue. The global error probabilities are shown to be functions of the local error probabilities and the fusion rule. The global test length is formulated in terms of the local test lengths, and the average global test length is derived. An example is presented for the case of two identical local detectors. The results obtained show clearly that for the same level of performance, the proposed scheme has a shorter average decision time than the single detector case.