Efficiency of the Generalized Neyman–Pearson Test for Detecting Changes in a Multichannel System

We propose a multialternative, multistage rule for detection of changes. The rule consists of fixed-length stages and it stops when the maximum likelihood statistic crosses the threshold for the first time. Parameter optimization and analysis of the rule are conducted for change detection in the mean of the Wiener process in a multichannel system. In the worst case, the proposed rule is inferior by a factor of two to the asymptotically minimax multialternative sequential rule with the mean time to false alarm tending to infinity. The results remain valid also in the general (non-Gaussian) case if we additionally require convergence of the hypotheses.