Robust sequential detection of signals in noise

The problem of robust sequential detection of weak signals in noise is studied, and two methods for designing the most robust detectors are presented. They give an upper bound on the probability of error and minimize the maximum expected risk asymptotically in the Bayes sense. The saving in the number of observations needed compared with nonsequential M-detectors is calculated for both detectors, and the two methods are compared for the family of p-point distributions.