Signal processing in an uncertain environment: Multivariate techniques of nonparametric and robust detection

The detection of known signals in additive noise is examined. The received signal is sampled, and the samples are used by the detector to determine whether or not a signal is present. The detectors are designed to provide nonparametric or robust performance when the data samples are statistically dependent. Nonparametric methods are proposed which result in detectors having a fixed false alarm probability. A new type of parametric test, the region test, is proposed. Two robust methods are proposed: the M-detector and the locally robust detector. False alarm and detection probabilities are estimated for both types of detectors, for specific noise distributions, and for signal waveforms. The performance of these detectors for nominal distributions and for classes of noise distribution functions is evaluated using finite-sample and asymptotic performance measures.