Short-Time Sample Robust Detection Study
PRINCETON UNIV NJ PRINCETON United States
Pagination or Media Count:
The optimum detector is well defined when the signal and noise characteristics are completely specified through deterministic andor probabilistic statements. Unfortunately this is not always the case and attempts to make it so through assumptions of one sort or another may lead to poor detector performance. This report studies the detection problem when certain of the signal or noise properties are unknown except perhaps within some wide class. Particular attention is directed to communications problems characterized by small time-bandwidth signals. Several applications emerge from the study. The first results from an exhaustive investigation of the nonparametric Wilcoxon rank sum detector which is found to be practical in implementation and nearly as good in performance as the optimum detector under a wide range of sample size, dependence, and noise distribution conditions. A second application from the study results from an investigation of optimum small sample linear coincidence detectors which are found to be superior to a Gaussian parametric detector when the normality assumption is violated. A third application from the study results from the analysis of a simple adaptive threshold receivers. In addition there are results of an incomplete or theoretical nature whose application is either not so important or as yet well defined. These include the study of an optimum robust detector for nearly Gaussian noise.