Signal Design Techniques. Volume IV. Distribution-Free Detection Procedures

摘要

A class of two-input detection systems for digital communication over random and unknown channels is investigated. The systems investigated possess false-alarm rates which are invariant for wide classes of channel statistics. Specifically, coincidence detection procedures with invariant or distribution-free false-alarm rates are proposed and investigated. The only information concerning the channel statistics which is required by these detectors is the median of the noise under no-signal conditions. The coincidence procedures are subsequently modified so that the detectors utilizing them become either learning systems with respect to slowly time-varying and/or unknown location parameters or adaptive systems with respect to rapidly varying and/or unknown location parameters. The classes of detection problems for which the false-alarm rates of the above procedures detectors remain distribution-free are also obtained. In addition to the distribution-free coincidence detectors, a detector based on the T-statistic, and well suited for the detection of stochastic signals in noise, is proposed and investigated. The T-statistic is then modified so that the detector utilizing it becomes an adaptive system with respect to rapidly varying and/or unknown location parameters. The wide classes of detection problems for which the above detectors remain distirbution-free are also obtained. The distribution-free detectors are then applied to various practical detection problems, and their performances are evaluated and compared to the performances of comparable likelihood detectors. (Author)