Detection of non-Gaussian signals

The resolution of closely spaced narrow-band signals is formulated as a probelm in multiple-hypothesis testing with non-Gaussian statistics. Two cases are considered. In the first, the noise model deviates from the Gaussian assumption. For the second, the signals contain random parameters which are governed by non-Gaussian distributions. It is shown that the optimum processor falls within the general estimator-correlator framework. Suboptimum detectors are investigated for the common situation where signal statistics are unknown or where the optimum processor is too complicated to implement. A hierarchy of non-linear processing is established which relates both to the implementation of the (conditional mean) estimator and to the degree of deviation from Gaussian statistics.