Detection in a non-Gaussian environment

Techniques for the detection of a weak signal in non-Gaussian, ill-defined noise are considered. Statistical characterizations used are moments, tail measures related to quantiles, and a measure related to the score function. For multivariate densities, the characterization is by means of a nonlinear transformation. Initial results seem to indicate that assuming a particular family of probability densities does not necessarily result in a significant degradation in performance when the observations actually come from a density outside the assumed family. More important to performance are accurate estimates of the moments, tail measures, or other parameters which are used to specify the detector.