This paper aims to provide a realistic modeling of a generic noise pro
bability density function (pdf). The target is to obtain a parametric
model dependent on few parameters (simple to estimate), and so general
to be able to describe many kinds of noise (e.g., symmetric or asymme
tric, with variable sharpness). To this end, three HOS-based models ar
e proposed. They present different levels of generality, which increas
e with the number of HOS parameters used in the model. The most genera
l and effective of these parameters are used in the asymmetric general
ized Gaussian, which is expressed in terms of kurtosis, providing vari
able sharpness (from flat to impulsive shapes), and left and right var
iances (whose combination provides the same information as skewness),
describing eventual deviation from symmetry. HOS-based pdf modelling i
s applied to optimize signal detection in non-Gaussian environments. T
he proposed models are used and compared in the design of locally opti
mum detection (LOD) tests. Promising experimental results are presente
d, being obtained by the application of the tests for detecting known
signals corrupted by real underwater acoustic noise. (C) 1998 Elsevier
Science B.V. All rights reserved.