Signal Detection in Non-gaussian Noise by a Kurtosis-based Probability Density Function Model

The problem of HOS-based signal detection methods applied in real communication systems is addressed. The Locally Optimum (LO) criterion is selected from a large number of detection criteria. It can be applied either under the ideal (but often not realistic) assumption of Gaussian background noise, or on the basis of realistic statistical models of channel noise. Conventional Fourier analysis (using first and secondorder statistics) allows a receiver to obtain optimum detection results in the presence of Gaussian noise. However, in many real communication applications using the ideal assumption of Gaussian noise causes the performances of a conventional approach to decay significantly. In these cases, Higher Order Statistics (HOS) has been selected as a powerful approach that allows complete signal and noise characterizations, and that optimizes detection performances. The present paper describes the applications of the LO criterion to both conventional and HOS approaches. Its performances have been evaluated in the field of underwater acoustics.