Statistical normalization of non-Rayleigh reverberation

Abstract

Low-frequency active sonar systems operating in shallow water are primarily limited by reverberation.

Reverberation is traditionally assumed to follow a Rayleigh probability distribution, from which the

detector and other signal processing algorithms are developed. Experimental studies have shown that

reverberation can be non-Rayleigh distributed with varying statistical character over range and

bearing. In such a situation, detectors designed under a Rayleigh assumption will exhibit increased

and varying false alarm performance. This report develops a technique for dealing with the unknown and varying reverberation statistics by using a non-Rayleigh reverberation model to perform a statistical normalization of the background reverberation. In this manner, range-bearing images will be

produced that have a constant background reverberation distribution (i.e., Rayleigh). The technique is evaluated through simulation, with particular attention to the false alarm performance, which is seen

to depend on the severity of the non-Rayleighness of the reverberation and the amount of data used

to estimate the parameters of the reverberation model. Application to real data has shown that a large

degree of the non-stationarity of the range-bearing image can be removed by statistical normalization.