Detection Of Signals In Non-Gaussian Reverberation Noise

Reverberation is one of the basic physical phenomena in under-water acoustic. It has been a long time since begin to research it. Suffering its Gaussian mechanism restrict, conventional techniques of underwater acoustical signal processing can not product ideal performance for anti-reverberation processing. This paper brings non-Gaussian signal processing techniques into anti-reverberation processing for active sonar. Based on this, the optimal detection of deterministic signal against reverberation background is found successfully.The probability density function (PDF) of reverberation transient amplitude can be fitted by Gaussian mixture (GM) model whereas the PDF of envelope by K-distribution model. GM parameters can be estimated by the expectation maximization iteration (EM) algorithm while K-distribution by the maximum likelihood algorithm and the moments estimate algorithm. A typical gaussianizing filter is studied in details. It is called U-filter, which is based on the PDF and its derivate. After the gaussianizing filter, the non-Gaussianity of reverberation would decrease.PSD of reverberation can be fitted by AR model. AR parameters can be estimated by the maximum likelihood estimation (MLE) and the weighted least squares estimation (WLSE). Then with the AR filter based on AR parameter estimation, reverberation can be prewhitened well.Utilizing coupling estimation for PSD and PDF parameters, combining prewhitening and gaussianizing modules, correlate prewhitening with gaussianizing (CPWG) filters is set up. The relationship between the CPWG and the REST is also studied.