Channel order selection in blind deconvolution based on eigenvector characteristics and using normal modes in conjunction with multipath compression for source identification

Abstract

Two related topics in blind deconvolution are presented. The first concerns channel order selection. Many blind deconvolution techniques assume the filter length to be known. One technique of filter length determination uses the dimension of a nullspace obtained by decomposing correlation matrices. Although the dimension is usually determined by calculations involving the eigenvalues, this alternative method relies on the known characteristics of eigenvectors among the subspaces. Multipath functions in underwater acoustics commonly have high kurtosis. The new method can give correct estimates on filter length for signal-to-noise ratios as low as 10 dB. The matrix methods described may not work as well when the sample length increases. However, the second topic shows how normal mode properties in conjunction with time compression in modeled multipath functions may be used to enhance blind deconvolution in a waveguide when the sample length is large. Given data from a vertical array of hydrophones, one may accurately obtain a source identification. The methods of the tracking and positioning of normal modes are incorporated with an input of a linear frequency-modulated pulse propagated through an ocean environment. [Research supported by ONR.]