Blind shallow-water acoustic channel estimation using frequency-domain natural gradient algorithm

Deconvolution with a signal received on a hydrophone in shallow water is made difficult by distortions caused by dynamic multipath propagation. It is acceptable to model the shallow water channel between the source and each hydrophone receiver as a multiple time-delay attenuation channel. Blind deconvolution techniques can adapt the processing to the acoustic environment if reasonable assumptions about the nature of the underwater channel used to represent the multipath propagation can be made. A blind deconvolution methodology for shallow-water channel estimation is employed and presented here, with no prior information on characteristics of the source or the channel. The methodology is based on the natural gradient in frequency-domain to minimize the mutual information. For its high efficiency and less computational complexity it is suitable for estimation of shallow-water channel. An application example of the approach using synthetic ocean acoustic data is also presented.