| dc.description.abstract | Source localization by signal subspace techniques in shallow water is difficult due to coherent multipath, resulting in non-planar wavefronts across the array. Also, the noise in shallow water is largely generated by wind action at the surface. Such a background noise is no longer white and uncorrelated among the sensors of an array as required by signal subspace methods. A knowledge of the noise coherence is necessary for satisfactory applications of the signal subspace method.
This thesis looks at these twin problems of source localization in shallow water. Initially, a signal subspace approach for source localization is developed using image theory, assuming the background noise is spatially white and the channel is known completely. It is observed that the algorithm performs well when all significant images are used in the process and it is robust against small changes in the channel. The algorithm is sensitive to bottom velocity variations, but this sensitivity can be used to estimate the true velocity with precision. A modification to the algorithm is suggested to estimate source parameters without having a priori knowledge about the bottom parameters.
Later, a generalized surface noise model is developed as a continuum of coherent point sources over a sector of an annular ring, which can be reduced to common geometries, viz., ring, disc, or infinite plane of source distribution. The noise sources are assumed to have general spatial spectral characteristics, and expressions for spectrum and cross-spectrum between two sensors along a vertical passing through the centre of the noise model are obtained. It is observed that the spectrum is an increasing function of directionality parameter ‘m’ of coherent noise sources, modeled as directional monopoles (cos ?)^m, in a rigid bottom channel, whereas it is a decreasing function of ‘m’ in a soft bottom channel, and the spectral level increases linearly with the radius of the disc of noise sources. The coherence is an increasing function of ‘m’ in both types of channels and is complex for disc sizes smaller than 10–50 times the channel depth.
Finally, the performance of the signal subspace algorithm has been studied in the presence of surface-generated noise background in a specific case of uncorrelated source distribution over a large disc. It is found that the algorithm performs well above a threshold SNR (>10 dB) and that averaging over many noise eigenvectors, which are not corrupted by background noise, yields better results. The performance has dependence on array position, giving the best results when the array is near the surface compared to its position in the middle or at the bottom of the channel. | |
