Adaptive Bayesian sparse representation for underwater acoustic signal de-noising

摘要

In this paper we specifically address the problem of denoising and localisation/separation of underwater acoustic sources. There have been a number of approaches to this problem. Here we evaluate a recently proposed adaptive sparse sequential Bayesian approach. This approach extends sparse reconstruction methods to sequential data. This is achieved by extending the classic Bayesian approach to a sequential Maximum a Posterior (MAP) estimation of the signal over time. A sparsity constraint is enforced through the use of a Laplacian like prior at each time step. An adaptively weighted LASSO cost function is sequentially minimised using the new measurement received at each time step. This algorithm was tested on the very challenging Portland03 dataset. This dataset was collected at Portland harbour in the UK using two linear hydrophone arrays laid on the sea floor. The target, a small fishing boat, then performed a number of transits in the harbour in various directions. This dataset is particularly challenging with a lot of noise from both natural and man-made sources. Therefore an effective method of de-noising and localisation is expected to significantly improve the results on this dataset. Our preliminary results show that the Bayesian sparse representation technique is effective in source localisation and denoising on this dataset.