Underwater acoustic localisation by GMM fingerprinting with noise reduction

In this paper, the underwater acoustic localisation is given by Gaussian mixture model (GMM) fingerprinting with noise reduction. Underwater acoustic measurement always contains a lot of noises and fluctuates seriously. The fluctuating measurement will make the localisation unreliable. To overcome this disadvantage, this paper gives two contributions. First, the singular value decomposition (SVD) technique is utilised to reduce the noise of underwater acoustic measurement. Second, the processed acoustic signal is statistically modelled by the GMM, which is a linear combination of multiple Gaussian functions, for fingerprinting and localisation. By using the SVD technique, one can suppress the noise-related subspace and reconstruct clean signals from the signal subspace only. Experiments are conducted in a large indoor tank to examine the boundary reflection effect. Note that our underwater localisation scheme is based on fingerprinting, which does not require any range or angle information. It can tolerate reflected, multi-path and random-noise components.