Application of the Sparse Low-rank Model in Denoising of Underwater Acoustic Signal

Abstract

Sound signals have good propagation effect in the marine environment, which is of great significance to under-water target positioning, underwater acoustic communication and so on. However, underwater acoustic signals are usually disturbed by a large amount of noise during the propagation due to the complexity of the marine environment. And we could not obtain the underwater acoustic signals precisely. Traditional denoising methods based on robust principal component analysis (RPCA) are limited by its incompleteness, and the denoised signal still has a lot of noise. We use the Go decomposition (Godec) algorithm in this paper, which is based on the RPCA algorithm to represent the noisy signal as sparse, low-rank and noise via sparse low-rank model. Then we use the non-negative matrix factorization (NMF) algorithm for the low-rank part to obtain the noise-free signal dictionary and the noise dictionary. Finally, the signal is reconstructed according to the noise-free signal dictionary, and we obtain the denoised underwater acoustic signal. To verify the effectiveness of this method, we perform denoising processing on the measured signals of the marine experiment. The results show that compared with the traditional RPCA algorithm, the denoised signal via our method in this paper has fewer noise components and has a better noise reduction effect.