ACOUSTIC SIGNAL DENOISING BASED ON ROBUST PRINCIPAL COMPONENT ANALYSIS

Abstract

Abstract
Robust principal component analysis (RPCA) is a powerful procedure which decomposes a matrix into its lowrank
and sparse matrix components. As such it can be used for signal denoising in situations where useful part of
the signal can be represented as a low-rank matrix, which is usually the case in acoustic signals with some
inherent periodicity. This paper examines the applicability of RPCA for cyclostationary acoustic signal denoising
by decomposing the Short-time Fourier transform of a signal and eliminating its sparse component. The main
purpose of this approach is improvement of the signal-to-noise ratio in acoustic signals obtained in noisy
industrial surroundings for the purpose of fault detection or machine state estimation. The procedure is tested on
artificially generated signals as well as on real acoustic recordings.
Key words
Acoustic signals, Noise removal, RPCA, Industrial state estimation.