Title :
Identifiability issues in noisy ICA
Author_Institution :
DSP Group, Univ. of London, UK
fDate :
5/1/2004 12:00:00 AM
Abstract :
We consider the identifiability of the statistical model for noisy independent component analysis showing that while the mixing process is identifiable, the noise covariance is only partially so. This raises questions as to the performance of certain maximum-likelihood algorithms for blind source separation in the presence of noise.
Keywords :
blind source separation; independent component analysis; maximum likelihood estimation; signal denoising; signal reconstruction; signal sources; blind source separation; maximum-likelihood algorithms; noise covariance; noisy ICA; noisy independent component analysis; statistical model identifiability; Background noise; Blind source separation; Covariance matrix; Digital signal processing; Gaussian noise; Helium; Higher order statistics; Independent component analysis; Noise reduction; Source separation;
Journal_Title :
Signal Processing Letters, IEEE
DOI :
10.1109/LSP.2004.826508
