Convolutive decorrelation algorithms form a class of powerful algorithms for blind source separation. In contrast to ICA, they are based on vanishing second order cross correlation functions between sources. We provide an analyze an unifying approach for convolu-tive decorrelation procedures. The convolutive decor-relation procedures impose the problem of simultaneously diagonalizing a number of covariance matrices. We examine diierent cost functions for simultaneous diagonalization with respect to the demixing matrix. It turns out, that best performance is achieved for a cost function, that takes the squared sum of the oo diagonal elements after the diagonal elements were normalized to unity. We then provide criteria for convolu-tion kernels, that are optimal for noise robustness and which can guarantee positive deenite covariance matrices , which are important for reliable convergence.
[1]
Terrence J. Sejnowski,et al.
An Information-Maximization Approach to Blind Separation and Blind Deconvolution
,
1995,
Neural Computation.
[2]
Hagai Attias,et al.
Blind Source Separation and Deconvolution: The Dynamic Component Analysis Algorithm
,
1998,
Neural Computation.
[3]
John C. Platt,et al.
Networks for the Separation of Sources that Are Superimposed and Delayed
,
1991,
NIPS.
[4]
K. Obermayer,et al.
Principal Component Analysis and Blind Separation of Sources for Optical Imaging of Intrinsic Signals
,
2000,
NeuroImage.
[5]
Schuster,et al.
Separation of a mixture of independent signals using time delayed correlations.
,
1994,
Physical review letters.