[Eeglablist] separation of white gaussian sources
dgroppe at cogsci.ucsd.edu
Mon Nov 22 20:31:54 PST 2004
There is no way to uniquely re-whiten a mixture of Gaussian
sources that are independent across time (i.e. the activity of a source at
time t tells you nothing about the activity of that source at time t+x).
This is because there are an infinite number of matrices that will whiten
the mixture. Any matrix that whitens the mixture can be rotated to
produce a new matrix that will whiten the data, because a rotated whitened
distribution is still a whitened distribution.
If the sources are not independent across time, Belouchrani et al.'s
Second Order Blind Identification (SOBI) algorithm might work:
A. Belouchrani, K. Meraim, J.-F. Cardoso, and E. Moulines, "A blind
source separation technique using second order statistics" IEEE Trans.
on Sig. Proc. 45, pp. 434-444, Feb. 1997
as might Gorodnitsky & Belouchrani's JCC algorithm:
I. Gorodnitsky and A. Belouchrani, "Joint cumulant and correlation based
signal separation with application to EEG data analysis" in ICA 2001, 2001
hope this helps,
On Mon, 22 Nov 2004, Marco Congedo wrote:
> can anybody point me on existimng algorithms for the separation
> of uncorrelated white gaussian sources (maybe not completely "blind")
> WITHOUT any additional statistical assumption?
> Thanks a lot,
More information about the eeglablist