To simplify the ICA algorithms, the following preprocessing steps are usually taken:
Subtract the mean
from the observed variable
so it has zero mean. By doing so, the sources
also become zero
mean because
. When the
mixing matrix
is available,
can be estimated
to be
.
Transform observed variables so that they are uncorrelated and have unit
variance. We first obtain the eigenvalues
and their corresponding
eigenvectors
of the covariance matrix
,
and form the diagonal eigenvalue matrix
and orthogonal eigenvector matrix
(
).
We have
(228) |
(229) |
(230) |
(231) |
(232) |
The whitening process reduces the independent variables, the
components of the mixing matrix
to half (
) due to the
constraint that
is orthogonal. Moreover, the whitening
can also reduce the dimensionality of the problem by ignoring the
components corresponding to very small eigenvalues (PCA).