OFI: OPTIMAL FILTERING ALGORITHMS FOR SOURCE SEPARATION
Andreas Ziehe y , Guido Nolte z , Gabriel Curio z and Klaus-Robert Muller y
We analyze second-order methods for signal separa-
tion and show how specic ltering operators can be
used to improve the separation performance. It is the
goal to exploit all spectral information in an optimal
way. A perturbation theoretical approach is used to de-
rive an explicit solution for the 2 2 mixing case and
an approximate solution for the M M case is given.
The usefulness of our optimal ltering method (OFI)
is demonstrated by simulations.