TFBSS performs Blind Source Separation of (over)determined multiplicative
mixtures of non-stationary real valued sources.
TFBSS is based on the joint-diagonalization of whitened and
noise-compensated Spatial Time-Frequency Distribution (STFD) matrices of the
observations, corresponding to single auto-terms positions, as described in:
C. Févotte and C. Doncarli. "Two
contributions to blind source separation using time-frequency
Signal Processing Letters, 2004.
Signal Processing Letters, vol.
11, no. 3, Mar. 2004. pdf
A. Holobar, C. Févotte, C. Doncarli, and D. Zazula. "Single autoterms
selection for blind source separation in time-frequency plane". In Proc. 11th
EUSIPCO, Toulouse, France, 2002 (Special Session on Source
The TFBSS package contains the following MATLAB files:
For a quick overview of TFBSS performance, download all the files in a
common directory and call demoTFBSS.m in a MATLAB command window.
- tfbss.m : main program
- joint_diag_rc.m : slightly modified version of J.-F. Cardoso's
code joint_diag_r.m, available on
J.F Cardoso web site and used with the kind permission of
its author (copyright owned by J.-F. Cardoso)
- tfrspwv.m & window.m : perform TFDs computation. Excerpts
from the Matlab Time-Frequency
Toolbox and used with the kind permission of their author F. Auger
(copyright owned by F. Auger).
- demoTFBSS.m : demo script of TFBSS, separation of 4 slightly
noisy instantaneous mixtures of 3 Time-Varying ARMA sources.
- data.mat : contains the sources used in demoTFBSS.
Download the TFBSS pack (*.zip)
(Last Update 02 Sep 2003)
% This program is distributed in the hope that it will be
% but WITHOUT ANY WARRANTY; without even the implied
% MERCHANTABILITY or FITNESS FOR A PARTICULAR
PURPOSE. See the
% GNU General Public License for more details.
Copyright (C) Sep 2003 C. FEVOTTE, A. HOLOBAR
Last Update: March 18, 2008