On the Convergence and Stability in Second-Order Statistics BSS

Part of the Lecture Notes in Electrical Engineering book series (LNEE, volume 3)

In Chap. 6, we have derived natural gradient SOS-BSS algorithms (NG-SOS-BSS) to adapt the separation matrix W. This chapter addresses the question of the convergence of these algorithms, that is, the convergence of W to a separating matrix. An examination of the global convergence is of practical interest because it may indicate the range of admissible step-sizes μ for which these NG-SOS-BSS algorithms globally converge. This issue is discussed in Sect. 7.1, which explains why the analysis of the global convergence is hardly tractable for NG-SOS-BSS algorithms, even in the case of instantaneous mixing. However, in the instantaneous case, an analysis is possible for a simplified decorrelation algorithm that is closely related to NG-SOS-BSS. Then, we propose an interpretation of the results of this analysis for convolutive NG-SOS-BSS. In contrast to the global behavior, the local behavior of convolutive NG-SOS-BSS may be analyzed. This is discussed in Sect. 7.2 where sufficient local stability conditions are given.


Equilibrium Point Global Convergence Global Stability Local Stability Separation Matrix 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Copyright information

© Springer Science+Business Media, LLC 2009

Personalised recommendations