eduzhai > Applied Sciences > Engineering >

Dynamic Independent Component/Vector Analysis Time-Variant Linear Mixtures Separable by Time-Invariant Beamformers

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 15 pages

Abstract: A novel extension of Independent Component and Independent Vector Analysisfor blind extraction separation of one or several sources from time-varyingmixtures is proposed. The mixtures are assumed to be separable source-by-sourcein series or in parallel based on a recently proposed mixing model that allowsfor the movements of the desired source while the separating beamformer istime-invariant. The popular FastICA algorithm is extended for these mixtures inone-unit, symmetric and block-deflation variants. The algorithms are derivedwithin a unified framework so that they are applicable in the real-valued aswell as complex-valued domains, and jointly to several mixtures, similar toIndependent Vector Analysis. Performance analysis of the one-unit algorithm isprovided; it shows its asymptotic efficiency under the given mixing andstatistical models. Numerical simulations corroborate the validity of theanalysis, confirm the usefulness of the algorithms in separation of movingsources, and show the superior speed of convergence and ability to separatesuper-Gaussian as well as sub-Gaussian signals.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×