TITLE:Independent Component Analysis Using the Spectral Measure for Alpha-Stable Distributions

AUTHOR: Preben Kidmose
Department of Mathematical Modelling, Building 321
Technical University of Denmark, DK-2800 Lyngby, Denmark
email: pk@imm.dtu.dk
www: http://www.imm.dtu.dk


A method for identifying the independent components of an alpha-stable random vector is proposed. The method is based on an estimate of the spectral measure for the characteristic function. Simulations with both synthetic and speech signals demonstrate that the proposed method can identify independent components in the so-called over-complete case of more sources than mixtures.

Appears in Proc. of NSIP2001, Baltimore, Maryland, June 4-6, 2001.