Header logo is ei

Subspace identification through blind source separation




Given a linear and instantaneous mixture model, we prove that for blind source separation (BSS) algorithms based on mutual information, only sources with non-Gaussian distribution are consistently reconstructed independent of initial conditions. This allows the identification of non-Gaussian sources and consequently the identification of signal and noise subspaces through BSS. The results are illustrated with a simple example, and the implications for a variety of signal processing applications, such as denoising and model identification, are discussed.

Author(s): Grosse-Wentrup, M. and Buss, M.
Journal: IEEE Signal Processing Letters
Volume: 13
Number (issue): 2
Pages: 100-103
Year: 2006
Month: February
Day: 0

Department(s): Empirical Inference
Bibtex Type: Article (article)

Digital: 0
DOI: 10.1109/LSP.2005.861581
Language: en
Organization: Max-Planck-Gesellschaft
School: Biologische Kybernetik

Links: PDF


  title = {Subspace identification through blind source separation},
  author = {Grosse-Wentrup, M. and Buss, M.},
  journal = {IEEE Signal Processing Letters},
  volume = {13},
  number = {2},
  pages = {100-103},
  organization = {Max-Planck-Gesellschaft},
  school = {Biologische Kybernetik},
  month = feb,
  year = {2006},
  month_numeric = {2}