Dimensional reduction using Blind Source Separation for identifying sources

Ganesh R. Naik, Dinesh K. Kumar

Research output: Contribution to journalArticlepeer-review

11 Citations (Scopus)


Separation of independent sources using Blind Source Separation (BSS) techniques requires prior knowledge of the number of independent sources. Performing BSS when the number of recordings is greater than the number of sources can give erroneous results. Techniques employed to estimate suitable recordings from all the recordings require estimation of number of sources or require repeated iterations. This paper demonstrates that normalised determinant of the global matrix is a measure of the number of independent sources, K, in a mixture of M recordings. This paper also shows that performing ICA on K out of M randomly selected recordings gives good quality of separation. The qualities of the outcome of this experiment were measured using Signal to Interference Ratio (SIR) and Signal to Noise Ratio (SNR). The results demonstrate that using this technique, there is an improvement in the quality of separation as measured using SIR and SNRs. ICIC International

Original languageEnglish
Pages (from-to)989-1000
Number of pages12
JournalInternational Journal of Innovative Computing, Information and Control
Issue number2
Publication statusPublished - Feb 2011
Externally publishedYes


  • Blind source separation (BSS)
  • Frobenius norm
  • Global matrix
  • Independent component analysis (ICA)
  • Overcomplete ICA
  • Source identification
  • Source separation
  • Undercomplete ICA


Dive into the research topics of 'Dimensional reduction using Blind Source Separation for identifying sources'. Together they form a unique fingerprint.

Cite this