Research Article Open Access

Blind Source Separation under Semi-White Gaussian Noise and Uniform Noise: Performance Analysis of ICA, Sobi and JadeR

Muna H. Fatnan1, Zahir M. Hussain1 and Hind R. Mohammed1
  • 1 University of Kufa, Iraq

Abstract

A comparative study is presented to evaluate the performance of three important Blind Source Separation (BSS) techniques under noisy conditions. The ability of FastICA, SOBI and JadeR is tested in separating several kinds of signals under noisy conditions, including human speech and frequency-modulated (quadratic and linear FM) signals. Additionally, different mixing matrices are used to inspect the effect of the mixing process. The influence of two types of noise (semi–white Gaussian and uniform) has been investigated under different Signal to Noise Ratios (SNR). The Pearson correlation coefficient (versus signal to noise ratio) between original and recovered signals is used as a performance metric. Despite the wide use of BSS techniques, there has been no extensive study in these directions. It is found that JadeR out performs other BSS techniques under semi-white Gaussian and uniformly-distributed noise.

Journal of Computer Science
Volume 15 No. 1, 2019, 27-44

DOI: https://doi.org/10.3844/jcssp.2019.27.44

Submitted On: 25 August 2018 Published On: 31 December 2018

How to Cite: Fatnan, M. H., Hussain, Z. M. & Mohammed, H. R. (2019). Blind Source Separation under Semi-White Gaussian Noise and Uniform Noise: Performance Analysis of ICA, Sobi and JadeR. Journal of Computer Science, 15(1), 27-44. https://doi.org/10.3844/jcssp.2019.27.44

  • 3,831 Views
  • 1,772 Downloads
  • 4 Citations

Download

Keywords

  • Independent Component Analysis (ICA)
  • Second-Order Blind Identification (SOBI)
  • Joint Approximation Diagonalization Estimation for Real Signals (JadeR)
  • Quadratic FM (QFM)
  • Linear FM (LFM)