Please use this identifier to cite or link to this item:
Title: Constrained blind source separation : frequency domain independent component analysis with reference
Authors: Chen, Wenda.
Keywords: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition
Issue Date: 2009
Abstract: Semi-blind separation of signals is useful for many applications in signal and image processing. Independent Component Analysis (ICA) method incorporates certain a priori knowledge of the interested sources and has been shown to be very useful in particular applications where the frequency of the estimated signal is known. While time-domain ICA needs to assume instantaneous mixtures and independent sources, frequency-domain ICA (FICA) is especially useful for solving convolutive mixtures of source signals since it can transform time domain convolution to a multiplication in the frequency domain. However, the permutation of the sources is exagerated in FICA because the individual demixing operations are used in different frequency bins and leads to different orders in the reconstruction process. In this report, we show how ICA with Reference (ICA-R) is extended to the frequency domain and convolutive mixtures as FICA-R, effectively solving the permutation problem in FICA while enhancing the optimization convergence using reference signals. We demonstrated the performance of our technique on synthetic data and real EEG and speech datasets. The results indicate that our proposed Frequency-domain ICA-R (FICA-R) is more effective and efficient than other FICAs and time-domain ICA-R in both convolutive and other types of mixtures.
Rights: Nanyang Technological University
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:SCSE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
Chen Wenda 09.pdf
  Restricted Access
2.95 MBAdobe PDFView/Open

Page view(s) 50

checked on Oct 20, 2020

Download(s) 50

checked on Oct 20, 2020

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.