Independent Component Analysis in the Presence of Correlated Gaussian Noise

• Activity Number: 698
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Rajesh Nandy*
• Companies: University of North Texas Health Science Center
• Keywords: Independent Component Analysis ; ICA ; Dimension reduction ; Model order selection ; Principal Component Analysis ; Factor Analysis

Abstract

Independent Component Analysis (ICA) is a computational method for separating a multivariate mixed signal into its original non-Gaussian sources. ICA algorithms were originally developed in the absence of any noise in the model and have later been extended with some success to incorporate the presence of white noise. However, no ICA algorithm has been developed so far for mixed signals in the presence of correlated noise with unknown covariance structure. A novel iterative algorithm will be presented to solve this problem, where the conventional non-noisy ICA will be used to first estimate the number of independent sources and then to estimate the noise covariance matrix under the weak assumption of stationarity for the noise. The estimated noise covariance matrix will be used to whiten the noise from which the sources can be extracted more reliably. Several iterations of this process will ensure accurate extraction of original sources. Results will be presented using simulated data where it will be established that the proposed method provides superior extraction of original sources compared to non-noisy ICA algorithms as well as noisy ICA algorithms with white noise.