Please use this identifier to cite or link to this item:

Maximum likelihood ICA of quaternion Gaussian vectors

Authors Vía, J.
Palomar, D.P. View this author's profile
Vielva, L.
Santamaría, I.
Issue Date 2011
Source ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings , 2011, p. 4260-4263
Summary This work considers the independent component analysis (ICA) of quaternion random vectors. In particular, we focus on the Gaussian case, and therefore the ICA problem is solved by exclusively exploiting the second-order statistics (SOS) of the observations. In the quaternion case, the SOS of a random vector are given by the covariance matrix and three complementary covariance matrices. Thus, quaternion ICA amounts to jointly diagonalizing these four matrices. Following a maximum likelihood (ML) approach, we show that the ML-ICA problem reduces to the minimization of a cost function, which can be interpreted as a measure of the entropy loss due to the correlation among the estimated sources. In order to solve the non-convex ML-ICA problem, we propose a practical quasi-Newton algorithm based on quadratic local approximations of the cost function. Finally, the practical performance and potential application of the proposed technique is illustrated by means of numerical examples. © 2011 IEEE.
ISSN 1520-6149
ISBN 9781457705397
Language English
Format Conference paper
Access View full-text via DOI
View full-text via Scopus