MIMO-AR system identification and blind source separation for GMM-distributed sources
IEEE Transactions on Signal Processing
Some results on the denoising problem in the elliptically distributed context
IEEE Transactions on Signal Processing
QML-based joint diagonalization of positive-definite hermitian matrices
IEEE Transactions on Signal Processing
Blind separation of non-stationary sources using continuous density hidden Markov models
Digital Signal Processing
Hi-index | 0.01 |
In this paper, two novel algorithms for blind separation of noiseless instantaneous linear mixture of independent sources are presented. The proposed algorithms exploit nonGaussianity of the independent sources by modeling their distribution using the Gaussian mixture model (GMM). The first proposed method is based on the maximum-likelihood (ML) estimator. According to this method, the sensors distribution parameters are estimated via the expectation-maximization (EM) algorithm for GMM parameter estimation and the separation matrix is estimated by applying nonorthogonal joint diagonalization of the estimated GMM covariance matrices. The second proposed method is also a ML-based approach. According to this method, the distribution parameters of the prewhitened sensors are estimated via the EM algorithm for GMM parameter estimation and a unitary separation matrix is estimated by applying orthogonal joint diagonalization of the estimated GMM covariance matrices. It is shown that estimation of the sensors distribution parameters amounts to obtaining a tight lower bound on the log-likelihood of the separation matrix, and that the joint diagonalization of the estimated GMM covariance matrices amounts to maximization of the obtained tight lower bound. Simulations demonstrate that the proposed methods outperform state-of-the-art blind source separation techniques in terms of interference-to-signal ratio.