Sharif Digital Repository

In this paper, the performance of the gradient method based on Score Function Difference (SFD) in the separation of i.i.d. and periodic signals will be investigated. We will see that this algorithm will separate periodic signals better than i.i.d. ones. By using this experimental result and the fact that voiced frames of speech signals are approximately periodic, a modified algorithm named VDGaradient has been proposed for separation of speech signals in synthetic convolutive mixtures. In this method, voiced frames of speech signal will be used as the input to the gradient method, then the resulting separating system will be applied to separate sources completely  

Score Function Difference (SFD) is a recently proposed "gradient" for mutual information which can be used in Blind Source Separation algorithms based on minimization of mutual information. To be applied to practical problems, SFD must be estimated from the data samples. In this paper, a new method for estimating SFD is proposed. To compare the performance of this new estimator with other proposed SFD estimation methods, we have applied them in separating linear instantaneous mixtures. It will be seen that our method performs superior to all other methods previously proposed for estimation of SFD  

Equivariant adaptive separation via independence (EASI) is one of the most successful algorithms for blind source separation (BSS). However, the user has to choose non-linearities, and usually simple (but non-optimal) cubic polynomials are applied. In this paper, the optimal choice of these non-linearities is addressed. We show that this optimal non-linearity is the output score function difference (SFD). Contrary to simple non-linearities usually used in EASI (such as cubic polynomials), the optimal choice is neither component-wise nor fixed: it is a multivariate function which depends on the output distributions. Finally, we derive three adaptive algorithms for estimating the SFD and... 

In this paper, a nonparametric "gradient" of the mutual information is first introduced. It is used for showing that mutual information has no local minima. Using the introduced "gradient", two general gradient based approaches for minimizing mutual information in a parametric model are then presented. These approaches are quite general, and principally they can be used in any mutual information minimization problem. In blind source separation, these approaches provide powerful tools for separating any complicated (yet separable) mixing model. In this paper, they are used to develop algorithms for separating four separable mixing models: linear instantaneous, linear convolutive, post...