1: \begin{abstract}

2: We revise the problem of extracting one independent component from 

3: an 

4: instantaneous linear mixture of signals.

5: The mixing matrix is parameterized by two vectors, one column of 

6: the mixing matrix and one row of the de-mixing matrix. %, like in 

7: %the Internal Linear Combination (ILC) method

8: %for extraction of a single component, but the difference is that 

9: %no part of the mixing matrix is known a priori.

10: %Compared to Independent Component 

11: %Analysis, a novel parameterization of the %mixing model is used. 

12: The separation is based on the non-Gaussianity of the source of 

13: interest, while the other 

14: background signals are assumed to be Gaussian. 

15: Three gradient-based estimation algorithms are derived using the 

16: maximum 

17: likelihood principle and are compared with the Natural Gradient 

18: algorithm for Independent Component Analysis and with One-unit 

19: FastICA based on negentropy maximization. The ideas and algorithms 

20: are also generalized for 

21: the extraction of a vector component when 

22: the extraction proceeds jointly from a set of instantaneous 

23: mixtures.

24: Throughout the paper, we address the problem of the size of the 

25: region of convergence for 

26: which the algorithms guarantee the extraction of the desired 

27: source. We show how that size is influenced by the ratio of powers 

28: of the sources within the mixture. Simulations confirm this 

29: observation where several algorithms are compared. They show 

30: various convergence behaviour in a scenario where the source of 

31: interest is dominant or weak. Here, our proposed modifications of 

32: the gradient methods taking into account the dominance/weakness of 

33: the source show improved global convergence property. 

34: 	\end{abstract}