Identifiability and Separability of Linear Ica Models Revisited

We prove theorems that ensure identifiability, separability and uniqueness of linear ICA models. The currently used conditions in ICA community are hence extended to wider class of mixing models and source distributions. Examples illustrating the above concepts are presented as well. In this paper we are concerned with conditions ensuring that the mixing system may be identified and sources separated in a linear instantaneous ICA (Independent Component Analysis) model. The fundamental problems of identifiability, separability and uniqueness of ICA models are addressed. We give formal definitions to each of these concepts. A theorem is given for each of the above conditions and proofs are constructed. These theorems generalize the results given in earlier presented theorems addressing the same problems [Com94, CL96, Car98, DLDMV99, TJ99]. In particular, some of the restricting assumptions made in those theorems may be relaxed. As a result, the identifiability, separability or uniqueness can be ensured for wider class of ICA models and source distributions. The proofs given in this paper stem from the results in [KLR73, Com94, TJ99]. This paper is organized as follows. In section 2, the concepts of identifiability, separability and uniqueness are discussed. Various representations for the linear ICA model are considered as well. In section 3, a theorem for ensuring the identifiability of mixing system in the linear ICA model is given. In section 4, theorem on separability is presented. Separability ensures that the source signals may be recovered up to some ambiguities. In section 5, the uniqueness of the ICA model is addressed. It is a very relevant concept, in particular for underdetermined source separation problems. Finally, section 6 concludes the paper and proofs of the theorems are given in an Appendix.

[1]  Jean-Francois Cardoso,et al.  Super-symmetric decomposition of the fourth-order cumulant tensor. Blind identification of more sources than sensors , 1991, [Proceedings] ICASSP 91: 1991 International Conference on Acoustics, Speech, and Signal Processing.

[2]  Pierre Comon,et al.  Independent component analysis, A new concept? , 1994, Signal Process..

[3]  Ruey-Wen Liu,et al.  General approach to blind source separation , 1996, IEEE Trans. Signal Process..

[4]  P. Pajunen,et al.  Blind separation of binary sources with less sensors than sources , 1997, Proceedings of International Conference on Neural Networks (ICNN'97).

[5]  T. Ens,et al.  Blind signal separation : statistical principles , 1998 .

[6]  B. De Moor,et al.  ICA techniques for more sources than sensors , 1999, Proceedings of the IEEE Signal Processing Workshop on Higher-Order Statistics. SPW-HOS '99.

[7]  O. Grellier NON LINEAR INVERSION OF UNDERDETERMINED MIXTURES Pierre COMON and Olivier GRELLIER , 1999 .

[8]  Christian Jutten,et al.  On underdetermined source separation , 1999, 1999 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings. ICASSP99 (Cat. No.99CH36258).

[9]  Terrence J. Sejnowski,et al.  Blind source separation of more sources than mixtures using overcomplete representations , 1999, IEEE Signal Processing Letters.

[10]  AmariShun-Ichi,et al.  Natural Gradient Learning for Over-and Under-Complete Bases in ICA , 1999 .

[11]  Shun-ichi Amari,et al.  Natural Gradient Learning for Over- and Under-Complete Bases in ICA , 1999, Neural Computation.

[12]  Terrence J. Sejnowski,et al.  Learning Overcomplete Representations , 2000, Neural Computation.

[13]  Aapo Hyvärinen,et al.  Two Methods for Estimating Overcomplete Independent Component Bases , 2001 .

[14]  L. Vielva,et al.  UNDERDETERMINED BLIND SOURCE SEPARATION USING A PROBABILISTIC SOURCE SPARSITY MODEL , 2001 .

[15]  Anisse Taleb,et al.  An algorithm for the blind identification of N independent signals with 2 sensors , 2001, Proceedings of the Sixth International Symposium on Signal Processing and its Applications (Cat.No.01EX467).