Acceptability conditions for blind source separation problems

Vincent Vigneron (1) , Sébastien Lagrange (2) , Christian Jutten (3)

1 LSC - Laboratoire Systèmes Complexes
2 LISA - Laboratoire d'Ingéniérie des Systèmes Automatisés
3 LIS - Laboratoire des images et des signaux

The Herault-Jutten (HJ) algorithm is a neuromimetic structure capable to perform blind source separation (BSS) of a linear mixture from an array of sensors without knowing the transmission characteristics of the channels, nor the inputs. The learning algorithm developed by Herault and Jutten is based on the generalized Hebb's rule in such a way that each output signal will be proportional to only one source by cancelling the influence of the other source. In this article, we show how theoretic stability conditions can be used for parameter estimation to restore the primary sources via interval computations.

Domaines: Théorie [stat.TH]; Statistiques [math.ST]

Vincent Vigneron, Sébastien Lagrange, Christian Jutten. Acceptability conditions for blind source separation problems. 11th European Symposium on Artificial Neural Networks (ESANN 2003), Apr 2003, Bruges, Belgium. pp.301-306. ⟨hal-00233005⟩