Blind separation of three binary sources from one nonlinear mixture
Konstantinos Diamantaras, Théophilos Papadimitriou · 2010
This paper presents a blind method for the separation of three binary sources from a single, nonlinear mixture. Since the problem is intractable, in general, we focus on the special case where the nonlinearity is an odd function. The proposed method is based on clustering of the observed data and the geometry of the cluster centers. The separation algorithm is analytical, it does not involve iterative optimization and it is computationally efficient since it involves the inversion of a small matrix. Due to the structure of the problem, the true sources are extracted together with spurious signals adding one more indeterminacy to the usual sign and order indeterminacy of the sources. However, in some applications, eg. the separation of binary images, this indeterminacy can be resolved by visual inspection.