Méthodologie de Conception d'une Implémentation Matérielle d'un algorithme par l'Analyse en Composantes Indépendantes pour la Séparation de Signaux

In this paper, we proposed a hardware circuit design methodology that optimises the learning algorithm of a neural network by independent component analysis (ICA) for blind signal separation. The aim of the proposed approach is to favour the achievement of an implementation architecture of FPGA circuits that consumes less hardware resources. The use of the stopped rule criterion on the updated weight adjustments allows the optimisation of the hardware architecture while keeping optimal performance of blind signal separation. The optimisation of the algorithm of independent component analysis yielded obtaining a less complex hardware architecture with an efficient adjustment of the updated weights.