ROBUST NEURAL NETWORKS WITH ONLINE LEARNING FOR BLIND IDENTIFICATION AND BLIND SEPARATION OF SOURCES

Two unsupervised, self-normalizing, adaptive learning algorithms are d eveloped for robust blind identification and/or blind separation of in dependent source signals from a linear mixture of them. One of these a lgorithms is developed for on-line learning of a single-layer feed-for ward neural network model and a second one for a feedback (fully recur rent) neural network model. The proposed algorithms are robust, effici ent, fast and suitable for real-time implementations. Moreover, they e nsure the separation of extremely weak or badly scaled stationary sign als, as well as a successful separation even if the mixture matrix is very ill-conditioned (near singular). The performance of the proposed algorithms is illustrated by computer simulation experiments.