Blind signal extraction using self-adaptive nonlinear Hebbian learning rule

A simple neural network with local, biologically plausible , non-linear Hebbian learning rule is developed to perform sequential extraction of independent signals from a linear mixture of them. Instead of separating all signals simultaneously, we rather extract them sequentially using a new de BLOCKINation technique. It is shown that with suitable designed nonlinear functions and applying self-normalization principle, we are able to extract source signals with (specied) predened order (e.g., extracting signals according to the maximum absolute value of normalized kurtosis). Computer simulation experiments conrm the validity of the proposed algorithm and demonstrate that it is useful extension of existing neural network solutions to blind separation problem as extraction of particular source signals with predened stochastic characteristic is possible.