A feedback neural network with Hebbian's learning rule which used to realize blind separation of wide-band signals for near-field situations was described. Then the convergence of algorithm and the stability of equilibrium states were discussed. Finally computer simulations were also presented.