Adaptive Blind source Separation with Dynamical Changing Source Number

Abstract

Abstract-Analyzed the reason why the overdetermined blind signal separation of natural gradient algorithm ultimate not stably converge, constructed an adaptive blind separation algorithm suitable for dynamically changing source number. The new algorithm not only overcomes the shortcoming of eventually not stable convergence, but it also need not to delete the redundancies according to the correlation of the output components, which greatly simplifies the computational complexity of the algorithm. Randomly reduce or increase the number of source signals is still able to achieve a better separation effect, simulation results demonstrate the convergence of the algorithm and its effectiveness for blind source separation.