An adaptive on-line algorithm for independent component analysis

This paper presents an adaptive on-line algorithm whose optimization criterion is based on maximum likelihood to perform Independent Component Analysis (ICA). A fast density estimation method is introduced , it can quickly achieve the true score functions of the unknown sources by estimating from the sample. In the meantime, the careful selection of the step size is often necessary to obtain good performance for the source separation tasks. We carry out the global minimum of the contrast function with the gradient adaptive step size. The results of simulation experiment show that the provided algorithm can perform the adaptive separation of real digital signal efficiently.