Signal extensions in independent component analysis and its application for real-time processing
Shuxue Ding, Jie Huang, Daming Wei, Sadao Omata · 2004
In this paper, we investigate some issues related to realtime processing for independent component analysis (ICA), based on gradient learning with simultaneous perturbation stochastic approximation (SPSA). Real-time ICA processing is especially necessary for an application in dynamic mixing environment, since a batch type of ICA processing can work well only in a static or stationary mixing environment. Although there are many choices for an ICA object function to which SPSA can be applied, in this paper, we choose a diagonality of the nonlinear correlation matrix as our object function. Theories and implementations of the algorithm are described. Results of computer simulation are also presented to demonstrate the effectiveness.