KICA-based feature extraction for mechanical noise data

Kernel Independent Component Analysis (KICA) which is advanced recently is a non-linear method for blind source separation (BSS). KICA can´t reduce the dimension of multidimensional data when extract its feature, that is to say, KICA can´t remove the disturbing noise in observed sample signal. For these reason, paper improved its ability to process the multidimensional data, recurring to the characteristic of dimensional reduction and noise-removing of PCA. Then paper used this method to process the mechanical noise data. Results of example show that PCA_KICA method can be used to remove the disturbing noise availably, and also to separate the original signal accurately. It has a better result compared with other feature extraction methods (such as ICA) by Amari error.