Blind Extraction of Nonstationary Signal with Four Order Correlation Kurtosis Deconvolution

As for nonstationary signal, such as subpixel peak detection,we could be difficult to suppress the noise of super- Gaussian and sub-Gaussian in the mixed signal with the traditional low order filter. The gradient search method is generally adopt in the filter algorithm based on higher order statistics, but it is difficult to avoid local convergence and large complexity in the gradient search process. Blind source separation method based on maximum entropy is not suitable for using correlation kurtosis to blind extraction signal. Therefore, a maximum four order correlated kurtosis deconvolution( M4CKD) algorithm is presented on the basis of the kurtosis variation of mixed signals and the corresponding inverse filter is designed. Moreover, the convergence of the algorithm are analyzed. The results shown that the super Gauss and Gauss noise of mixed signal can be effectively suppressed, and this algorithm has faster convergence speed and higher signal-to-noise ratio which compared with FastICA algorithm.