An Improved Algorithm for Blind Source Separation with Newton Method
Xiaogang Deng · Computer and Information Technology · 2011
A novel algorithm for blind source separation with Newton method is presented to solve problem with both super-and sub-Gaussian sources.Instead of using one nonlinear function,an alternative switching criterion using the kurtosis to select the different nonlinear functions is proposed for Newton method learning algorithm.Computer simulation results showed that,the algorithm can be effective in blind separation of mixed super-and sub-Gaussian sources,it can converge speedily.In addition,the separation efficiency was improved relative to the Ext-Informax algorithm.