Natural gradient for temporally dependent component analysis
Zhenwei Shi, Dalong Cheng, Xueyan Tan, Zhiguo Jiang · 2010 Sixth International Conference on Natural Computation · 2010
The temporally dependent component analysis (TDCA) method for blind source separation (BSS) is introduced. As a new principle, it is shown that maximizing the mapping of autocorrelation of source signals can be used to perform BSS. We use the natural gradient algorithm for TDCA and study the mathematical properties of TDCA. Simulations by square temporal autocorrelation sources verify the efficient implementation of the proposed method.