An efficient measure of signal temporal predictability for blind source separation

作者：Ye, Mao^*; Li, Xue

来源：Neural Processing Letters, 2007, 26(1): 57-68.

DOI：10.1007/s11063-007-9042-0

• blind source separation
• principal component analysis
• neural networks
• eigenvector

An efficient measure of signal temporal predictability is proposed, which is referred to as difference measure. We can prove that the difference measure of any signal mixture is between the maximal and minimal difference measure of the source signals. Previous blind source separation (BSS) problem is changed to a generalized eigenproblem by using Stone's measure. However, by using difference measure, the BSS problem is furthermore simplified to a standard symmetric eigenproblem. And the separation matrix is the eigenvector matrix, which can be solved by using principal component analysis (PCA) method. Based on difference measure, a few efficient algorithms have been proposed, which are either in batch mode or in on-line mode. Simulations show that difference measure is competitive with Stone's measure. Especially, the on-line algorithms derived from difference measure have better performance than those defived from Stone's measure.

• 出版日期2007-8
• 单位电子科技大学