A kind of independent component analysis（ICA） method with global optimal property is proposed

which is based on low frequency part of wavelet transformation as moving factors and built a signal-to-noise ratio objective function on predictability error of denominator.Optimizing leads to get generalized eigenvalue and eigenvector which is the demanding separable matrix.Iteration is not necessary.The experiments show that computation is small and separable precision is higher

which can respectively get resembled coefficients of 1 and 0.999 8.Meanwhile it can choose wavelet basis flexibly based on different requirement.