The matrix structure has an important influence on the signal processing effect of the singular value decomposition (SVD), and the idea in traditional algorithm that matrix structure should be fixed is not adopted, while the varying matrix structure is proposed to be used in SVD. For every SVD, the matrix structure is changed once to adapt the different periodical components of the original signal, and then in every decomposition the signal of the previous layer is decomposed into the principal component and secondary component. The secondary component is retained, while the principal component is decomposed by the varying matrix structure SVD again, so repeatedly. Finally the original signal is decomposed into a series of principal and secondary components. The signal processing examples show that this method has a good separation effect on the different periodical components in the original signal.