Successive variational mode decomposition

Abstract

Variational mode decomposition (VMD) is a powerful technique for concurrently decomposing a signal into its constituent intrinsic modes. However, the performance of VMD will be degraded if the number of modes available in the signal is not precisely known. In this paper, we introduce a new method, namely successive variational mode decomposition (SVMD), which extracts the modes successively and does not need to know the number of modes. The method considers the mode as a signal with maximally compact spectrum, as VMD does. It achieves the mode decomposition by adding some criteria to the optimization problem of VMD: the mode of interest has no or less spectral overlap to the other modes and to the residual signal. Our simulations on some artificial and real world data have demonstrated that the new method without knowing the number of modes converges to the same modes as VMD does with knowing the precise number of modes. Moreover, the computational complexity of SVMD is much lower than that of VMD. Another advantage of SVMD over VMD is more robustness against the initial values of the center frequencies of modes.