In this paper, a novel Successive Multivariate Variational Mode Decomposition (SMVMD) is presented. Different from most existing multichannel signal decomposition approaches, the proposed SMVMD does not need to predefine the mode number and is able to extract the joint or common modes successively. Firstly, a general decomposition form for multichannel multicomponent signals is formulated based on the instantaneous linear mixing model, which is commonplace in the blind source separation (BSS) problem. Then, four key criteria are introduced to establish the successive variation optimization function. Finally, the alternate direction method of multipliers (ADMM) algorithm is employed to solve this optimization problem. The effectiveness and advantages of the proposed SMVMD are demonstrated by a series of numerical examples. The utility of the proposed approach is also highlighted by the analysis of real-life EEG and vibration signals.

在本文中，提出了一种新的连续多元变分模式分解（SMVMD）。与大多数现有的多通道信号分解方法不同，所提出的 SMVMD 不需要预定义模数，并且能够连续提取联合或共模。首先，基于瞬时线性混合模型制定了多通道多分量信号的一般分解形式，这在盲源分离（BSS）问题中很常见。然后，引入四个关键标准来建立连续变化优化函数。最后，采用乘法器交替方向法（ADMM）算法来解决该优化问题。一系列数值例子证明了所提出的 SMVMD 的有效性和优势。