A two-stage adaptive multivariate variational mode decomposition method for noise reduction

Abstract Variational Mode Decomposition (VMD) is a powerful signal decomposition technique for noise reduction. Its generic extension version Multivariate Variational Mode Decomposition (MVMD) possesses unique advantages in noise suppression across multiple channels. However, the performance of these techniques is highly sensitive to the initial parameters. To address the challenge of initial parameter selection in MVMD, this paper proposes a two-stage adaptive MVMD method based on a binary search strategy. In the first stage, based on the convergence properties of the initial center frequency, we propose a binary-search strategy that traverses the entire frequency space to rapidly obtain mode functions (CMFs-1) corresponding to the convergence center frequencies, in which the intrinsic mode functions (IMFs) are included. In the second stage, to eliminate spurious mode functions (SMFs) caused by noise, we input the CMFs-1 as initial parameters into the balance factor adaptively adjusted MVMD method to obtain the corresponding modes (CMFs-2). By leveraging the distinct characteristics of noise and IMFs on the convergence center frequencies (CCFs) under two different acquisition methods during the mode acquisition process, SMFs are effectively filtered out by setting a threshold for center frequency variation. We compared the proposed method with existing MVMD, GWO-MVMD and recent adaptive MVMD variant (SA-MVMD). Additionally, parameter sensitivity analysis and complexity analysis of the proposed method were conducted. Simulation results demonstrate that the proposed method exhibits superior noise reduction performance both in harmonic signals and periodic impulse signals. Experimental results under several test scenarios also validate the effectiveness of the proposed method.