Mechanomyography signals processing method using multivariate variational mode decomposition

Abstract

With the development of human-computer interaction technology, the wearable power-assisted robots gradually change from passively accepting users' instructions to actively recognizing users' intentions. More and more wearable power-assisted robots use human Mechanomyography (MMG) signals to recognize users' motion intentions. There is still a bottleneck problem in how to remove motion artifacts in the MMG signals. In this study, we propose the MMG signals processing method using Multivariate Variational Mode Decomposition (MVMD). The multi-channel MMG signals of the same muscles group are decomposed into several the Bandwidth-limited Intrinsic Mode Function (BIMF) components. Combined with the frequency and energy distribution of the MMG signal, the BIMF components are selected. The selected BIMF components are summed to obtain the MMG signals filtering noises such as motion artifacts. Compared with the Multivariate Empirical Mode Decomposition (MEMD), the Instantaneous Frequency (IF) of each channel BIMF components decomposed by MVMD is more stable, and the degree of mutual aliasing of the components is significantly reduced. The MMG signals energy is maximized while the noises such as motion artifact are filtered. The recognizability of the MMG signals is improved.