Partial noise assisted multivariate EMD: An improved noise assisted method for multivariate signals decomposition. (July 2017)

• Record Type: Journal Article
• Authors: Huang, Weiping; Zeng, Jing; Wang, Ziyang; Liang, Jun
• Is Part Of: Biomedical signal processing and control. Volume 36(2017)
• Page Start: 205
• Page End: 220
• Publication Date: 2017-07
• Subjects: Multivariate EMD -- Noise assisted MEMD -- Motor imagery -- Multivariate signal -- Mode mixing -- Direction vector number; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2017.04.003
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An improved partial noise assisted multivariate empirical mode decomposition (PNA-MEMD) method has been proposed. High-frequency band-limited noise is adopted in the proposed PNA-MEMD method. Our new method improves the decomposition accuracy and computational efficiency in contrast with standard NA-MEMD. Experiment on motor imagery EEG data indicates that the proposed method performs well in locating basis frequencies. Abstract: Empirical mode decomposition (EMD) is widely used in the decomposition of non-stationary signals, contributing to accurate time-frequency analysis in different scales. With the increasing demand for dealing with multivariate signals, it's of practical significance to extend univariate EMD to multivariate EMD (MEMD). Following the MEMD, a noised assisted MEMD (NA-MEMD) method has been proposed recently to restrain the mode mixing phenomenon. However the standard NA-MEMD method is prone to obtaining unstable performance, and suffering high computational complexity. In this paper, a partial noise assisted multivariate EMD (PNA-MEMD) method is proposed to resolve the problems. In PNA-MEMD, high-frequency band-limited noise instead of the original white noise is utilized to assist the decomposition process. The numerical simulations and application to motor imagery EEG data demonstrate the superiority of proposed method, in contrast with the standard MEMD and NA-MEMD.