A Novel Filtering Framework for High-Density sEMG Based on Variational Mode Decomposition and Independent Vector Analysis

Abstract

During the acquisition process, surface electromyography (sEMG) recordings are unavoidably contaminated by different types of noise signals, including baseline noise (BLN), powerline interference (PLI), and white Gaussian noise (WGN). The inclusion of these noise signals significantly diminishes the quality of sEMG signals and impairs the accuracy and resilience of their subsequent utilization. There are two significant issues concerning existing filters: 1) the primary technique for filters targeting BLN and PLI remains IIR filters. However, this methodology makes it challenging to prevent the filters from suppressing sEMG signals in the stopband and 2) the noise signals that are filtered out by the filters targeting WGN do not adhere to a Gaussian distribution. In this study, we propose a novel filtering framework that combines split spectrum processing (SSP) and mix source separation (MSS) to effectively eliminate the three types of noise signals and address the two aforementioned concerns. In this article, other four well-designed filtering methods, including the infinite impulse response (IIR) filter, ensemble empirical mode decomposition (EEMD) method, variational mode decomposition (VMD) method and the independent vector analysis (IVA) method, were evaluated for comparison. The proposed filtering framework demonstrated superior performance in eliminating all three types of noise signals. The simulated signals showed an improvement in SNR of 29.7, 22.4, and 12.9 dB for sEMG signals corrupted by BLN, PLI, and WGN with input SNRs of −10 dB, respectively. The experimental results indicated that the proposed method achieved an average improvement in SNR of 15.9 dB. The proposed filter is highly effective in eliminating all three types of noise and can be utilized for various applications that necessitate sEMG signals, such as gesture recognition and sEMG decomposition.