An algorithm study for FES evoked artifacts removal based on the combination of CEEMDAN and GS

Volitional electromyography (vEMG) used in motion intention driven functional electrical stimulation (FES) is often contaminated by stimulation artifact and M-wave. It is necessary to extract vEMG signals from the contaminated signals. This paper presents an algorithm combining complete empirical mode decomposition with adaptive noise (CEEMDAN) with Gram Schmidt (GS) to remove the electrical stimulation artifact and muscle response from raw surface EMG signals. First, the raw surface EMG signal recorded during FES was decomposed into several intrinsic mode functions (IMFs) using the CEEMDAN method. Next, energy entropy analysis was performed on each IMF component, classifying them into two categories: effective signal dominant and artifact noise dominant. Then, the GS de-noising method with the Laida criterion was applied to process the artifact noise dominant IMFs and the initial spike. Finally, we reconstruct the signal using the CEEMDAN method, obtaining vEMG signal. To validate the proposed method, semi-simulated signals were generated by fusing the recorded pure EMG signal and artifacts evoked by stimulation current with different current amplitude and frequency. The validity of the proposed method was assessed by analysis of normalized root mean square (RMS) and signal to noise ratio (SNR) values between the original EMG signal and the extracted vEMG data. All results showed that modified CEEMDAN-GS is capable of extracting vEMG signals during FES with time-variant parameters.