Estimation of wavelet threshold value for surface EMG baseline removal

Abstract

High quality of surface electromyography is vital during investigation on muscle activity. Low quality of surface EMG signals causes extracted signals to be inaccurate and lead to misinterpretation and misclassification of the signals. A surface EMG signal quality is determined by the ratio of muscle contraction to its baseline during muscle relaxation period. Baseline noises are originated from powerline, cable motion artefact, electronics of the amplification systems and skin-electrode interface. The noises are quite difficult to be removed by digital or active filter since they do not have specific frequency range like powerline interference and corner frequency noise. However, wavelet de-noising enables users to remove noise by accessing both frequency and time information. Baseline surface EMG noise is possible to be removed by estimating de-noise threshold based on mean absolute value and root mean square of its baseline. The result of this study shows that the proposed estimation of threshold method is better than the conventional threshold setting.

A New Threshold Estimation Method of SEMG Wavelet De-noising for Prolonged Fatigue Identification

“…This method is a combination of the research presented in [15] and [11]. The method is estimated to remove corner frequency and baseline noises.…”

“…Quality of SEMG signals determined by their signal-to-noise ratio (SNR) and baseline value [15]. Smaller baseline values indicate higher quality SEMG signals.…”

“…The baseline noises may come from powerline interference, cable motion artefact, electronics circuit design and skin-electrode interface [16]. Filtering baseline noises through digital or active filter is not possible since there is no specific frequency range and it may interfere at strong SEMG energy frequency range [15]. Thus, time-frequency wavelet de-noising method can be employed to remove irrelevant signals.…”

“…In general applications, there are conventional estimation methods commonly used such as Stein's Unbiased Risk Estimate (SURE), minimaxi and universal threshold method. These thresholds are estimated based on statistical approach to minimize risk and error in eliminating unwanted signals [15]. However, these methods are not suitable for SEMG because it tend to remove significant SEMG information and keeping noises [16], [17].…”

“…Previous research in [11] demonstrates that corner frequency noises can be removed through wavelet denoising, but did not demonstrate how to remove baseline noises. The research in [11] was extended to remove baseline noises using the similar technique and presented in [15]. Thus, this paper will combine both techniques to remove corn er frequencies and baseline noises holistically through wavelet de-noising.…”