Simplified Optimal Estimation of Time-Varying Electromyogram Standard Deviation (EMGσ).

Abstract

Previous works have shown that whitening improves the processed electromyogram (EMG) signal for use in end applications such as EMG to torque modelling. Traditional whitening methods fit each subject from calibration contractions, which is a hindrance to their widespread use. To eliminate this cumbersome calibration, a universal whitening filter was developed using the whitening filters from a pre-existing data set (64 subjects, 8 electrodes/subject). Since the shape of each subject-specific whitening filter was observed to be relatively consistent across subjects, the universal whitening filter was formed as their ensemble average. The processed EMG was then used to model surface EMG to torque about the elbow. Traditional and universal whitening provided the same EMG-torque benefit, each improving statistically over unwhitened processing by ~14% during dynamic contractions. We further studied the use of root difference of squares (RDS) post-processing to attenuate additive measurement noise in EMG channels. With and without whitening, RDS processing (vs. no RDS processing) better attenuated additive noise, reducing it from 2-4% (on average) of the processed EMG from a 50% contraction down to < 1%. The combined use of universal whitening filters and RDS processing should be a particular benefit in real-time applications such as prosthesis control.