With recent advancements in wearable sensors, wireless communication and embedded computing technologies, wearable EMG armbands are now commercially available and accessible to most laboratories. Due to the embedded system constraints, however, these armbands typically sample EMG signals at a lower frequency (e.g. 200 Hz for the Myo armband) than professional versions. It remains unclear whether existing EMG feature extraction methods, which have largely been developed based on EMG signals sampled at the Nyquist rate (generally 1000 Hz) or above, are still effective for use with these emerging lower-frequency systems. In this study, we investigate the effects of sampling rate (low: 200 Hz vs. high: 1000 Hz) on performance in classifying eight classes of hand motion in 20 able-bodied subjects for eleven commonly used time-domain features. The effect of within- and between-day variation on the performance of EMG features was also investigated. The results show that classification accuracies drop significantly with the lower sampling rate for all of the evaluated features, when using either a support vector machine or a linear discriminant analysis classifier. Furthermore, the within-class variability increased significantly with reduced sampling rate, although the level of inter-session repeatability was not affected. In comparing the performance of single features, waveform length outperformed the others for both high- and low-sampling rates. The optimal feature sets found using sequential forward selection for the two sampling rates, however, were found to be different. These results suggest that feature selection results for myoelectric control, previously determined using EMG data sampled at 1000 Hz, may not directly apply to this new generation of low-sampling rate wearable EMG sensors.

A feature extraction issue for myoelectric control based on wearable EMG sensors