Optimal EMG amplitude detectors for muscle-computer interface

To develop an advanced muscle-computer interface (MCI) based on surface electromyography (EMG) signal, a suitable signal processing and classification technique has a key role to play, particularly the selection of EMG features. Two sufficient and well-known methods to extract signal amplitude are root mean square (RMS) and mean absolute value (MAV). Their classification performance is comparable to an advanced and high computational time-scale feature, e.g. discrete wavelet transform. The performance of RMS and MAV, however, depends on a probability density function (PDF) of EMG signals, i.e., Gaussian or Laplacian, and the PDF of motions associated with EMG signals is still not clear yet. In addition, both features provide the same distribution in feature space, thus only one of them should be used to avoid redundancy in a classification scheme. This study investigated the PDFs of eight hand, wrist and forearm motions and then estimated the signal-to-noise ratio (SNR), defined as a mean value divided by its fluctuation, of both amplitude detectors. On average, the experimental EMG density was closer to the Laplacian density, and MAV had slightly higher SNR than RMS for both forearm extensor and flexor muscles and both genders. Lastly, the accuracy of both features in MCIbased EMG classification was reviewed. For MCI applications, MAV is recommended to be used as an optimal EMG amplitude detector.