The experimental demonstration of a multichannel time-series myoprocessor: system testing and evaluation.

A multichannel time-series myoprocessor which combines the advantages of the parallel filtering limb function classifiers of Doerschuk et al. and the "optimal myoprocessor" muscle force estimators described by Hogan et al. have been developed and evaluated experimentally. Magnitudes and directions of knee movements were identified accurately and robustly from EMG sites intermediate to the major thigh muscles of intact individuals. Electrode placement criteria were tested, and system performance and sensitivity to contraction level as functions of channel number were computed. By including spatially distributed information into the structure of the processor, gains in accuracy and reliability over systems with fewer channels were demonstrated. Operating range increased with the number of channels included in the processor. Joint moment was estimated from multiple channels of temporally correlated data, extending and generalizing previously reported techniques. Identifying the parameters of AR models of the EMG at low levels of contraction resulted in more robust classification and joint moment estimation. Optimal electrode position could not be predicted a priori. The system may be applicable to the proportional control of myoelectric prostheses.