Read/Search this Article
Reflexive feedback is necessary when the Functional Electrical Stimulation (FES) system is considered as a means to restore function to spinal-cord-injured patients. Muscle tension feedback is required if movement smoothness and regulation of joint torque are to be maintaind under variable loading. Tension feedback could be simply performed in laboratory animals if a force transducer is placed in series with the muscle. A similar approach, however, is unlikely if human subjects are considered. With the advent of the technology associated with the analysis of the electromyographic signal (EMG), the possibility of using it as a feedback variable representing tension is highly promising. For such feedback schemes to be performed, the EMG-tension relationships need to be determined. Much has been reported regarding the EMG-joint force relationships during voluntary contraction. However, systematic analysis relative to the EMG-tension relationship during electrical stimulation has not been reported, because some problems of crosstalk and noise exist when stimulation was applied to the muscle with surface electrodes. We propose a system eliminating crosstalk of the unexpected stimulus artifact from the EMG during electrical stimulation, and a method estimating the tension from the EMG in real time by using the auto-regressive moving average (ARMA) model. In order to test the possibility of use of EMG as a feedback variable for the closed-loop and tension-regulated FES system, we constructed a control system which used the estimated tension from EMG as a feedback signal, and carried out an experiment with normal human subjects. In the system for estimating tension, we used a relay connecting EMG amplifier and A/D converter to eliminate crosstalk noise from EMG. The relay served to turn the input terminal of A/D converter to the ground level only when stimulation was applied. The experiment was carried out with upper extremity muscles of three normal human subjects. The muscle tension elicited by the stimuli was measured as the isometric torque around the elbow joint axis. The EMG signal was sampled during a period of 4 seconds at 250Hz sampling frequency. In order to estimate the muscle tension within 20msec, we used a second-order ARMA model. To evaluate the results of the estimated tension, Power Normalized Error (PNE) was obtained as one of criteria for suitable estimation. The PNE was defined as the power of tension error between the estimated and measured values, which was normalized by the power of measured tension. The PNEs obtained from the results were almost within a 10% margin of error. Besides, we tried to use the estimated tension from EMG as a feedback parameter for a closed-loop, tension-regulated system. The validity of the present system eliminating crosstalk artifacts of electrical stimulation on EMG was proved, and the muscle tension from the EMG signal could be estimated in real time. The possibility of use of EMG as a feedback parameter for the closed-loop system was also proved.