EMG processor based on the amplitude probability distribution

The EMG signal is being used increasingly for the control of prostheses and for investigating and retraining human movements. The raw (unprocessed) EMG is normally unsuitable for such applications and some processing is necessary. Due to the statistical characteristics of the EMG, it is usually difficult to achieve a processing delay of less than 100 ms. We introduce here a new EMG signal processing technique, implemented on an 8-bit microprocessor. It uses changes in the amplitude probability distribution of the EMG signal to discriminate between a number of tension levels in an individual muscle. The program employs procedures often used in pattern and speech recognition, in that it is trained to identify several tension levels. With a processing delay of 100 ms, on-line tests show that the new processor achieves a recognition rate of 84.8% when discriminating between five tension levels (including relaxation) in the biceps brachii. When the number of tension levels is reduced to four, the recognition rate becomes 96.7%, which compares well with other similar systems, some of which were tested in parallel with the new technique.