A New EMG Decomposition Framework for Upper Limb Prosthetic Systems

Abstract

Neural interfaces based on surface Electromyography (EMG) decomposition have been widely used in upper limb prosthetic systems. In the current EMG decomposition framework, most Blind Source Separation (BSS) algorithms require EMG with a large number of channels (generally larger than 64) as input, while users of prosthetic limbs can generally only provide less skin surface for electrode placement than healthy people. We performed decomposition tests to demonstrate the performance of the new framework with the simulated EMG signal. The results show that the new framework identified more Motor Units (MUs) compared to the control group and it is suitable for decomposing EMG signals with low channel numbers. In order to verify the application value of the new framework in the upper limb prosthesis system, we tested its performance in decomposing experimental EMG signals in force fitting experiments as well as pattern recognition experiments. The average Pearson coefficient between the fitted finger forces and the ground truth forces is 0.9079 and the average accuracy of gesture classification is 95.11%. The results show that the decomposition results obtained by the new framework can be used in the control of the upper limb prosthesis while only requiring EMG signals with fewer channels.