The prosthetic hand has been aimed to restore hand functions by estimating the user’s intention via bio-signal and providing sensory feedback. Surface electromyogram (sEMG) is a widely used signal, and transcutaneous electrical nerve stimulation (TENS) is a promising method for sensory feedback. However, TENS currents can transmit through the skin and interfere as noise with the sEMG signals, referred to as “Artifact,” which degrades the performance of intention estimation.
In this paper, we proposed an adaptive artifact removal method that can cancel artifacts separately across different frequencies and pulse widths of TENS. The modified least-mean-square adaptive filter uses the mean of previous artifacts as reference signals, and compensate using prior information of TENS system. Also temporal separation for artifact discrimination is applied to achieve high artifact removal efficiency. Four sEMG signals—two from flexor digitorum superficialis, flexor carpi ulnaris, extensor carpi ulnaris—was collected to validate signals both offline and online experiments.
We validated the filtering performance with twelve participants performing two experiments: artifact cancellation under variable conditions and a real-time hand control simulation called the target reaching experiment (TRE). The result showed that the Signal-to-Noise Ratio (SNR) increased by an average of 10.3dB, and the performance of four TRE indices recovered to the levels similar to those without TENS. The proposed method can significantly improve signal quality via artifact removal in the context of sensory feedback through TENS in prosthetic systems.