사람의 움직임 의도를 파악하기 위한 표면근전도 시스템 개발

Abstract

In this work, design of Surface ElectroMyoGram (sEMG) interface to decode human motion intentions and pattern recognition algorithm for myoelectric control are introduced. There are many applications using bioelectric signals, nonetheless, understandings of instrument design are insufficient. This thesis try to summarize properties of sEMG and introduce how to design sEMG preamplifier circuit using these properties. Amplifying and filtering are main parts of the bioelectric amplifier. In order to enhance the effect of filter, two filtering methods are introduced in this thesis. One is the most well-known filter named sallen-key filter and another is directly coupled quasi-high-pass filter which is comprised of adding a capacitor into instrumentation amplifier to improve common mode rejection ratio (CMRR) of instrumentation amplifier. Furthermore, the effort to eliminate ground loop which is a big problem when an entry-level student designs analog circuit is discussed and characteristics of electrode are represented. An attempt to connect two different grounds directly and an attempt to connect different ground-level systems are generated undesired signal. Having experimented the development of EMG system, using dry electrode is not easy since impedance of dry electrode is more higher than that of wet electrode. This means that having more higher input impedance preamplifier is needed. In practical issue session, these problems and solutions will be introduced. The sEMG signal is influenced by electrode shape and interelectrode distance. Therefore, electrode issue is also important. In the last work, fast learning online classifier using sEMG signal from eight electrodes on forearm is introduced. This algorithm is introduced by Park et al. and attempt to apply online remote myoelectric control system using own made preamplifier have been introduced. In this work, preamplifier introduced in the second chapter will be verified using this algorithm through five subjects. Eight skin surface electrodes were placed on a forearm to detect the sEMG signals corresponding to four different hand motions and the rest is presented. In order to enhance performance of the classifier, feature extraction using class information has been developed. The randomly assigned non-update learning method achieves high speed classifier learning. Algorithm was verified using fifth data set gathered from each subject for training set and assessment set.