High performance in the elbow joint angle estimation using electromyography (EMG) is important in the devices based on EMG control. EMG is bioelectric which generated from muscle when the muscle performs a contraction. EMG signal can provide an information of the kinematic parameters such as angle, force, and torque of the joint of human limbs. A digital signal processing (DSP) is required to obtain those parameters. Smith developed a pattern recognition to control a device using EMG signal. In order to obtain high accuracy in the pattern recognition, Smith (2011) explored a number of window length. Previous studies (Subasi, 2012; Chu, 2006) preferred to use a fixed window length in the feature extraction process. However, mostly, previous studies investigated the window length for machines learning to classify a number of pattern into some classes. Therefore, choosing a correct windowing technique is important in the feature extraction process in order to obtain high performance in estimating the parameters (angle, force, and torque of elbow joint). The purpose of this study is to evaluate the effect of the windowing technique to the performance of the elbow joint angle estimation using EMG signal.

Materials and Method. In order to perform the purpose of this study, four male subjects were involved. The EMG signal was collected only from bicep muscle using disposable electrode (Ag/AgCl) while the elbow performed a flexion and extension motion. To estimate the elbow joint angle, the DSP of the EMG signal was performed. The DSP consisted of rectifying, normalizing, feature extraction, and a lowpass filtering. The EMG signal is extracted using twelve of time-domain features. Those time domain features were the root mean square (RMS), integrate EMG (IEMG), variance (VAR), mean absolute value (MAV), logarithmic (LOG), waveform length (WL), average amplitude change (AAC), difference absolute standard deviation value (DASDV), zero crossing (ZC), sign slope change (SSC), Wilson amplitude (WAMP), and myopulse percentage rate (MYOP). Windowing technique in feature extraction process used are adjacent and overlap window. In this study, various window length used to evaluate the performance of the windowing technique are 50, 100, 150, 200, 250, 300, 350, 400, 450, and 500 samples. In the overlap windowing technique, the percentage of overlap are 10, 20, 30, 40, 50, 60, 70, 80, and 90%. The effect of window length and percentage of overlap to the performance of the elbow joint angle estimation was evaluated by calculating the root mean square error (RMSE) and Pearson’s correlation coefficients.

The results showed that the window length and percentage of overlap in windowing technique affected the performance of the elbow joint angle estimation for all time-domain features. In the adjacent windowing technique, a window length of 100 samples has the highest performance in estimation. In the overlap windowing technique, the percentage of overlap of 10% has the highest performance of the estimation. Features ZC, SSC, WAMP, and MYOP have the best RMSE and Pearson’s correlation coefficients. In the window length of 100 samples, the mean and standard deviation of RMSE and correlation coefficients for feature ZC, SSC, WAMP, and MYOP are 14.41°±3.86° and 0.98±0.008, 15.20°±4.59° and 0.97±0.015, 14.50°±3.33° and 0.98±0.008, and 13.56°±3.53° and 0.97±0.0085, respectively.

Conclusion. The results of the study suggested that window length of 100 sample are effective in the feature extraction process for elbow joint angle estimation using EMG signal. Feature ZC, SSC, WAMP, and MYOP have higher performance than the other features.