SNR estimation in EMG signals contaminated with motion artifact

Abstract

An electromyography (EMG) recognition system is essential for enabling a variety of applications. However, motion artifact contaminated with the EMG signal as it passes through or by various tissues may degrade the recognition performance. We present the algorithm for signal-to-noise ratio (SNR) estimation in EMG signals contaminated with motion artifact. Six features derived from the EMG signals are used as the neural network input: skewness (SKEW), kurtosis (KURT), mean absolute value (MAV), wavelength (WL), zero crossing (ZC), and mean frequency (MNF). The estimated SNR values are the neural network output. The best mean and standard deviations of the correlation coefficient (CC) between the actual and estimated SNR values are provided by the MNF $(0.9699 \pm 0.0076)$. Future research may concentrate on determining SNR values using real EMG signals in their natural surroundings.