Using the discrete wavelet transform for time-frequency analysis of the surface EMG signal

The frequency content of the surface electromyographic (SEMG) signal is used to study neural activity, and force development and fatigue in muscle. The fast Fourier transform (FFT), or short time Fourier transform (STFT), are commonly used to determine the frequency content of the SEMG, but have the drawback of assumed signal stationarity. A relatively new technique, the wavelet transform (WT), is well suited to nonstationary signals, and has gained widespread use in speech and image processing. We applied the discrete wavelet transform (DWT) based on the Daubechies wavelet to SEMG data. The DWT decomposed the SEMG into 11 time-frequency bands; the data was also processed with an FFT algorithm. Comparison of these results show that the DWT provided information in the correct frequency bands. These results are encouraging, as time-frequency signal decomposition will allow movement and force generation patterns to be directly related to SEMG frequency components. The main disadvantage of the DWT seems to be that because the signal is down sampled at each successive DWT scale, the transform is sparse at lower frequency scales. However, we believe that the continuous discrete wavelet transform will overcome this deficiency and provide an additional method of SEMG frequency analysis.