Instantaneous Frequency Analysis of EMG Signals during Isometric Exercise of the Paraspinal muscles

Introduction The surface Electromyography (EMG) signal detected during a voluntary isometric muscle contraction is a random, non-stationary signal. It is well known that during a sustained contraction the EMG signal becomes progressively ‘slower’. This is a manifestation of muscle fatigue and can be quantified by monitoring specific variables, which indicate the change that occurs over time. The initial value and the rate of decrement of spectral variables, such as the Median Frequency (MDF) and the Mean Frequency (MNF), have been widely used for this purpose. They are commonly assessed from the intercept and slope of a Least Squares Error linear regression of the MDF or MNF data (Duchene and Goubel, 1993, Merletti and Conte, 1995). However, it should be noted that when estimating the power spectral density of a random signal, the latter is assumed to be stationary over the analysis time window. Thus, the EMG signal is effectively considered quasi-stationary, which makes the selection of the epoch length an issue that requires careful investigation. The choice of the size of the analysis window is also important for the Power Spectral Density (PSD) estimation, and it is known that small windows tend to give inaccurate results. On the other hand, longer windows violate the condition of quasi-stationarity and increase the variance of the estimates of the intercept and slope of the regression line (Farina and Merletti , 2000). In this paper, a new approach for monitoring the signal’s frequency descent is proposed, which is based on the concept of the Instantaneous Frequency. The new method is compared to the traditional MNF and MDF methods and is shown to be superior in many respects. In particular, the slope and point of intercept of the regression line seems to be almost unaffected by the epoch length. Furthermore, it exhibits smaller variation around the regression line. It should be noted that the proposed method does not require any spectral estimation, neither does it necessitate any quasi-stationarity assumption.