Mean frequency estimation of surface EMG signals using filterbank methods

This paper focusses on the accurate estimation of the Mean Frequency of surface electromyogram (EMG) signals during voluntary isometric contractions. This particular type of analysis is commonly used by kinesiologists to gain important information relating to muscle fatigue. These EMG signals are typically processed to extract theMean Frequency (MNF) and studies often follow how these parameters evolve through time. Traditional approaches to estimate the MNF variables are based on the periodogramor Burg's autoregressive approach, but these methods suffer from a high degree of variability due to the choice of window size and/or significant bias in frequency estimation due to other inherent limitations. In this paper we propose the use of a data-adaptive filterbank spectral analysis technique, namely the Power Spectrum Capon (PSC) to overcome the problems associated with the traditional methods. This new method is shown to provide significant reductions in MNF parameter bias and variability over a wide range of data window sizes. Experiments are performed on simulated data with known spectral characteristics in order to compare the relative performance of the different techniques. This paper follows on from previous work by the authors showing that the filterbank methods outperform currently used methods in terms of consistency on real patient data.