Application of LMS adaptive predictive filtering for muscle artifact (noise) cancellation from EEG signals

The presence of muscle artifact (noise) affects the electroencephalograph (EEG) analysis. This paper deals with the filtering of the muscle artifact (noise) from a muscle artifact contaminated EEG, by a hybrid approach. In this, the muscle artifact component outside the EEG band is removed by lowpass filtering and the component within the EEG band by the least mean square gradient adaptive predictive filtering. Further, the effect of the muscle artifact on the parametric representation of EEG and the improvement achieved by the proposed filtering, are considered for simulated and real EEG data. The results indicate that the proposed filtering facilitates a reasonably valid parametric representation of EEG even when it is contaminated with the muscle artifact. The adaptive predictors realized by tapped delay line and lattice structures have been considered.