Removing power line noise from recorded EMG

Three methods for offline removal of power line interference (hum) from electromyograms (EMGs) were compared using both simulated and recorded EMG signals. The first method was a simple recursive digital notch filter. In the second method (Regression-Subtraction), the amplitude and phase of the interference were estimated by regressing sine and cosine functions onto a 'quiet period' before the start of the muscular contraction. A sinusoid with this frequency, magnitude and phase was then subtracted from the entire length of the signal. In the third method (Spectrum Interpolation), it was assumed that the magnitude of the original component of the signal at the frequency of the interference can be approximated by interpolating between the adjacent frequency bins in the power spectrum. While Regression-Subtraction was found to give the highest SNR for the output signal under ideal conditions, Spectrum Interpolation was found to be comparable if the phase of the interference was not constant and superior if the interference contained strong harmonic components.