Non-distorting post-acquisition line-frequency filter for evoked potentials

Two types of waveforms were chosen to be filtered. The first was actual visual evoked potentials contaminated with 60-Hz pickup. The second was a set of simulated waveforms consisting of a sum-of-Gaussians evoked potential with varying amounts of added 60-Hz noise. Both consisted of 128 samples with a sampling frequency of 250 Hz. For digital notch filtering, an FIR filter kernel was constructed with a length of 1024 and a Blackman window. Filtering was accomplished by convolution. A series of notch widths was used. With the simulated data the FIR filter did not perform as well as the subtractive filter, especially for large contaminations. Neither distorted the waveforms, but the FIR filter left significant contamination. With real data, however both filters performed similarly when a suitable notch width was chosen for the FIR filter. The results can be explained by the fact that the line-frequency contamination in real data is not of constant amplitude, phase, or even frequency.<>