I'm using the function to filter out line noise at 60 Hz (and 120, 180 Hz) caused by electrical noise: ecogNotch

In the case below I can effectively filter out the line noise with a notch filter at 60 Hz +- .5 Hz unfiltered, low line-noise

For noisier electrodes, like below, I am unable to filter out the line noise with any range on my notch filter.

unfiltered, high line-noise

I understand that this is due to my notch-filter being gaussian in the frequency domain and my artifact exceeding all filter shapes significantly. Is there another way to effectively notch-filter out the extreme cases of line noise.

My two thoughts were either interpolating across the frequency domain or some sort of statistical baseline. I thought I'd ask here first to see if there is a well accepted answer for how to approach this.

  • $\begingroup$ Hi, it's kind of hard to guess the spectral shape of your second "broader" grid hum peaks. Maybe you could get a power spectrum with more points? My wild guess is that either you're fine with a slightly broader bandstop filter, or you're better off locking a PLL to the tone and cancelling it. But that depends on how the system actually behaves around the net frequencies $\endgroup$ Mar 23, 2017 at 21:57
  • $\begingroup$ Perhaps this can help you as it is a simple notch filter implementation with the ability to easily set the bandwidth with the change to one parameter. You could cascade three of these structures to null your three harmonics (this could be combined to one filter but for stability purposes I would not recommend that; for the same reason we reduce higher order fiters to biquad sections): dsp.stackexchange.com/questions/31028/… $\endgroup$ Apr 3, 2017 at 10:58
  • $\begingroup$ I think this adaptive approach for narrow band interference could also be quite applicable: dsp.stackexchange.com/questions/37902/… $\endgroup$ Apr 3, 2017 at 11:04


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.