1
$\begingroup$

enter image description hereIm filtering an EMG signal, more precisely the 50 Hz frequency and its harmonics. For that, I've constructed a comb notch filter (on filterdesigner in Matlab), with its fundamental frequency being 50 Hz, sampling frequency of 10kHz, and notches's bandwidth of 0.1 Hz. That means i have a filter order of 199.

Given the filter specifications, its magnitude response is bellow. I don't understand why the first two notches almost aren't attenuated. I've read that high quality factors (Associated with low bandwidth) are related with low attenuation rate, but i don't know in what extense it explains the low attenuation of the first two harmonics. Can you explain?

$\endgroup$
  • $\begingroup$ When remembering correctly the human emg should have a range upto ~500Hz, in other words your signal is oversampled. I would recommend to first decimate the input down to 2kHz. This will also effect the order of the comb notch filter which you are applying afterwards. $\endgroup$ – Irreducible Sep 11 at 13:41
1
$\begingroup$

An order-200 IIR filter? You're playing with fire you know that? High-order IIR filter are prone to instability and/or misleading behavior even if they are stable.

If I understand correctly, you designed a 50-Hz notch filter, then a 100-Hz notch filter and so on?

Do you really need to filter out every harmonic? Do you know how high your harmonics are?

Solution #1 : Why not filter only the fundamental and the other major harmonics. Usually even harmonics are not a problem and typically harmonics higher than 11 or 15 are small enough that you can simply ignore them. This would change your filter from an order 200 to a more manageable order 12 IIR filter.

Solution # 2 : Instead you could use a 200-tap moving-average that would allow you to extract the 50-Hz component and its harmonics. You could subtract the output of the moving-average from your signal and you would get rid of the 50-Hz component and its harmonics as long as they don't change too fast.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.