Or in other words, you may need to make the Q lower.
If you want to be fancy and make something dynamic,
 , try to keep the Q high when the tone is steady and lower the Q just  before there are changes to the tone.

On 26/11/15 16:16, radams2000@gmail.com wrote:
> I've designed hum removal algorithms for guitars in the past and they all have audible time-domain effects. You can't get around the basic time/frequency trade offs. What you refer to as oscillations are just the linear filter response to changes in the input. At 50hz any high-order filter will have an impulse response measured in seconds. So you need to relax the frequency-domain specifications if you want a shorter impulse response. The absolute minimum length filter with a notch at 50 hz is 1 + a delay corresponding to the half wavelength of 50 hz. But this is not a very selective filter , and also will have notches at odd harmonics. Alternatively you can use 1 - delay of full wavelength (20 ms) which has notches at all harmonics but also has a notch at DC which may be a problem for you. Note that if the source of your noise is a power line , chances are you will have harmonics as well, especially if you have a dimmer nearby, so using a filter that also attenuates harmonics may b
e a good thing.

It is worth actually looking at a 50Hz mains signal. Quite apart
from it not being 50Hz, it doesn't even /look/ like a sine wave
anymore: the peaks are noticably flattened by all the low-power
electronic PSUs attached to the mains.

The last time I did an FFT of it,
150Hz: -28dBc
250Hz: -32dBc
350Hz: -42dBc
450Hz: -46dBc

I've designed hum removal algorithms for guitars in the past and they all have audible time-domain effects. You can't get around the basic time/frequency trade offs. What you refer to as oscillations are just the linear filter response to changes in the input. At 50hz any high-order filter will have an impulse response measured in seconds. So you need to relax the frequency-domain specifications if you want a shorter impulse response. The absolute minimum length filter with a notch at 50 hz is 1 + a delay corresponding to the half wavelength of 50 hz. But this is not a very selective filter , and also will have notches at odd harmonics. Alternatively you can use 1 - delay of full wavelength (20 ms) which has notches at all harmonics but also has a notch at DC which may be a problem for you. Note that if the source of your noise is a power line , chances are you will have harmonics as well, especially if you have a dimmer nearby, so using a filter that also attenuates harmonics may be a good thing. 

Bob

Hi all,

I'm trying to filter out 50 Hz noise which intermitelly appears on the input signal, I've been trying to use a 6th order IIR filter but results are unsatisfied:

http://www.jrdltd.co.uk/data/noise.png

Is there any way to filter out such intermittent noise correctly without massive oscillation at transitions, please? Can you please point me to the resource where I could read about this? :-)

Here is a CSV for reference:
http://www.jrdltd.co.uk/data/noise.csv

Kind regards,
Krzysztof