resonator, filter and delays??

[Nubeat7]

hello, i`m working on a resonator plugin which is done with a delayfeedback and a filter in the feedbackloop but it seem that when i turn the cutoff of the filter down it also produces a tuning down effect?

is this a normal behaviour?

i`m using biquad filters, do i need to do some delay compensation when changing the cutoff? to keep the resonator tuned?

[Drnkhobo]

Hi Nubeat7,

I'm sorry I can't help you with what you are looking for, rather I wish to learn about what you are doing. . .

If you don't mind, could you explain why you put a filter in the delays feedback loop?

Surely if you have a HP in the feedback loop, for example, then wouldn't it be the same as having the filter after the delay module ? Bar one delayed loop? Am I missing something?

Good luck!

[Nubeat7]

normally there is a LP filter is in the feedback loop to damp the high frequencies and make it sound 'warmer' or not too metallic, i`m also not sure if it produces this tuning effect or if it just sounds like this - it makes a kind of 'miou' effect when i turn the cutoff down..

watching it in the scope it looks like that the filtered signal gets phase shifted when i turn cutoff down, what i did was to look for the latency of the filter which seems to be 1 sample so i reduced the delay with 2 samples 1 for the filter and one for the feedback but i`m not sure if this is enough..

[KG_is_back]

I believe that is normal behavior. Lowpass filter not only changes the amplitude response of the signal, but also introduces delay (phase shift) to the frequencies. In case of Lowpass, frequencies close to and above cutoff frequency are delayed the most. If you put a Lowpass filter into feedback, what happens is, that the actual delay for each frequency is feedback delay + filter delay (phase shift). When you pull down the cutoff down, you gradually introduce more delay to lower frequencies = time stretching = pitch shifting. The effect should be very subtle however.
Only way how to completely remove the effect is to use linear-phase filter. You may try to compensate the shifting by lowering the delay time slightly - proportionally to the phase shift of some prominent frequencies.

[Nubeat7]

thanks KG for the explanation, things are much clearer now, looks like i leave it like it is after i don`t think it makes much difference in sound when using some FIR filters ( after days of finetuning everything to sound good with the biquads ) and i think it would be nearly impossible to compensate the phaseshift of the biquads after they also beeing used with modulation sources

only idea about this - could it be done with an allpass filter?

[KG_is_back]

only idea about this - could it be done with an allpass filter?

What you are actually asking is: "can I turn a filter into linear phase by adding an allpass filter into the series?"
...and the answer is not that simple. Theoretically yes. The practical problem is, that you have to make an allpass filter that has an inverse phase response opposed to the lowpass filter... from what I know about filters (which is not much, so people with better math education, feel free to correct me if I'm wrong), such filter would be infinitely complex (infinite-order filter and with each order the thing gets +1sample delay) which is not doable (we can do pretty much up to order 8 with 64bit floats) nor practical (who'd what infinite latency anyway xD ). That's where approximations chime in which may reduce the filter order to a reasonable level (for example, frequencies above cutoff that are near Nyquist are almost completely removed, so we don't really care about their phase). Someone more skilled would have to help you with that - I once tried to make a filter with custom frequency/phase response and failed miserably cos' I failed to understand the math behind.

[martinvicanek]

I am also into waveguide synthesis these days and have encountered a similar problem. Perhaps you may use a first order lowpass in the feedback loop, then the phase shift is smaller. Attenuation of high frequencies should still be sufficient as the filter is passed once per cycle and the effect gets cumulated fast.

The other idea I had is maybe you don't really need to compensate the phase shift for the entire spectrum (which, as KG points out, is somewhat involved ). It may be sufficient to get the low frequency end right (essentially where the fundamental is).

[Nubeat7]

thanks guys, this gives me some enlightning into the world of filters..

It may be sufficient to get the low frequency end right (essentially where the fundamental is).

i think this is beyond my skills atm.. anyway with all this information i can dig some deeper into different filter topics and much things are explained with this, thx

[martinvicanek]

Here is a simple Karplus-Strong schematic with a first order lowpass in the feedback loop. There is also a switch for phase delay compensation. You can check the improvement by playing octaves. There is also some cursory explanation inside. Hope it helps.

[Nubeat7]

thx martin for the example,

after i did some research in the www i found this site http://chromium.googlecode.com/svn/trun ... phase.html
where you can see the phase response for the the different biquad filters related to cutoff and resonance (and also the magnitude) and it seems that only the cutted part get phase shifted while the passed part isn`t shifted - specially with higher resonance values (except for the bandpass where it is 90°)..

so if this is right i only need to take care of the groupdelay which is 2 samples at second order filters like biquads?

so am i right when i just subtract 3 samples from from delaytime to compensate delay and biquad filter?