Feedback Alters Freq Balance

[Perfect Human Interface]

Hi all. So I've noticed that if you create a (non-infinite) feedback loop in Flowstone, simply mixing a signal with itself, it will progressively "warm up" the sound, attenuating the high end and emphasizing the lower freqs. My expectation was that it would simply amplify the audio and not effect it in any other way.

I'm mostly just curious, but does anyone have an explanation as to why this happens?

Feedback.PNG (29.05 KiB) Viewed 18105 times

[tester]

Well - if you exceed the (-1;+1) blue signal ranges, which are crucial for what is happening on main input/output of your plugins/apps, then the signal will be trimmed, producing very hard distorsions. And with direct feedback loop and signals around maxima - you know how it happens (it's not simple 1+1, it's a continuous flow of output - going back to the input).

Try to incorporate this. Feedback level is from 0 to input, and as the feedback arises - input is decreased, to fit the range. If in second module you just set division by 2, then your output will be always 0.5 (no feedback) to 1 (feedback = input). Remeber, that in attached scenario - knob must send values from 0 to 1.

[KG_is_back]

Ti is called one-pole low pass filter. Typically it is used as a dezipper (also in FS as a module) or in envelope followers.

When you sum input and previous output the result is that high frequencies cancel out. input white noise and analyze output with FFT and you will see for yourself.

[Perfect Human Interface]

I think I found my answer here. A signal delayed by 1 sample then mixed with the original signal seems to cause a slight phase cancellation at only the high end. This alone is barely audible, but if performed repeatedly, the attenuation becomes more and more noticeable.

I recreated the effect like so:

Feedback 2.PNG (52.8 KiB) Viewed 18098 times

So it seems a feedback loop causes a 1 sample delay each time it passes through.


Here's another question: there are like three "delay" modules that ship with FS, only one of which specifies in the tooltip that there's always a minimum delay of 1 sample (which I used above). Is this also true of the other delay modules? I also have an optimized interpolated delay by Trog in my toolbox. Is this something inherent of all delay modules?

[Nubeat7]

Ti is called one-pole low pass filter. Typically it is used as a dezipper (also in FS as a module) or in envelope followers.

When you sum input and previous output the result is that high frequencies cancel out. input white noise and analyze output with FFT and you will see for yourself.

you can find a good description on this at the beginning of this site : http://www.katjaas.nl/complexintegrator ... nator.html
which someone posted in the forum or maybe it was SM forum..

[Perfect Human Interface]

I read the beginning of that until it started to sound too much like math gibberish. I liked my "phase cancellation" idea better... Is that not correct? I mean, it just seems so DRAMATICALLY more simple than that whole webpage of mathematical explanation. And I imagine the phase cancellation would effect the high-end more because a tiny shift in timing would be more dramatic a difference in high frequencies since the wavelength is shorter. That all makes sense to my mind.

Clearly I was right about the one sample delay thing though. I just never realized that a feedback loop would have any delay at all, though of course thinking about it now it would be necessary to prevent the program from hanging.

[Nubeat7]

what you can read on the site i've posted explains exactly what you did in the 1st picture you posted and refers to the thread name, the second picture you posted is a different design and has nothing to do with feedback.. (isnt the 2nd example the basic logic of a phaser?)

i`m not a dsp - math man but i found this site explains this things in a very good way also for not math freaks

btw: nice knobs you are using here

[Perfect Human Interface]

the second picture you posted is a different design and has nothing to do with feedback..

I disagree. If I'm not mistaken it's logistically identical to feedback aside from the fact that it terminates after 5 "cycles" (otherwise would require infinite modules).

[Nubeat7]

no, you just add delayed signal, in a feedbackloop you always feed the output back to the input so once the output is feeded back it is not the original signal anymore it is the original signal added whith its own output - now you feed a new changed signal back, while in the second design the delayed signals are always the original signals but just delayed which all you add at the end...

[Perfect Human Interface]

You're right. It would have to be summed after each delay. It did attenuate the high frequencies somewhat when I tried that though. Seemed similar.