Phaser Feedback - OTA desing

[Epameinondask]

Don't forget that there's positive and negative feedback both possible. Some designs include a switch for this as an option. If the phase-shift stages are based on an inverting configuration, you can swap between them by changing the stage at which the feedback is inserted. If not, you can invert (or not) the actual feedback signal itself.

I also found different results from feeding back the signal from the output mixer (which includes the input) rather than the signal from the last stage (which doesn't). There are a lot of options, and that's *before* you start putting any frequency-dependent tone-shaping components like caps into the audio path.

HTH,
Tom

Thanks a lot Tom. 

Well, my design is an inverting configuration. I tried to feed the last phase stage signal back to many different places of the circuit. I used a cap and a trimmer for limiting purposes.  I didn't have much luck finding a sweet spot where the feedback sounds the way I would like it to sound (that is the ehx style feedback, deep and rich).

Am I doing something wrong?  . I suppose that my current values and configuration of the phaser circuit aren't getting in the way of achieving this type of ehx type of sound. Or do they? The phaser alone without any type of feedback sounds very nice, deep and vintage. I just want the feedback to make the phaser go from subtle to berserk.  Any suggestions would be very welcome and appreciated.

[DrAlx]

I wrote a document on how to analyse this sort of thing a while ago.
http://www.diystompboxes.com/smfforum/index.php?topic=106273.msg960649#msg960649

Short answer to your question, is that regardless of whether the feedback is inverted or non-inverted, you will get strongest filtering effect by feeding back to the very first all-pass stage.

For a 6-stage phaser where there is no feedback at all (just the clean added to the output of the 6 all-pass stages) you get this sort of frequency response:

https://www.wolframalpha.com/input/?i=plot+abs(1%2BP)+where+P%3D((pi*ix-6)%2F(pi*ix%2B6))%5E6+from+x%3D0+to10

If you feedback 30% of the output of the last all-pass stage back to the input of the first then you get a stronger response but the dips no longer go to zero.  (I use 30% in this example so the pictures don't get truncated.  If you want to see the effect of more feedback change the "0.3" value (e.g. 0.5 for 50% feedback):

https://www.wolframalpha.com/input/?i=plot+abs(1%2BP%2F(1-0.3*P))+where+P%3D((pi*ix-6)%2F(pi*ix%2B6))%5E6+from+x%3D0+to+10

If you feedback 30% of the final output (i.e. the sum of the clean signal and the output of the last all-pass stage) back to the input of the first all-pass stage you get a stronger effect in the sense that the dips still go to zero.

https://www.wolframalpha.com/input/?i=plot+abs((1%2BP)%2F(1-0.3*P))+where+P%3D((pi*ix-6)%2F(pi*ix%2B6))%5E6+from+x%3D0+to+10

Whether or not you prefer the sound or if it's audibly different is another story.

[Epameinondask]

Thanks a lot DrAlx for your info. 

[ElectricDruid]

I wrote a document on how to analyse this sort of thing a while ago.

Excellent stuff. Thanks, I'd missed that.

Whether or not you prefer the sound or if it's audibly different is another story.

It's audibly different. I tried it. Whether or not you prefer it...like you say.

Tom

[StephenGiles]

What happens if you put the feedback signal only through say 4 or 6 further fixed stages?

[DrAlx]

What happens if you put the feedback signal only through say 4 or 6 further fixed stages?

That's tricky to write down in a way that gives a single picture. With all the pictures above, phasing just involves taking a picture and stretching/compressing it in the horizontal direction, so the peaks change their positions. You cant do this stretching for a fixed stage.  I can easily model the effect of putting extra variable all-pass stages into the feedback path though which might provide a clue...

E.g. regular feedback from last output stage to first...
https://www.wolframalpha.com/input/?i=plot+abs(1%2B(P%5E6)%2F(1-0.3*(P%5E6)))+where+P%3D((pi*ix-6)%2F(pi*ix%2B6))+from+x%3D0+to+10

To model putting an extra 4 variable stages into the feedback arm (in series with the feedback factor of 0.3), just incease the power of P in the denominator from 6 to 10.
https://www.wolframalpha.com/input/?i=plot+abs(1%2B(P%5E6)%2F(1-0.3*(P%5E10)))+where+P%3D((pi*ix-6)%2F(pi*ix%2B6))+from+x%3D0+to+10

Result looks similar to before but as if EQ applied. Now increase feedback factor to 70% and the effect of the extra stages is increased. Sort of have 5 dips rather than 3.
https://www.wolframalpha.com/input/?i=plot+abs(1%2B(P%5E6)%2F(1-0.7*(P%5E10)))+where+P%3D((pi*ix-6)%2F(pi*ix%2B6))+from+x%3D0+to+10

So with extra fixed allpass stages in the feedback arm, you would get extra notches at some point of the sweep.