Phaser Feedback

[daverdave]

Hey all, I've been designing a phaser for my first year uni project. I've designed a pretty decent lfo and I've rigged up some pretty standard allpass filter stages, I'm using ldrs to sweep the stages. I've got 8 stages at the moment, it's working fine with and I'm getting a pretty good phase signal at the output.

Today I was messing around with feedback, I had an mp3 player going into the input and the output to a powered speaker. I had the feedback coming from after the last stage to the input of the first stage through a 15k resistor, then to the input of the second stage through the 15k. I couldn't hear any difference at all, I tried lowering the resistor to a 10k and then a 1k and I still couldn't hear any difference. I also tried taking the feedback from right after the allpass filters, and from after the summing amp.

I did get some weird noise just shorting the feedback with no resistor, some slight ocillation. Can anyone point me somewhere with some info on phaser feedback, or tell me why it's not affecting the signal?

[Mark Hammer]

If you are using op-amp stages, note that they are inverting in the classic phase-shifter design.  This requires you to use an odd number of stages for the feedback path if you wish to accentuate the notches and peaks..  The tradition is to return the signal to the second phase-shift stage.  The two exceptions I can think of are those where the signal is returned from the output to the original input stage (and not to a phase-shift stage) or things like the orange Ross phaser, where the unit is set up like a 4-stager, but the signal comes back through a 5th fixed phase-shift stage to the feedback node.

[daverdave]

I was looking at the ross phaser schematic today, yeah I noticed that. I tried applying the feeback to the input of the second stage and couldn't hear any difference. I thought that it might have been because I was using an ipob with music, so the bandwidth of the uadio would be greater compared to a guitar. Not sure what to try really. I did mess around with a few series caps as well, still though no results.

I'm a bit confused as well as to what positive and negative feedback is for the stages. I thought that positive would be adding the feedback to an even number of stages and negative to an odd number, take it this is wrong?

[R.G.]

Positive and negative feedback get to being slippery terms where the thing you're feeding around does large phase changes inside it. The original idea was that there were inverting things and noninverting things, and that was that. That worked fine until the change of phase with high frequency limitations inside devices made the feedback that was negative (and stabilizing) at DC become positive (and therefore destabilizing) at high frequencies.

The classic allpass stage generates a phase shift of 180 degrees to star with below the phase shift effects of the phasing capacitor. From there the phase lag increases to another 180 degrees minus the thinnest of margins. The second allpass stage generates another 180 minus a cat's whisker. Therefore two stages will always generate some frequency in the middle where the sum of the two phase shifts is 180 degrees, and when you add that to the original signal, you get a notch.

The frequencies where the total phase shift gets to 360 (or near as matters) the original signal and the phase delayed signal reinforce, and produce a bump of twice the original signal size, a broad, not very peaky hum of +6db.

Notice that you can get the same notches and humps out of the phase line itself by feeding its output back in. What is tricky is that all allpass phase lines have close to unity gain. If the actual gain in the line is strictly less than 1.000000, then the phase line will find a frequency where the phase shift is N times 360 degrees, reinforce itself, and go into classical gain-phase oscillation. If it's 0.999... then it will not be able to truly oscillate, but will be really, really peaky at Nx360 points. As a result, all phasers with feedback make sure they don't let the gain be greater than unity, usually by reducing the amount of feedback with some kind of divider.

In the standard allpass phase line, this is done by making the feedback come through a series resistor to one of the inverting input nodes, and by making the feedback resistor be larger than the phase line gain setting resistors.

From your description, I think there is something that is not letting your feedback be effective, some quirk about how you're attaching the feedback. Just a guess.

[daverdave]

Ah, got it working today, sent the feedback from the output of the last stage back to the second stage inverting input through a 10k resistor resistor and a pot wired as a voltage divider. I put a 330pf cap in parallel with the feedback resistor to get rid of some noise that was there. Sounds pretty damn good now, although with the pot at max it starts to oscillate slightly.

Gonna try to run the feedback through an inverting amp and see how that sounds on monday. Cheers for the advice.

[daverdave]

Well, I've built this thing up on stripboard now, I've installed a switch to select either 4 or 8 stages. When it's selected for 4 the feedback seems really prominent, but with 8 it's hardly noticeable. Not sure where I've gone wrong here, but I'm sure somet isn't right. Sounded alot better when it was on breadboard.

[Mark Hammer]

Where have you added the second set of 4 stages?  Were they added after the 4th stage or inserted between stages 2 and 3?  Or something entirely different?

[daverdave]

they're after 4, just straight in series with the first 4. The feedback goes from before the summing amp to the 2nd stage feedback resistor.

[Mark Hammer]

Okay, so you have the feedback recycling through either 3 stages or 7.
I whipped up an 8 stage Ross/Ropez with the little extension boards that Francisco Pena has, and found that the regeneration/feedback with 8 stages (additional four inserted between stage 2 and 3, making the feedback source always the last stage) easily verged on oscillation.

In a perfect world, each phase shift stage is unity gain such that mixing the full last stage output back in equal proportions with stage 1 should not produce oscillation since there is no cumulative gain.  Of course, the world is not perfect, and with 8 stages and 5% tolerances on resistors, the odds are pretty good that at least one of those stages adds just a teeny bit of gain.  With gain being multiplicative, it doesn't take that much gain across the post-feedback-return stages to produce a feedback signal that is greater than what the feedback-return stage sent out originally.  My sense was that it might be wise to include a trimpot to adjust max regen separately for the 4-stage feedback path and the 8-stage feedback path.  All it took was one instance of cranking the regen in 4-stage mode and flipping to 8-stage mode without easing back the regen!

[daverdave]

Sounds like that might be the case, I used standard yellow leds driving ldrs to sweep the phase shift. But when I had it on breadboard the feedback through 8 stages was really prominent. I didn't use the same components to make the stripboard though so that could be the case.

I've been looking over the stripboard and can't see anything wrong, I'll have a better look though. I'm only using a 10k resistor for the feedback at max, so when it's full it's very very near oscillation with 4 stages, I'll try reducing it slightly and see how the 8 stages react to it.

[daverdave]

Ah, got it going. I noticed I was getting a very low signal at the 8th stage, I'd accidentally shorted pin 5 of the opamp to ground. Sorted it and it works fine, the feedback oscillates a bit when at max but I think it sounds ok, might reduce the feedback resistor a bit.