PT-80 delay build report: crazy runaway feedback

[geertjacobs]

Finally finished my PT-80. Well finished... There are still a few things things that need to be fixed.

The level control for the repeats seems to be too strong, with the control at 9 o'clock the level is already equal to the input signal.
The control for the number of repeats makes the delays go into oscillation(edit: this is runaway feedback)  from almost 8 o'clock! This means that until now I can only use it for very short echo's.
The delay pot works great, but of course I wired it the wrong way. :roll:

The sound I get with my limited settings are great, but is it normal that I can't set the control for number of repeats over 8 o'clock? Maybe I need to re-route my wiring there. When I wired it up and tested it without putting it in the box, it also oscillated like crazy. I shortened the wires and ran them under the board, but it doesn't seem to help...

Here's a picture of the box and the inside. Still need to find a way to mount the pcb properly. I'll probably glue in some metal standoffs.

http://users.telenet.be/orangefg/GJA_PT80/PT-80open.jpg
http://users.telenet.be/orangefg/GJA_PT80/PT-80.jpg

[Valk]

Looking nice.

Where did you order the parts?

[Peter Snowberg]

Great job! That looks really beautiful!!!

[birt]

nice

[Connoisseur of Distortion]

i like the interior alot... but that box is too sharp looking. square.  :?

again, beautiful interior!

[geertjacobs]

Thanks.
The parts all came from local stores. The IC's were quite expensive, I didn't know about  http://www.musikding.de at the time, otherwise I would have ordered them there. I got this box because at the time there were no aluminium boxes in which the pcb would fit. It's painted already, but of course it's not 100% anymore after drilling.
The chickenheads came from on old tube amp.

[mathflan]

Hi,

Where did you take the schematic of the PT-80, General guitar gadgets or Ustomp???
I built the PT-80 too with moosapotamus mods and Chorus mod.
I had some problems with level repeats, I change 22k resistor which is solder to SA571 with an adjustable resistor, take higher than 22k if you want less volume and lower if you want more volume but if you decrease below 22k there's a saturation.

[geertjacobs]

I used the layout and schematic from GeneralGuitarGadgets.
No mods done. I double-checked all components yesterday and it should be completely according to the original schematic.

[Mark Hammer]

The problem of excessive regeneration is common enough that even the best designs include some way of "taming" it.  One often sees a trimpot in the regeneration path of phasers, flangers, and delays.  I suspect the reason is that when you throw in 5% tolerances on this resistor and that one, it can be the case that the level of the regen signal can, in some cases, be high enough to keep it going.

Here is how it works.  At the factory, the regen pot is set for maximum.  The trimpot is then adjusted until you reach a point where oscillation (runaway feedback) stops.

If you look through the schems of some flangers (remember, chorusses, while they might use a VERY similar circuit, do not usually employ recirculation, so you will typically only see this in flangers), you will often see an op-amp stage two or three semiconductors "into" the circuit with a pair of diodes (or more, in the case of the A/DA flanger) in the feedback loop to form a crude limiter that keeps the addition of new and recirculated input from exceeding a certain level and turning into runaway feedback.  A resistor in series with those diodes "softens" the clipping so that this mixing stage doesn't just turn into a fuzz (although the extra harmonic content probably makes the effect sound more interesting).

In the case of delay/echo boxes, there is often a compander chip to do this, instead of an op-amp clipper circuit.  Still, the compander doesn't really set an absolute limit.  As far as I know, it just reduces input proportionally, making it still possible to insert a regen signal that stays alive longer than desired.  

In the case of the PT-80, probably the simplest fix, and most sensible thing to do is to to insert either a well-selected fixed resistor, OR a trimpot, between the 1uf output cap of the compander and the 50k REPEAT pot.  If, for example, a 4k7 resistor is placed in series with the input lug of the pot, the pot "behaves" as if it is a 54k7 pot that can never be turned up more than about 90% of the way.  In a sense, it is as if you have converted the pot so that 100% of its rotation is used for only 90% of its range - i.e., the pot now affords greater precision.  I suspect some sort of series resistance of 25k or less, in either trimpot or fixed form will allow you to nail maximum regen up to the point of oscillation.

From another perspective, when it comes to delay lines, I am a big fan of adding more lowpass filtering in the regen path.  Normally, everything going through the regen path also passes through all the assorted lowpass filtering before and after the delay chip (digital or analog) all over again.  But this filtering is intended to provide maximum usable bandwidth at all delay times so the cutoff frequency remains the same, but the effectivee steepness of the filter above that cutoff increases with each repeat.  In the "real world", long delays and secondary/additional reflections lose a lot of high frequency energy and are mere bandwidth shadows of their former selves.  As such, additional lowpass filtering in the feedback path tends to increase the "realism" of longer delays with multiple repeats by shaving off more and more highs and mids below the cutoff frequency of the initial lowpass filters.

I mention this here because additional filtering in the feedback loop will effectively reduce the feedback signal amplitude over repetitions such that it can't keep summing with itself to the point of runaway feedback quite as easily.  It CAN happen, but with extra filtering, a regen setting that might normally start to produce howls will not do so.  I might point out that this particular mod will probably eliminate the need to do the one mentioned earlier, since it attacks the same problem, just in a different way.

Probably the simplest way to do it in this case is to take the 22k resistor that feeds pin 12 of the compander chip, and replace it with a 10k and 12k resistor in series to achieve the same series resistance.  Now, from the junction of the two resistors, run a small nonpolarized cap to ground, and voila!  I would suggest a value, but unfortunately the computer I'm working on has the Windows applets like the calculator locked out.  I am guessing, though, that with a 10k resistor, something like a .015uf to .022uf cap might be sufficient to do the deed.

The nice thing about this is that you can easily install a toggle to lift the ground end of the cap and restore full bandwidth if you so choose.

Finally, while it doesn't sound at all more "natural", cutting the low-end of the regen signal will also reduce the accumulated signal level and reduce howling.  Here, what you would do is stick another cap in series with the 1uf feedback cap to create a lower effective value series capacitance.  A value of .047uf to .1uf is probably a good place to start.  This will shave off several octaves from the bottom of wherever the passband starts.  If you want to defeat it, simply install another SPST toggle than shunts the added cap and you'll be back to original bandwidth.

I recommend both these latetr twomods.  I've used them before and you'll find it opens up a lot of sonic possibilities and textures, at the same time as solving a usability problem.

[geertjacobs]

Mark,

It's really appreciated that you took the time to write all this down.
I will try out your suggestions over the weekend.
Thanks!

[T-rask]

Mark,

  Once again, an excellent post that I've stumbled across. I was wondering if you're saying in other words, that the additional low-pass filtering in the regen path helps because in essence, it's making the components after it not have to deal with all the extra work from the frequencies that don't affect the acutal audio signal that we're trying to keep, by cutting them out and "lessening the load" so to speak?

 Which brings me to the other question of, if you were to throw in gain stages in the regen path, when hooking up multiple delay chips in series, could you, in essence, create a more stable signal that could last for longer regen times, by constantly boosting it along the way. I know natural degradation will still occur, but basically, to what extent do you think it could be taken?

Also, I'm assuming that this may be used already in some effects out there. Any schematics I should look at to get the visuals???

~trask


~trask

[NaBo]

Yep, if Mark ever pulled out of this forum with his posts, I'd probably cry.  What a wealth of knowledge he's shared.  Always goes the extra mile not only to get you up and running, but to help you understand WHY it works the way it does.  Thanks Mark, you rock!!!

Gotta bookmark this thread for when I build mine!

[JT]

One of  the reasons i love the PT80 IS the runaway feedback  

[Mark Hammer]

Thanks for your support, gents.   Think of me as just another storage cap in the great bucket brigade chip of life, passing on whatever charge I have stored to the next stage..... :roll:

The treble rolloff does what it does because you simply end up with slightly less signal each iteration so that he regen can't snowball.  The problems arise when the signal gets recirculated and another copy of the signal, at the same amplitude gets added on top of that, and another on top of that, etc.  That's why the diode circuits are used - because they set a fixed maximum on the signal, no matter how many iterations are summed.

The thing about both analog and digital delays is that, while we assume that "copies" are being repeated, each copy is slightly different than the original.  In effect, the delay circuitry changes the signal over time.  This happens in a few ways from what I have understood.

In the case of analog delays and bucket brigade chips, the tiny storage caps in the chip that make up each stage tend to leak.  They don't leak a huge amount (most such chips show less than a few db signal loss under most normal conditions), but they leak enough that after a number of iterations/repeats, between the leakage and the filtering, the signal isn't precisely what it once was.  The slower you clock the chips (i.e., aim for longest delays possible), the greater each cap has to leak, and the less faithful a copy of the signal is passed onto the next stage.  If you are willing to sacrifice a LOT of bandwidth, apparently you can stretch delays out and go for many repeats with a reasonable copy of the original at the end of it, but we are NOT talking enough bandwidth to hear pick attack or the difference between old and new strings in this instance.

What such chips have to their advantage, though, is that they have the equivalent of what is essentially infinite resolution; i.e., a-to-d conversion that has an infinite number of bits, all the time.  In Dave Hunter's recent guitar pedal book, for all its shortcomings, the interviews contain some real nuggets.  One of them is Roger Mayer's comments on analog-vs-digital differences.  One of the things he points out is that most musicians he discusses such matters with tend to give negative opinions about the decay phase in digital delays.  What he points out is that while it is wonderful to have 24-bit resolution, each bit is essentially dedicated to a certain amplitude range of the signal, such that by the time you let a note decay and it has dropped to a fractionof its initial attack amplitude, you have considerably fewer bits left to encode that signal level because the remaining 16 bits are off twiddling their thumbs, mumbling "Hell no, my contract stipulates that I only deal with signals 256mv or higher".  If I've understood him correctly, his contention is that digital is essentially indistinguishable from analog on note attack and when the overall amplitude is high enough, but where analog retains its "infinite resolution" across all possible signal levels, digital declines in quality as the signal level declines.  Whether he is entirely accurate in this view is another matter, but it's what he proposes and there is some logic to it.

The problem this creates, in turn, is that when you have multiple repeats, or when you do transfers of tracks to and from the digital domain, quantization error is exaggerated for lower amplitude nuances.  Keep in mind that quantization error is what happens when analog signals are coded into digital information, "forced" into the closest digital level classification erroneously (by virtue of not having enough resolution in the conversion process) and essentially misrepresented when converted back again from code to some analog form.

While digital delays have this problem, the problem they DON'T have is leakage of signal over time like analog delays do.  So you have six of one and half dozen of the other.  Use analog delays and multiple repeats over long delays start to lose fidelity, although they may have it over digital delays for the first iteration or two because of this "infinite resolution".  In the digital domain, the nuances of note "tails" may lose something in the a-to-d translation, but you can keep rerouting that puppy back through the input and as poor a representation as it might be, at least it won't get any worse, even with long delays and many repeats.

I am telling you this (oh, so you asked!) because wanting to keep analog signals alive for a long time is not an effort bound to be rewarded.  Fact is, the signal gets cruddier with each repeat.  This is, in fact, one more reason why I like extra lowpass filtering in the regen loop, because the usual lowpass filters are enough to keep audio crud out that is due to being clocked through *once*, but tends not to be enough to reduce the additional crud that accumulates after being clocked through several more times.  In essence, you NEED more lowpass filtering the more the signal passes through delay stages, and passing through 4096 stages 3 times is the same as passing through 12288 stages.

In the case of digital delays like the PT-80, remember that the regen signal being fed back has been converted to digital, then reconverted back to analog prior to recirculation whereupon it is once again redigitized upon reentry to the PT2399.  Each of those conversions will result in quantization error, and an accumulation of audio crud, especially since the PT2399 is not exactly a 24-bit a-d clocked at 96khz.  Less resolution and slower clocking WILL result in more quantization error, and recirculation will only make it worse.  This error WILL require heavy lowpass filtering to blur the error (the error shows itself most at the higher frequency end of the spectrum).

The long and the short of it is that with a delay of this quality, keeping the signal alive longer is not exactly in your best interests.  Moreover, taking out more high end with each repeat is likely to be good practice whether working in the analog OR digital domain.

Dear lord, I hope that's clear.

[wui223]

Sorry could somebody explain to me which one is Feedback loop?
I am just too new to effect buidling.
Thanks Mark, u really enlighten me alot. If u could do a complete analysis of this delay effect, that would be even better   . ( i am too greedy i think :twisted:  )

[geertjacobs]

I mention this here because additional filtering in the feedback loop will effectively reduce the feedback signal amplitude over repetitions such that it can't keep summing with itself to the point of runaway feedback quite as easily. It CAN happen, but with extra filtering, a regen setting that might normally start to produce howls will not do so. I might point out that this particular mod will probably eliminate the need to do the one mentioned earlier, since it attacks the same problem, just in a different way.

Probably the simplest way to do it in this case is to take the 22k resistor that feeds pin 12 of the compander chip, and replace it with a 10k and 12k resistor in series to achieve the same series resistance. Now, from the junction of the two resistors, run a small nonpolarized cap to ground, and voila! I would suggest a value, but unfortunately the computer I'm working on has the Windows applets like the calculator locked out. I am guessing, though, that with a 10k resistor, something like a .015uf to .022uf cap might be sufficient to do the deed.

It's all there...  :wink:

[Mark Hammer]

The "feedback loop" is the path from the "repeat" pot back to the input of the compander chip.  Sorry for the confusion.  I realize the term "loop" is often used to refer to some sort of patch-point for external connections and this may have confused some folks who started looking for it, only to find it wasn't there.

Just so people know for the future, flangers, delay lines and phasers often have a "feedback loop" or regeneration path, where a portion of the output is fed back to the input.  Chorus pedals do not have one, and neither do Uni-vibe type applications of phaser circuits, since recirculating the signal does produce any usable improvement or alteration.

[wui223]

Thanks, i tried to add a trimpot in between the repeat pot and the compander chip ( feedback loop ) but it still produce alot of oscillations.
Although i tried to trim it for sometimes, seems that no subtle effect at all
WHY?

[STOMPmole]

Probably the simplest way to do it in this case is to take the 22k resistor that feeds pin 12 of the compander chip, and replace it with a 10k and 12k resistor in series to achieve the same series resistance.  Now, from the junction of the two resistors, run a small nonpolarized cap to ground, and voila!  I would suggest a value, but unfortunately the computer I'm working on has the Windows applets like the calculator locked out.  I am guessing, though, that with a 10k resistor, something like a .015uf to .022uf cap might be sufficient to do the deed.

Mark, I just wanted to say...THANKS THANKS THANKS THANKS THANKS!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
I tried your tip for the 10k and 12k resistors with a .022uF cap and it worked fantastic!  My runaway feedback problem is gone (although I can still get DESIRED runaway feedback when I crank the regen pot) and as an added bonus my delay sounds a little more mellow and a bit nicer.

I know many times people make detailed posts to help somebody (myself included) only to never get any feedback on if their help was useful or appreciated.  All I can say is that I really appreciate it...and that it worked like a charm to help my PT80 earn a permanent place on my pedalboard.

[powerplayj]

I'm still a bit puzzled by the "repeats" pot and how it affects self-oscillation.  When I increase the repeats, I eventually get into the oscillation at the last 10% range of the pot.  I assume the resistance is increasing when I turn the pot () so, I replaced my 50K with a 25k but I get the same oscillation at the end of the pot's rotation.  Am I missing something here?

Other than the unwanted oscillation, this pedal sounds beautiful and has no noise whatsoever.  It reminds me quite a bit of the delay I hear in Eric Johnson tunes like "Manhattan" and such.  Sweet!!!