The lie of smoothing caps on OD's

[dschwartz]

I consider myself quite good at designing distortions and ODs..but i have just realized i know nothing.

I was designing a multistage distortion/overdrive with inverting opamps with diodes in the feedback for clipping.
To me, a little fizzy decay is acceptable on high gain distortions, but in overdrives, its just horrid.

So i was tweaking and tweaking trying to get rid of the fizzy decay, tried different types of diodes, series resistance, voicing, etc, nothing worked. I tried increasing the feedback caps more and more and it only got worse.

Then i realized that since these caps work together with the feedback resistor, when the diodes are clipping there's almost no resistance! So any filtering made at the feedback loop gets cancelled.
Hence, the harmonic content changes drastically from filtered to non filtered, and if you are soft clipping, it just fizzes like a deep fried turd.
So..i removed all the feedback caps, and instead heavily filtered the output, and boom!..no fizz..well. just a little bit due to intermodulation, but it was 99% better.

Everyday you learn something new..i love that feeling.

[Mark Hammer]

While I think you are correct in asserting that feedback caps are not the be-all and end-all of smoothing out overdrives, there IS something to be said for limiting high-end in the clipping path.  Yes, harmonic content will be generated above the rolloff point set by the cap value, but it will be harmonic content based on the bandwidth of the signal being clipped.  So if one rolls off treble at 2khz in the feedback loop, you will be getting plenty of harmonic content above 2khz, but you won't be getting nearly as many multiples of 4khz or 5khz content.  This will be true whether any clipping diodes are in the feedback path OR going to ground on an op-amp's output.

The final tonal goal is a product of both what one has clipped, and what one does to that clipped output.

[dschwartz]

Yes i agree. Of course the harmonics generated will be proportional to the filtered (non clipping) signal. My point is that the cap loses its effect and if you go to far with the cap value, you'll get the unfiltered harmonics (clipping) riding above the unclipped, filtered signal, and it will sound like unrelated buzz. And as you say, same happens with shunted to ground diodes.

My design is basically a drivemaker ( a commercial pedal i make) which has two CMOS stages plus previous opamp filters and gain conditioning stages. I translated the circuit into all opamps, using inverting amplifiers, but using leds for clipping. They sounded soooo different, and realized that a) opamps have much more gain than cmos inverters (duh!) Due to open loop gain.  And B) smoothing the clipping was not working like with the 4049.

After meditating..i got the idea that the smoothing was different on the inverters because it is clipping by saturation to the rails, therefore, the rc filter on the feedback loop was always working, clipping or not ( same applies to jfets, mosfets and transistors)

So..if you have an opamp OD with buzzing decay, try reducing or removing the feedback cap..it may solve the issue.

[bool]

... and when you remove them all you will have to design-in a noise-gate stage ...

[dschwartz]

... and when you remove them all you will have to design-in a noise-gate stage ...

Not necessarily. I managed to keep it very quiet by filtering outside the feedback loops. An active 2 pole low pass at the input and output gets it really quiet. At full gain ( 5150 style gain) it gets noisy, but not more than a tube amp.

[Mark Hammer]

And that's a big part of why feedback-cap variations don't, and can't, do enough.  By their very nature, they are single-pole filters, so any treble/fizz-reduction is modest at best, unless the circuit uses multiple op-amp stages, each contributing another 6db/oct.  And if one seeks to seriously curtail unwanted fizz, you have to set the high-end rolloff low enough that you end up losing bite.  As much as I advocate for treble-taming in the feedback loop, use of a passive 2-pole lowpass at the output is a smart and effective solution.

[dschwartz]

Yes, and it takes a lot of experimentation to get the right balance of bite and noise.
As a hint. On the first gain stage, i used an active low pass filter that varies the cut frequency at the same time that controls the gain. Higher gain, narrower bandwidth, and viceversa..

[PRR]

> these caps work together with the feedback resistor, when the diodes are clipping there's almost no resistance!

Good point.

Relates to the folly of putting an e-Cap across a Zener. Does hardly any good.

[amptramp]

I think you could get good results from a state variable filter which gives you lowpass, bandpass and highpass outputs and only apply the overdrive to the bandpass then recombine all three.  You don't lose bite because the highs are there, you don't get anything muddy because the lows are filtered out and the midband is the only part of the spectrum that is changed.  Then before the signals are recombined, you can filter the overdrive output to soften the response as necessary.

[dschwartz]

I think you could get good results from a state variable filter which gives you lowpass, bandpass and highpass outputs and only apply the overdrive to the bandpass then recombine all three.  You don't lose bite because the highs are there, you don't get anything muddy because the lows are filtered out and the midband is the only part of the spectrum that is changed.  Then before the signals are recombined, you can filter the overdrive output to soften the response as necessary.

It is a great idea, but managing the balance while changing the gain will need a blend potentiometer.
With a little ingenuity you can make a resonant LPF filter as a gain stage that varies the Q. And keep the Q controlled to avoid sounding like a fixed wah (like other crappy metal pedals)

[Elektrojänis]

One More thing to remember with all the stuff in the feedback loop: Non inverting opamp configuration can't go below gain of 1.

That means even without the diodes your high end will get the gain of 1 anyway. If you add the clipping diodes it will limit the signal level anyway so even with "high gain" the signal will not be that loud and that top end that recives that gain of 1 can be significant.

Another thing that minimum gain of one causes is the clean signal that seems to be always mixed in. The clipping diodes actually cause the instantaneous reduction of gain when the signal get's over the threshold of the diodes conduction... But... It can't get below one... So you get your clipped signal + clean with gain of 1. Btw. That seems to sound a bit fizzy to me. If I remember some previous experiments correctly, inverting opamp configuration sounds less fizzy and does not have that clean mixed in. It was quite some time ago I tried though, and it will have some other issues (noise if not anything else).

[dschwartz]

Yeah, as RG put it somewhere, always invert, when possible.

[anotherjim]

Well, you know the feedback cap has Xc impedance at some frequency, so if you assume the diodes are on enough to be considered zero ohm, then if you insert some resistance in series with the diodes, then you can work out the low pass effect at the "clipping" point and the cap will be doing something. Could even make the filter asymmetric with a different resistor for each diode?

[dschwartz]

It would have to be a pretty big resistance to make the cap filter at audible range..
Maybe using a low resistance, like 1K and a big cap in parallel, so the resultant RC is the same when clipping..
Still it's easier to just put a lpf at the opamp output..

[iainpunk]

the reason the fizzyness got worse with bigger caps is because you use inverting gain stages,
the other side of the cap was essentially at a fixed voltage level creating a low ac resistance from output to a fixed voltage. this will lead to higer current draw and thus, in case of most opamps, result in crossover distortion in the signal, especially audible in the decay because there is less other artificial harmonic content to mask it.

taking out this cap really does change the clipping characteristic, but it also affects crossover distortion. maybe try it with another opamp type. another option is a loading resistor on the output so the output stage is always sourcing or sinking current and doesn`t change states (go from source to sink and back). this can give you the desired smoothing of the clipping stage along with no fizzyness.

you can also do this:


it provides smoothing AND filtering in a simple way.

[Rob Strand]

it provides smoothing AND filtering in a simple way.

I guess the more obvious solution is to put a simple low-pass filter at the output like the TS-9, Rat, Timmy.  That affects both clipped and clean conditions.

If you think of the virtual ground point (opamp -ve input) as ground then the 1k and cap form a low-pass filter.  Having the diodes in place ensures the opamp is operating in closed loop under clipping so the virtual ground point is a ground point under all conditions.   

For the non-clipping case, the 1k appears in series with the cap so that reduces the roll-off at high frequencies.   As it turns out the low-pass filtering effect restores the roll-off to be the same as the case where there is no 1k resistor.

For the clipping case however, you get a bit more roll-off due to the low-pass filter effect but only when the effective diode impedance is less than about 1k ohm.   For small feedback caps the filtering effect is going to be small as the low-pass filter behaviour is in the frequency region of f3 = 1/(2 pi 1k * Cf).

A more elaborate scheme to separate filtering for the clean and clipped signal is:

[phasetrans]

I think you could get good results from a state variable filter which gives you lowpass, bandpass and highpass outputs and only apply the overdrive to the bandpass then recombine all three.  You don't lose bite because the highs are there, you don't get anything muddy because the lows are filtered out and the midband is the only part of the spectrum that is changed.  Then before the signals are recombined, you can filter the overdrive output to soften the response as necessary.

When I decided to stop doing pro audio tinkering, and play with FX instead, this SV with clipping on the middle passband was exactly my first idea to try. Still haven't built one, though.

[bool]

The two important states that a clipp-stage works in are basically 2: silence (noise-gain) and saturation. At saturation, you obviously need tone-shaping post-clipp-stage; and at "silence" (noise-gain) you obviously need to do whatever necessary to limit the closed-loop clipp-stage bandwidth to "keep the hiss out" (aka small cap across the nfb).

Imnho these two conditions are blatantly obvious when you look at something as simple as a TS (when it was designed the notion of "high gain" was remarkably lower than today).

Another angle (over-simplified): Clipping the signal usually creates a third-harmonic of the incoming signal, so if you band-limit the NFB at say 5kHz, you get (over) 15k banwidth out of the clipper (even at the transition point of breakup).

So post-filter it is.

[marcelomd]

Dividing the signal in 3 bands, then clipping the mid isn't what the Klon & friends are all about?