Yesterday I was tinkering on the breadboard, trying to get the Diezel VH4 pedal a bit more palatable to my tastes. Just for the sake of "what happens if", I replaced the last clipping stage (TL072, whose hitting-the-rails sound I am not too fond of) with the equivalent gain version of a Boss BD-2 discrete opamp stage.
Now, in the Spice sims, it was giving me more or less the same frequency response and FFT regardless of which pnp or jfet models I was using (give or take, few mV of bias up or down, close enough for rocknroll, etc etc).
I tried with the first devices that I had at hand for the diff input pair, which happened to be LND150; 2N3906 for the output. Nice distortion, slightly muffled, nothing that can't be fixed by adjusting the cap on the feedback loop, but still a bit coarse and woolly (subjective terms anyway, I know).
I replaced them with J113, and the sound became more sharp, focused and aggressive, even readjusting the passives around for the same frequency response as the previous version.
Now, what leaves me wondering is:

- Where does the distortion happen in a differential opamp? My first deduction, only based on the sims, would be that, given its rail-to-rail bashing, is that the pnp output stage is responsible for the clipping. But changing the input pair gave me a SUBSTANTIAL change in texture.

- At which point, if any, does the input clip in such configuration? Is slamming almost 7V ptp the reason behind that change in "colouration" of the distortion? In consequence, are there two (or more) "clipping points" occurring?

- Why (again, based on the sims) doesn't the type of Jfet or PNP used seem to matter much to the overall gain?

Hopefully more knowledgeable brains than mine will help me understand all this, thanks in advance!

On a side note, the discrete opamp with LND150 input stages would sound quite good in a preamp or dist aimed at stoner/doom and such genres, so Please Steal My Idea If It Hasn't Been Done Before 

I'ld start briefly by stating that a classic push pull tube amp is a kidn of discrete opamp as well.

Funny that you should mention that. This afternoon I was thinking of modifying the output of the discrete opamp in order to get some crossover distortion
The Peavey Dirty Dog or The later version of the SD Pickup Booster look like good places to start... But first, I guess understanding what makes the circuit tick should take priority over blind grafting and modding.

Best guess answers:

Quote from: Umlaut on December 20, 2024, 09:58:42 AM- Where does the distortion happen in a differential opamp? My first deduction, only based on the sims, would be that, given its rail-to-rail bashing, is that the pnp output stage is responsible for the clipping. But changing the input pair gave me a SUBSTANTIAL change in texture.

I would guess the output PNP too. It might be useful to do some more sim work by abusing the feedback network to let the thing work with 100% feedback at DC for bias point, but shunt all AC feedback above a few Hz to ground, so it's effectively open loop above that, and then use tiny, tiny attenuated input signals to see whether the inputs or the PNP distort first.
This comes with the question of what "distorts first" means. The LND150 has a Vgsoff of -1 to -3V. It has a transconductance of typically 2 mmho, or 2 milliamperes per volt. The gates will sit at about (guessing in the middle) 2V lower than the sources, so with the gates at Vd (??4.5V??) then the 4.7K resistor has 6.5V across it and then 6.5/4700= 1.38ma through it. Presuming this splits equally at balance, each JFET has 690uA through it.
690uA through the 2.2k driver resistor is 1.52V, so there is some of the 690uA left over for driving the PNP. With the PNP base 0.6V down from the positive rail, there's only 0.6V/2200 = 272uA in the resistor, so the base of the PNP is getting 690-272 = 418uA of base drive. Ish. So yeah, there is base drive for the PNP, and only a gain of a bit less than 5 needed from the PNP to pull the 2.2K output resistor up to balance at 4.5V.
In fact, there is a LOT of base drive available for the PNP. My best guess is that "balance" on the input devices is with the inverting input JFET sucking most of the bias current away from the inverting input. Your simulator ought to be able to tell you how imbalanced this is.

Quote- At which point, if any, does the input clip in such configuration? Is slamming almost 7V ptp the reason behind that change in "colouration" of the distortion? In consequence, are there two (or more) "clipping points" occurring

I suspect that the differences in Vgsoff and forward transconductance between the two types of JFETs is making for differently imbalanced inputs. 2N3906 has a typical gain of 80 or so at lower currents, so it only needs a base current change of about 32uA to drive it from full on to full off with a 2.2K load. Changing input JFET transconductances will change how fast that 32uA will be obtained.

With the LND150, the transconductance of ... about ... 2ma/V means that changing the drain currents only needs .000032/.0002 = 16mV difference between the gates to drive the full swing on the output PNP. Changing the devices to J113 gives about a 10:1 increase in transconductance, so yeah, this could radically affect the distortion as about 1/10 as much voltage change on the input will drive the output full on to full off.

Quote- Why (again, based on the sims) doesn't the type of Jfet or PNP used seem to matter much to the overall gain?

I think because overall feedback is hiding the real changes inside the "op amp". This is what negative feedback was designed to do. Only running the amp effectively open loop can the internal funny stuff be teased out.

Quote from: Umlaut on December 20, 2024, 04:39:52 PMFunny that you should mention that. This afternoon I was thinking of modifying the output of the discrete opamp in order to get some crossover distortion
The Peavey Dirty Dog or The later version of the SD Pickup Booster look like good places to start... But first, I guess understanding what makes the circuit tick should take priority over blind grafting and modding.

The key is in dynamic cross over distortion. At onset of clipping a gradual gap of cross over. This introduces not only a sizzle on the classic distortion, it "bites" away signal profile which means loss in power. Compression!
The dirty dog has a simple way of doing is. Quilter has one too. The T Dynamics circuit by Peavey does it but complicated. I love that circuit. I added a master volume internal in my redstripe bandit so I can crank the clean channel constantly. Great tube like amp which does all the treble boost and rounding off fuzz face-in stuff!

R.G., thanks A LOT for the detailed explanation and the maths. Some of it is above my level, so I'll have to let that sink in and reread as needed.
I did what you suggested on the sim: it seems indeed that the output pnp distorts before the diff stage, although not by much. Now, my reasoning might be erroneous,  but probing the drain showed asymmetric distortion way below rails after veeery few mv, I'd assume this is effectively the input distortion? Probing the input just gives me the source signal.
When comparing LND150 and J113 on the sim, the former has indeed a lower transconductance and takes longer to clip. I suspect in my use-case the difference doesnt matter that much except in the "texture" nuances.

Steben, my go-to amp (and actual favourite) is a silver stripe bandit, which I sadly can't crank very often, but I am quite fond of the transtube concept and tones. Would you mind sharing where in the circuit you implemented the MV? That might be the ticket to fully enjoying it at apartment volumes

Quote from: Steben on December 21, 2024, 04:51:34 AM

Quote from: Umlaut on December 20, 2024, 04:39:52 PMFunny that you should mention that. This afternoon I was thinking of modifying the output of the discrete opamp in order to get some crossover distortion
The Peavey Dirty Dog or The later version of the SD Pickup Booster look like good places to start... But first, I guess understanding what makes the circuit tick should take priority over blind grafting and modding.

The key is in dynamic cross over distortion. At onset of clipping a gradual gap of cross over. This introduces not only a sizzle on the classic distortion, it "bites" away signal profile which means loss in power. Compression!
The dirty dog has a simple way of doing is. Quilter has one too. The T Dynamics circuit by Peavey does it but complicated. I love that circuit. I added a master volume internal in my redstripe bandit so I can crank the clean channel constantly. Great tube like amp which does all the treble boost and rounding off fuzz face-in stuff!

I don't understand. Crossover distortion is almost by definition away from clipping. Class A has no crossover distortion because the tubes are always on, class AB places the transition somewhere at midswing and class B tends to have the most objectionable distortion at low signal levels (...ish, with transistor devices class B actually makes more sense than class AB, oftentimes).

I almost got sucked in by that one too. I think he was joking.

Hm, unless I am misunderstanding something here, our off-topic is related to mimicking the behaviour of a bog-standard marshall power stage, where you get both clipping and crossover distortion. See f.ex. the transtube power stage circuit, the Ampeg SvT-5pro (or 8pro?) or dis 'ere contraption that I posted on another forum a while ago (see post #3): https://www.ssguitar.com/index.php?topic=5371.0
There's another patent that I found recently which uses a less complicated way of achieving the same end resuls. Can't remember the name now, Poletti or similar... Ah, I will have to rummage through the archives.

47k pot. Not exactly master... but reduction

Quote from: fryingpan on December 21, 2024, 12:38:49 PM

Quote from: Steben on December 21, 2024, 04:51:34 AM

Quote from: Umlaut on December 20, 2024, 04:39:52 PMFunny that you should mention that. This afternoon I was thinking of modifying the output of the discrete opamp in order to get some crossover distortion
The Peavey Dirty Dog or The later version of the SD Pickup Booster look like good places to start... But first, I guess understanding what makes the circuit tick should take priority over blind grafting and modding.

The key is in dynamic cross over distortion. At onset of clipping a gradual gap of cross over. This introduces not only a sizzle on the classic distortion, it "bites" away signal profile which means loss in power. Compression!
The dirty dog has a simple way of doing is. Quilter has one too. The T Dynamics circuit by Peavey does it but complicated. I love that circuit. I added a master volume internal in my redstripe bandit so I can crank the clean channel constantly. Great tube like amp which does all the treble boost and rounding off fuzz face-in stuff!

I don't understand. Crossover distortion is almost by definition away from clipping. Class A has no crossover distortion because the tubes are always on, class AB places the transition somewhere at midswing and class B tends to have the most objectionable distortion at low signal levels (...ish, with transistor devices class B actually makes more sense than class AB, oftentimes).

Hmmm..... cathode bias shift? Power sag dropping the bias? It happens along with clipping.
Quilter and transtube made a whole concept out of it.
And it works... Not so meaningless.

Quote from: Steben on December 21, 2024, 01:50:09 PM

Quote from: fryingpan on December 21, 2024, 12:38:49 PM

Quote from: Steben on December 21, 2024, 04:51:34 AM

Quote from: Umlaut on December 20, 2024, 04:39:52 PMFunny that you should mention that. This afternoon I was thinking of modifying the output of the discrete opamp in order to get some crossover distortion
The Peavey Dirty Dog or The later version of the SD Pickup Booster look like good places to start... But first, I guess understanding what makes the circuit tick should take priority over blind grafting and modding.

The key is in dynamic cross over distortion. At onset of clipping a gradual gap of cross over. This introduces not only a sizzle on the classic distortion, it "bites" away signal profile which means loss in power. Compression!
The dirty dog has a simple way of doing is. Quilter has one too. The T Dynamics circuit by Peavey does it but complicated. I love that circuit. I added a master volume internal in my redstripe bandit so I can crank the clean channel constantly. Great tube like amp which does all the treble boost and rounding off fuzz face-in stuff!

I don't understand. Crossover distortion is almost by definition away from clipping. Class A has no crossover distortion because the tubes are always on, class AB places the transition somewhere at midswing and class B tends to have the most objectionable distortion at low signal levels (...ish, with transistor devices class B actually makes more sense than class AB, oftentimes).

Hmmm..... cathode bias shift? Power sag dropping the bias? It happens along with clipping.
Quilter and transtube made a whole concept out of it.
And it works... Not so meaningless.

But that's not crossover distortion, it's bias shift, which can lead to crossover distortion but that's a side effect, not the main effect.

The reason I thought it was a joke is that crossover distortion is fundamentally a push-pull issue, and this opamp is fundamentally single ended. As it sits, it simply can't do any crossover distortion because it does not cross over.

Crossover distortion comes from a discontinuity in well, crossing over, from a positive going to negative going signal output around the average or DC/0 level. There are tricks to introduce this by inserting an anti-parallel pair of diodes in series with the output, for instance, probably others.

This opamp will naturally have asymmetrical distortion unless the open loop gain is huge, as the PNP will run out of pull up in a different way than the resistor will run out of pull down. That will naturally reflect back through the feedback path to shift the bias point inside the amplifier section.

Found it! https://patentimages.storage.googleapis.com/26/09/61/06441121cd3b24/US20080049950A1.pdf
Indeed this particular circuit is single ended, but my "what if" musing was about replacing that pnp with a push-pull stage capable of generating xover. Either replacing, or sticking such stage (the dirty dog bit, or quilter/pignose diode thingameebob, or the boss Mega Distortion double bjt stage, or didnt Roland blues cubes have a similar stage at certain points of the preamp?)... Then there's the question of whether to put that inside or outside the feedback loop, but here I am speculating before fully understanding the circuit As Is.

IIRC antiparallel diodes do cros sover, but past a certain voltage it kinda "evens out" and loses it? That was certainly the drawback on my Keelhauler circuit. I kinda bumped into a similar issue when experimenting with abused LM358's, crossover was accentuated when clipping receded and "faded to clean", instead of doing the opposite as in the Transtube circuit.

Quote from: Umlaut on December 20, 2024, 09:58:42 AMWhy (again, based on the sims) doesn't the type of Jfet or PNP used seem to matter much to the overall gain?

The amount of open-loop gain can affect the sound.

The original JFETs in the Boss pedals produce quite a bit more gain than common JFETs.   That's not the issue here but the point is using the lower transconductance devices does produce an audible effect.

In this thread I matched BJTs and JFETs.   In this case I reduce the transconductance of the BJTs with emitter resistors:
https://www.diystompboxes.com/smfforum/index.php?topic=129424.0

In your example here the LND150's are likely to have lower transconductance than the J113's.    That's the sound you prefer.

So it is possible to reduce the transconductance of the J113 JFET diff-pair stage to be more like the LND150 by adding resistors in each of the source leads.  Similar to what I did with the BJTs.

You could do an experiment to see if it works the J113's.    J113 specs vary a bit across manufacturers so I can only give a rough estimate for the source resistors of 470 ohm.   However, it's possible you might get a better match by trying values in the 220 ohm to 1k range.

FWIW, using frequency response is only one aspect to the sound.   Also trying to judge the sound of the amplifier from the a time domain simulation is unlikely to pickup the fine detail you hear.

QuoteIndeed this particular circuit is single ended, but my "what if" musing was about replacing that pnp with a push-pull stage capable of generating xover.

You really need to get that going on simultaneously to clipping. It's not as trivial as just sticking an opamp somewhere, discrete or not.

QuoteEither replacing, or sticking such stage (the dirty dog bit, or quilter/pignose diode thingameebob, or the boss Mega Distortion double bjt stage, or didnt Roland blues cubes have a similar stage at certain points of the preamp?)

None of those were OpAmps though.

And yes, modern Cube amps used that stage, modern Blues Cubes are a different thing.

You can find an example of working "antiparallel" diode configuration from Ampeg SvT-8PRO IIRC.

The cross over thing you like is not achieved by the push pull design but other mechanisms. It is far easier to simply copy the quilter or dirty dog principles.