Pros and Cons of OD circuits

[uneumann]

The way to experiment with crossover distortion is to add a pair of back-to-back diodes in series with the signal path. These can have a bypass pot to vary the depth of the effect (similar to Peavey's diode noise reduction circuit). The Boss HM-2 also uses diodes in the same way. You can experiment with diode types to give different levels of crossover. I had a distortion circuit breadboarded with a variable crossover feature (as well as varying asymmetric clipping). It was interesting and as spin off got rid of hiss and noise.

Good point - there is a passive version of this idea and it has another attribute that's useful.

[uneumann]

There's a "folding gain" -type of action in the circuit: At output amplitudes below diode Vf the gain is - as usual - defined by the feedback ratio, and typically high-ish. Above Vf the forward biased diodes sort of act like switches and the gain drops to about unity (or some other value if we introduce series resistance to diodes-). If you plot transfer function you can see how it "folds" instead of brickwalling, i.e. "clipping". So yes, compression is evident.

The instantaneous gain compression is what provides the "soft clipping". We could achieve a similar effect also with inverting amps and easily too: just introduce ample series resistance to diodes, enough to limit gain to about unity.

But problem of this type of operation is that if we pump the output level too high majority of signal amplification now takes place at the compressed part of the transfer curve, which is linear but just lower gain. The low amplitude parts of the signal on the other hand are then amplified too excessively and the "crossover region" thus becomes distorted.

So we really should limit the input signal to proper range to get this circuit to work like archetypal diode clippers. Wink. Peavey's "Saturation" generation of amps feature a clever voltage divider for the diodes that effectively increases Vf in relation to output signal and thus allows the circuit to operate at wider levels of output signal amplitudes than about 1 to 2 volts.

Yes - agreed - levels matter in all of these OD circuits. Thanks for the Peavey pointer. The Studio Pro 50 looks like it has this control. I'll take a closer look.
I see the Pro 50 also has a simple diode clipper after the tone stack. It does seem like the two behaviors are complementary.

[Helmholtz]

I use mine all the time to push an ‘on the edge’ amp into overdrive.

I guess that's what I'm doing as well as a 35W Fender with low efficiency 10" speakers really has to fight in a bluesrock band.

BTW, I never liked Gibson PUs with strong magnets like fully charged A5 or even stronger ceramic types.
Original PAFs and 50s P-90s used rather weak A4 magnets.

[teemuk]

The Studio Pro 50 looks like it has this control. I'll take a closer look.

Yes it does: there's a 1K5/470R resistive attenuator for the feedback diodes that "fold the gain". Effectively about three times higher output amplitudes are tolerated (than sans divider) before the diodes conduct and decrease the stage gain. Even the "Normal" channel of this gen of Peavey amps tend to have that distortion circuit, permanently on and non-user-adjustable.

[Helmholtz]

A typical TS "clipper" should have the "folded gain" as well though maybe less pronounced.
There's no reason similar circuits should behave completely different.

[teemuk]

The way to experiment with crossover distortion is to add a pair of back-to-back diodes in series with the signal path. These can have a bypass pot to vary the depth of the effect (similar to Peavey's diode noise reduction circuit). The Boss HM-2 also uses diodes in the same .... It was interesting and as spin off got rid of hiss and noise.

Peavey/Boss circuit introduce crossover distortion true (roughly blocking all signal content below Vf), but overall they are indeed more akin to noise gates: In a highly clipping distorted signal the crossover region is actually very narrow because of high amplification factor for even smallest signal content. For example, if you think of a square wave the "crossover region" is actually just a vertical rise and fall of the waveform, iow non-existent. Such crossover distortion is thus rather inaudible in highly clipped signals, plus you have harmonics of clipping distortion masking harmonics of the crossover distortion. And nearly everything in the crossover region is just hiss and noise anyway.

This is quite different than introducing crossover distortion to a rather clean-ish signal where the crossover region is broader and distortion chops off a lot of signal content, AND proportionally more the lower the overall signal amplitude is. Now you get a constant buzz layered to the audio signal, like listening to an off-tuned radio station. What makes matters even worse is that you can reference this to basically clean signal. The harmonics are very audible. I guess it's a nice lo-fi effect but personally I'm not fond of it at all.

A yet different crossover effect is that of overdriven class-AB push-pull (tube) amplifiers (or emulations of such), where ample overdrive shifts the bias and starts to cause crossover distortion. In this mechanism there is no constant buzz disturbing low amplitude signals but harmonics of crossover distortion are introduced dynamically to a clipping distorted signal only once a specific period of overdrive is exceeded. This just adds a bit of "swirl" to overdriven tones and starts to sound obtrusive only when the bias shift is deep enough to start "blocking" low amplitude signal content in disturbing degree.

[Mick Bailey]

My own experiments with this were inconclusive; at certain settings I could hear that some crossover distortion could be a worthwhile and musically useful feature, but I was constantly making adjustments - any time gain or the guitar's volume was adjusted the crossover needed to be fine-tuned to get some kind of sweet-spot. I think if something that emulated power-amp crossover distortion was included in a distortion pedal it would be quite interesting and musical.

[Helmholtz]

As long as frequency dependent or dynamic effects can be ignored, cossover distortion shows best with a triangular inpiut signal as the slopes of the triangle correspond to the transfer characteristic.
Straight slopes mean linear response, faithful reproduction of the signal and no distortion.
Decreasing slope means decreasing gain (causing signal distortion) and vice versa.

So if the output shows a straight section through the zero crossing, there can't be "crossover" distortion in the true sense of the term.
Small signals within the straight section will be reproduced faithfully.

BB diodes in the series path will cause a gap in the response, meaning that small signals within the gap will not be reproduced at all.
Consequently, when a note decays, the signal eventually drops out as well as the noise.

[teemuk]

Straight slopes mean linear response, faithful reproduction of the signal and no distortion.
Decreasing slope means decreasing gain (causing signal distortion) and vice versa.

The transfer function of typical non-inverting amp with diodes in the feedback loop reflects that: The function is initially linear, typically steep, slope (for "small signals"), which then folds to another linear but decreasing "unity gain" slope at higher signal levels. Dynamic impedance of the diode at forward bias rounds up the knee of this folding.

So if the output shows a straight section through the zero crossing, there can't be "crossover" distortion in the true sense of the term.
Small signals within the straight section will be reproduced faithfully.

Well, IMHO, to degree.

As said, we typically want to employ this kind of circuit as subtle "soft clipping" circuit that achieves the goal by providing distortion via instantaneous gain compression of signal PEAKS. (Technically it's never clipping the peaks, just distorting them by similar mechanism of compression as clipping).

...But if we exceed the boundaries of such operation the circuit begins to amplify (at unity gain) major parts of the waveform instead of just those peaks. We have it majorily operating at the upper part of the transfer curve which is equally linear but where the gain is low. Lower amplitudes are naturally amplified linearly, but with much greater gain.

There's a grey, undefined area where signal "peaks" turn to "major parts of the waveform" but we do agree there's a big difference to overall operation because it turns from compressing the signal peaks to expanding the small signals, which - IMHO - effectively distorts the crossover region of the waveform. If we look at the waveforms the "peak compression mode" distorts the waveform by squashing its peaks, the "expanding mode" distorts it by stretching the crossover region. Neither is "faithful reproduction" of the signal.

[Helmholtz]

[QUOTE=teemuk;n952134]
The transfer function of typical non-inverting amp with diodes in the feedback loop reflects that: The function is initially linear, typically steep, slope (for "small signals"), which then folds to another linear but decreasing "unity gain" slope at higher signal levels. Dynamic impedance of the diode at forward bias rounds up the knee of this folding.
/QUOTE]

Yes, but as the transfer function is initially linear - as you confirmed - small output signals within that linear region will not get distorted.
That's essentially the same as saying there's no crossover distortion.
With crossover distortion even and especially small signals would get distorted.
(The ugly thing about real crossover distortion is that its percentage increases with lower signals amplitudes.)

Furthermore the linear crossover region itself will always be reproduced faithfully even with larger signals (though at a much higher gain than the rest of the signal in our case).

The output of the OP's circuit is linear between around +/- 0.4V, so it doesn't have crossover distortion.
Gain drops at higher signal amplitudes causing distortion by instantaneous compression.
The signal "roof" stays triangular because the large signal incremental gain has a finite lower limit of maybe 2.
A clipped top would mean zero local gain.

My intention is/was to clarify the term "crossover distortion". Otherwise I fully agree with your statements.

[pdf64]

...
With crossover distortion even and especially small signals would get distorted.
(The ugly thing about real crossover distortion is that its percentage increases with lower signals amplitudes.)
...

Due to HT sag and grid rectification, an amp may have no crossover distortion at low signal levels, but exhibit some at full power / overdriven.
It's especially the case with cathode bias, where in addition to the above, there will be bias shift.
Bear in mind it may not be zero crossing crossover distortion, but rather non linearity around the A to B transition (gm doubling https://music-electronics-forum.com/...on-in-class-ab )
I think that's why cathode bias doesn't give good results unless the conduction angle is pretty high, close to class A. Colder cathode bias will probably sound ok at lower signal levels, but due to bias shift may become awful at high power output / overdriven.
Aiken refers to cathode bias shifting with signal level as 'squish' https://www.aikenamps.com/index.php/what-is-sag

[Helmholtz]

Due to HT sag and grid rectification, an amp may have no crossover distortion at low signal levels, but exhibit some at full power / overdriven.
It's especially the case with cathode bias, where in addition to the above, there will be bias shift.
Bear in mind it may not be zero crossing crossover distortion, but rather non linearity around the A to B transition (gm doubling https://music-electronics-forum.com/...on-in-class-ab )

Bias shift distortion is a dynamic effect (remember that's something I excluded in my explanation above) that occurs at larger signal.
With cathode biased amps the reason is increased cathode voltage in the class B region and with fixed bias it is caused by grid current charging the coupling caps.

As it doesn't happen at crossover and has nothing to do with takeover between tubes, it doesn' really make sense to call it crossover distortion.

[pdf64]

I suppose it depends whether the crossover being referred to is zero crossing, or the crossover transition from A to B areas of operation. If the conduction angle isn’t much more than 180° then they’ll be the same thing. Whereas with higher conduction angles the A to B transition will move away from the zero crossing region.

[Helmholtz]

I suppose it depends whether the crossover being referred to is zero crossing, or the crossover transition from A to B areas of operation. If the conduction angle isn’t much more than 180° then they’ll be the same thing. Whereas with higher conduction angles the A to B transition will move away from the zero crossing region.

Now that would be a completely new (to me) approach .
In German crossover distortion is called Übernahmeverzerrung, meaning take-over distortion, referring to the take-over between the tubes.

Transition distortion might actually be a better term for the bias shift effect - in order to distinguish it from real crossover distortion.

But I think we're deviating from the actual topic, meaning the OP's circuit.

[uneumann]

After mulling over the responses in this thread, I think that the breakdown of the OD circuits can be improved. The initial post is more of a list than a structured classification of methods or results.

I think signal operating levels need to be taken into consideration since they often determine the results that are achieved. The same circuit behaves one way with small signals and behaves another way with large signals, so the circuit itself doesn’t determine the outcome. The outcome is what I’m interested in.

The music distortion world is full of variations of circuits and I started this in hopes of getting some organization to all of that. If we can classify all (or at least most) circuits and methods into a small group of categories, it helps (me at least) in understanding what each category does to a signal.

OD circuits are waveform butchers. That’s just a fact – they hack away at the input waveform and thereby produce harmonics. I’m not trying to grasp at the harmonics content of the output (not yet at least). I’m just trying to understand what is happening to the waveform. That’s a start point.

With a focus on waveform distortion, it seems obvious (in hindsight) that the main variant of all the circuits is the shape of the transfer function they implement. To produce distortion, there needs to be a nonlinear portion of the transfer function and the key to classifying methods is to focus on where the non-linear portion lies relative to the input signal amplitude.

I’ve drawn a set of transfer functions that I think represent the various options that I’m aware of. These are just examples of a class or category. Variations exist within each class, and the boundaries between categories are admittedly fuzzy – but we live in a fuzzy world so maybe this is as good as it gets. It’s a start at least.

Before we get into an argument over terminology – let me just say, this is my chart so I’ll use the terms I find descriptive or common. If you don’t like my terms – pick your own and make your own chart. Terminology is not the point anyway.

The charts show the input along the horizontal axis. The output is the vertical axis. I’m showing a non-inverting stage so a bipolar signal will fall into quadrants I and III. (An inverting stage would use quadrants II and IV.) The red line is the transfer curve that relates inputs and outputs. Let’s assume that the nominal input signal runs along the entire length of the red curve. This is important, since at different input levels the circuits behave differently. We have to know the nominal level for which the circuit is intended.

Yes, my linear sections are straight lines and the nonlinear transitions are sharp edges. They are easier to draw. I hope you can imagine curves rather than sharp bends. In drawing 5, I took some time to make smooth curves, so you can see what I mean. Also, the straight lines I’ve drawn are often not exactly straight (linear) in practice. Most OD circuits with diodes or tubes will produce smooth curves. But there are cases where sharp bends do occur, as in clipped opamps or SS amplifiers.

The 5 curves are described below
1) shows an ideal linear stage. No distortion is produced and this would not be useful as an OD stage. It’s just shown as a reference.

2) shows classic crossover distortion. I’ll generalize my term to include any distortion near the zero crossing so I’ll call it “Crossing Distortion.” In this case, any input near zero produces no output. Yes, a push-pull stage with incorrect bias is a classic circuit that produces this result. Any number of other circuits (like Mike Bailey’s) may produce the same or similar transfer functions. What matters is the end result - there is a nonlinear mapping of input to output values around zero.

3) shows another form of crossing distortion where the distortion occurs around a zero input (in fact right at zero) where the output suddenly jumps as the input crosses zero. The opamp circuit I posted earlier produces this this type of transfer function. I would argue that circuit falls into this category. I didn’t mean to single out this circuit as anything special in the earlier posts, but I did happen to be playing with it prior to these posts so I included it in my earlier list and I think it belongs in this chart.

Class 3 differs from class 2 in the slope of the non-linear section. One could argue they’re the same class of crossing distortion, but their behaviors strike me as different enough to warrant separate classes.

4) shows a transfer function that clips the peaks of the input signal. (Also called “clamping” in some circles.) No matter what it’s called, signals are cleanly passed and there’s no distortion until a signal peak reaches a threshold where the gain drops so the output no longer increases as rapidly as the input does. Most OD clipping circuits (like a DCCF) admittedly do not clip as abruptly as the curve I’ve drawn, but the idea is the same. The nonlinear distortion occurs at high signal amplitudes and it tends to flatten the waveform top.

5) shows a combination of classes 3 & 4 (Clip & Crossing). It’s not really a separate category, but it’s important to realize that systems often have combinations of the simple categories. I took time to draw nice curves, and the section around zero is not fully vertical, but its slope is high. I drew this because I think the power of having a set of categories is that it helps formulate what’s possible and decompose complex situations like this one. As stated previously, amps are systems, so it’s likely that combinations of OD categories happen all the time. Pedal OD may overdrive an input stage and eventually overdrive a DCCF. This is a plausible preamp transfer function.



Maybe all of this is obvious, but it wasn’t to me before working through this thread.
Thanks for everyone’s patience and input.

This is only a start to a more complete description and set of classes. I only show symmetric functions. I’m omitting EQ before or after the OD stage. I’m not including time- or freq-varying dynamics. Those seem orthogonal issues that can be added on top of this.

I imagine there must be other perspectives and categories and charts.
Digital processing could implement arbitrary transfer functions.
What is the perspective in that domain?
There must be other survey articles or papers out there – anyone have pointers?