A pedal idea that I need more thought about....maybe YOUR thoughts

[Mark Hammer]

I was contacted recently about the modded Gretsch Controfuzz that I had called the Contrafrizz.  One of the mods adds an option for introducing crossover distortion on top of traditional clipping. 

When back-to-back diodes are placed in series with the signal, such that the signal has no way around the diodes, the diodes prevent the signal from passing until it meets the threshold set by the forward voltage.  The net effect is to chop/block those parts of the signal below the V_f.  Since the series diodes don't do anything to the overall peak amplitude of the wave, the audible effect is similar, but not identical, to changing the duty-cycle on a square wave: you hear everything but the signal is "thinner" and reedier. 

The ZVex Machine pedal deliberately uses crossover distortion (one of the few to rely on it completely), and while I found it "interesting" it wasn't especially musical to my ears.  Listening to some synth demonstrations, I started thinking about the diode effect in terms of different duty-cycle versions of the signal, and wondered "What if one could blend multiple versions of the chopped signal to get different textures, and blend them with something more palatable, instead of just crossover distortion?".

So, I whipped up a little dual op-amp circuit with two inverting stages: a gain and a mixer.  The output of the first stage is split three ways and feeds a clean 2-pole passive lowpass filter for blending in (@480hz), and two series diodes paths.  One uses a pair of Schottkys to chop the sides of the waveform below roughly 190-200mv.  The other uses a pair of Si diodes to chop the waveform "sides" at roughly 680mv.  I figure the Schottkys give me most of the original signal (with more "edge"), while the Si diodes crop the signal even more to yield a "skinny" version of the signal. The premise is that I could blend in how much of each I wanted.  The whole shebang gets mixed by the second op-amp (that has a bit of recovery gain and a master output level control)

I made the input stage have a gain of 33x, thinking that I wanted to be able to derive a clean bottom for blending, but still have enough "push" to get the signal up to V_f for both sets of diodes.  Since the clean bass is not attenuated by the lowpass filter quite as much as the diodes cut overall amplitude for the X-over sections, I adjusted the mixing resistors for the paths appropriately.

So how does it sound?  Meh.  I can hear the intended effect, but the proportions are insufficient to be usable.  Here are my followup questions:

• Did I pick the right (or at least useful) "spread" of forward voltages to make a difference in the content of each path?  E.g., t'wer you, would you maybe use a pair of Si and a pair of LEDs?
• Should I apply more than a gain of 33x to push the typical guitar signal through such diodes?
• If I were to chop out everything below +/-600mv how much proportional reduction in signal amplitude should I expect? (the idea being: how much do I need to adjust relative gain for that path in the mixer stage)
• If I got the drive and the balance just right, is this the sort of thing that you think might have some musical value?  Equally important, does having the ability to blend in the clean fundamental give it more musical value?

[amptramp]

Crossover distortion has been around for a long time and one way to get a continuously variable version of it is to use a push-pull transistor stage such as you may have seen in an early transistor radio.  They usually had a forward-biased diode in the base supply to compensate for Vbe temperature variations and you can do that too but you can also vary the bias to get a continuously variable amount of crossover distortion.  Maybe parallel that with a Class A push-pull with no crossover distortion and vary the relative gain of each.  Instead of trying to use diodes that have a fixed voltage drop, you get a continuous variation that you can adjust to taste and even use a pot to make it adjustable.

[ashcat_lt]

I had a quoted reply where I addressed each point, but it wouldn't post, and I don't think it makes much difference, so I'll just throw it all out here in case it helps any.

I am concerned about your assertion that the P2P voltage is not affected, especially since you seem to contradict yourself later.  The P2P level if the output should be two diode drops less than the input (or 0, whichever is greater).

Anywho...

While its not exactly true, I generally figure that a Schottky has about half the Vf of an Si which has about half that of an LED.  So I'd figure the difference between Si and LED is about the same as the difference between a Schottky and an Si.  If that's already not enough difference...

While nobody really likes to admit it, the V/I curve for these different types of diodes are very close to exactly the same, just scaled up or down proportional to what we call Vf.  This means that a 1V input though an Si diode will be pretty hard to tell from a 2V input though an LED except about half as loud.  So, I'm thinking you'll get much more mileage just picking one and then varying the gain into it.

Then you were wondering about makeup gain, but the question you asked is awkward.  We can't say what "proportion" we're throwing away unless we know what the original signal, right?  IDK how to tell you how much makeup gain is required in this situation.  I tried a couple things in digital a while back to try and do the same, but never had satisfactory results.  If you're feeding it known level square waves, then the math is simple, but for arbitrary waves...

...if you take my suggestion to vary the gain in, and vary the makeup gain on the other side by a complimentary amount, I think it comes close to what you're asking for.

[Transmogrifox]

My theory is it tightened the palm-muting effect in this circuit.

[ashcat_lt]

Yeah, that's something I forgot when I had to rewrite my post.  There are a couple of commercial designs out there with "metal" in their names which do this.  I never really thought about the palm muting thing, but it certainly does act as a kind of gate, and probably helps some with noise too.  At extremes you get that gated, stuttering,  no sustain thing which is similar to a poorly biased fuzz.

Another way that you can "decide" how much of the wave gets cut out is to vary the bias on one end of the diode.  This shouldn't really change the curve of the knee, but it will move the actual cutoff point up and down.  To keep it symmetrical, you have to be able to bias each of the diodes in opposite directions simultaneously. 

I had the idea a while back that it would interesting to be able to vary that bias for each diode completely independently, to allow all sorts of asymmetry and even allow "overlap" distortion which is very much like what a Tubescreamer does.  I never got too far on it in analog, but I did implement it as a plugin for Reaper.  It's kind of fun to play with.  I use it as a module in a couple of my other "pedals" for a very small amount of noise reduction, and I've done some more extreme things with it in my noise work.  It's not a sound I tend to look for in a guitar, but I haven't really explored all possible options with it.  If there's a time and place for a sputtering fuzz, I'm sure there's uses for a crossover distorter.  Most sane people probably won't turn it on that often, though. 

[anotherjim]

Could this really be done by a fast gate?
That is -  Full wave precision rectifier, Comparator with threshold control and an analogue switch. When the signal exceeds the threshold going pos or neg, signal is passed until it falls below threshold. Really just a normal gate scheme but NOT averaging the rectified signal for the envelope. The threshold set is the width of the zero crossing "off" region, and if it was provided by an LFO?

[ashcat_lt]

Hmmm...seems maybe we're not giving responses that Mark finds interesting enough to comment on?

This pretty much is a gate with 0 attack and release time, though I think the diode curve is probably at least a little important to the final result.  There may be some mileage to be explored in using shorter than normal time constants in a noise gate, and "embrace the ripple".  I guess that would be almost the opposite of what he's doing here with filtering out the low end first, but it might be worth messing with.