CMOS and MOS distortion

[DIY Bass]

OK, so I am interested in making a Darkglass clone, so I have looking into CMOS distortion.  I read the run down on different distortion types on R.G's we site.  That got me to thinking.  I am sure I am not the first one to have had similar thoughts and so I brought it here to see what others have done.  The suggestion is that CMOS distortion is quite tubey, as it compresses rather than just clipping.  The explanation from R.G suggests that the major difference is that CMOS distortion is symmetrical, whereas tube distortion is asymmetrical.  The suggestion seems to be that CMOS inverters are made from matched sets of MOSFETs.  That got me wondering if you could get an asymmetrical distortion if you had a set of slightly mismatched MOSFETs, arranged in a similar way to the internal schematic of the CMOS chip.  Has anyone tried that?

[teemuk]

The suggestion is that CMOS distortion is quite tubey, as it compresses rather than just clipping.

Instantenous gain compression is not any different from soft clipping. If soft clipping is a "tubey" characteristic to you then fine. On the other hand, IMO, many of these solid-state circuits portray much "softer" clipping characteristics than generic common cathode gain stages or vacuum power amps, both SE and PP.

The explanation from R.G suggests that the major difference is that CMOS distortion is symmetrical, whereas tube distortion is asymmetrical.

Let's evaluate a CMOS inverter for starters: It's a push-pull complementary MOSFET circuit. You will have hard time finding MOSFETs with identical characteristics to cut off or saturate, and you will have equally hard time finding complementary pair of "N" and "P" MOSFETS that would have identical characteristics. Come to think of it, that applies to all bipolar solid-state devices.

Yet symmetry vs. asymmetry is still largely a circuit - not device - characteristic! You can tune SS and tube circuits to clip symmetrically or asymmetrically. ...That is, within their inherent limited capabilities of "symmetry".

Not to mention, many inverters include internal "clamp" diodes for input protection, which will introduce asymmetry to clipping characteristics.

Solid-state circuits without feedback are also subject to much larger bias shifts than tube circuits so larger asymmetry than from a tube should be expected. Then again, many generic solid-state circuits do implement feedback (and lots of it) so this characteristic is obscured. That along with inherent soft clipping of solid-state devices.

Symmetry of MOSFET gain stages (source follower, common source) depends largely on circuit's bias conditions. If bias ensures equal output voltage swing for negative and positive halfwaves the net effect is symmetric clipping. If you tune the circuit for something else than such "center bias" then you gain asymmetric clipping characteristics. Same thing as with generic vacuum tubes. Gain stages of tube guitar preamps, for example, are often deliberately tuned for asymmetric clipping. They could be tuned for more symmetric characteristics as well.


CMOS inverter configured as a gain stage biases its input by reference to its output so i't's practically an automatic "center bias" configuration. Clipping is therefore quite symmetric. If you insert some offset voltage to the circuit you can turn clipping characteristics more asymmetric.

Differential and push-pull -type circuits are also inherently more symmetric than "single-ended" circuits because their architecture cancels even order (symmetric) distortion by taking differential of two opposite phase signals that are asymmetrically distorted. Differential of such signals is a signal with odd order (symmetric) distortion. Distortion is also "softened" by the process because differential is that of clipped and non-clipped half wave. So, it must have less harmonics than the clipped signal and more harmonics than the clean one. An intermediate.

Therefore push-pull circuit (that including tube -based circuits like push-pull power amps) will likely produce largely symmetric distortion (often moderately "soft") while a single-ended circuit will more likely produce more asymmetric distortion (often moderately "hard").

[anotherjim]

There is a big problem with a custom "complementary" pair using discrete MOS transistors. The available parts are not like the ones in the CMOS logic families. The big difference is the on-resistance of the drain-source channel is far lower than in a logic chip, except maybe those families that are only usable at low voltages, 5v and below.

If you make a complementary pair and bias it just like an inverter, the shoot-thru current will be too much for practical use. If you run them at lower supply voltage to reduce the current, The gate thresholds may not work at half supply volts bias.

4049UB is not really complimentary. The lower N transistor has lower minimum on-resistance than the upper P type. It was originally designed to meet a big customers requirement to sink load current rather than source it. That was before there was a 4000 logic series as we know it, hence the odd power pin position. 4069UB and 4007UB inverters are complimentary.

The inverters "tube" sound is a difficult topic. It does not have similar structure of any part of tube amp as far as I know, although it's kind of like a tube class A-B power stage since each half is really in class-A bias and works differentially, the transfer curve is totally different. Compared to tubes, the inverter becomes non-linear very early. IMHO, this gives a more congested, less open sound character.

[PRR]

Tube amplifiers are non-complementary. Why are you trying to fake it with a near-complementary pair?

That said.... this field has been plowed extensively. While all active devices work the same way, the parasitic elements are all different and you don't get quite the same action with different devices. Vacuum tubes are terrible conductors; crystal devices conduct better and cut-off harder. You sure can find interesting soundz. But probably not tube-equvalent to a very discerning ear.

[Rixen]

The CD4007 allows independent use of some of the P and N channel MOSFETS, allowing asymmetrical distortion. I've done this, and it's smooth:

https://www.rixenpedals.com/uploads/1/7/9/3/17932295/ninja_tiger.pdf

[DIY Bass]

Thanks for the input people. So - CMOS matched pairs are OK but don't do a perfect job.  You can get asymmetrical distortion by altering the bias on a MOSFET.  I had noticed that the schematic for the CMOS chips looked a lot like an A/B power amp.  It sounds like slightly mis-biasing a single MOSFET would be a simpler and easier than trying for a pair.  That is what I will research then. 

Rixen - are there any sound clips of the ninja tiger?  It sounds interesting.  Oh, and those enclosures!!!! :-O.  So awesome.  I don't need any enclosures at the moment but I will remember them they are spectacular.

[Steben]

Therefore push-pull circuit (that including tube -based circuits like push-pull power amps) will likely produce largely symmetric distortion (often moderately "soft") while a single-ended circuit will more likely produce more asymmetric distortion (often moderately "hard").

"Often" is the correct expression.
Some single ended solid state circuits can behave rather soft, thinking of Fuzz face, while most push pull opamps are straight hard clipping, mostly because of negative feedback.
A typical vintage marshall plexi power amp is a simple "minimum" opamp, it clips rather hard yet very controlled, no opamp latch-up, BJT saturation latency, ... That's why a high gain diode clipping circuit is so easily made "marshall-esque": hard clipping with the right frequency shaping.
The most difficult thing to emulate is a non-feedback soft distorting push-pull amplifier (think of VOX AC30 or 1974 18W), since it behaves so differently according to how much it is pushed.

[jubal81]

Tube amplifiers are non-complementary. Why are you trying to fake it with a near-complementary pair?

That said.... this field has been plowed extensively. While all active devices work the same way, the parasitic elements are all different and you don't get quite the same action with different devices. Vacuum tubes are terrible conductors; crystal devices conduct better and cut-off harder. You sure can find interesting soundz. But probably not tube-equvalent to a very discerning ear.

This is some real wisdom and a hard lesson to learn. What we think of as tube distortion isn't necessarily about the tubes, it's about the whole circuit they're in. Taking a cranked JCM800 as an example, look at the whole circuit and you see a series of gain stages and filtering.

If you want to make a great sounding distortion, you can follow those principles with mosfets, diodes, jfets, etc. The 'magic' comes from getting the gains and input/output frequencies dialed in so each clipping element isn't creating too much distortion  - you need to share the load among multiple stages to get more distortion. Push any clipping element too far and it sounds nasty - even tubes.

If what you really want is a tube preamp in a pedal, there are projects for that. Search for GTFO on this forum. I haven't built one, but demos sound great.

[DDD]

I think it's possible to "adjust complementarity" and gain by changing source resistors' values (8,2kOhm) in the below circuit, based on CD4007 inverters.
http://forum.guitartonelab.ru/cgi-bin/yabb2/YaBB.pl?num=1514838111

[teemuk]

Taking a cranked JCM800 as an example, look at the whole circuit and you see a series of gain stages and filtering.

Key word: filtering. Frequency response shaping -before- and -after- introducing distortion is an extremely crucial design element. In fact, far more crucial element than the stage(s) that actually distort and how those stage(s) distort.

The 'magic' comes from getting the gains and input/output frequencies dialed in so each clipping element isn't creating too much distortion  - you need to share the load among multiple stages to get more distortion.

This actually IS NOT a characteristic you'd commonly find from classic tube amps. In most of them there are maybe just 1 - 3 stages that actually distort and it is extremely typical that just one stage is distorting very "sudden" and "hard". Just look at something like 1959, 2203 or perhaps SLO preamps, for example. The first tube stages are usually just amplifying "cleanly" to pump up signal amplitude and distortion happens in just select few stages.

Multiple distortion stages are usually employed simply to acquire "dynamically shifting" DC bias (and consequently shifting harmonics) due to introducing two different clipping styles: symmetric and asymmetric.

For example, in SLO preamp the main clipping stages are the "cold cathode stage", which clips HARD and with distinct asymmetry, and the stage that follows it (direct-coupled CC+CF). That stage ALSO clips hard, but now with with more symmetric characteristics. Rest of the tube stages contribute almost nothing to clipping.

I can't remember a single tube amp that would introduce "soft" clipping in several cascaded stages. Not one. Carvin's "Quad X" - as a very unorthodox design - cascades more than three distorting stages (in fact, eleven!!!), but even in many of them the clipping is actually quite "hard" and the purpose of the design was solely to acquire the aforementioned effect of dynamically shifting harmonics. In fact, Quad X design deliberately overdid that particular characteristic in comparison to other tube amps out there. No, the design didn't gain any noteworthy popularity. 1 vs. 11 clipping stages still sound about the same.

The problem of soft clipping - especially cascading several soft clipping stages together -  is the intermodulation distortion, which quickly messes the signal and ruins note separation. The reason for "congested" sound of CMOS inverters is that they actually clip much, much softer than any real triode gain stage. They start to introduce harmonic distortion way too early. Cascade a few and the net effect is that IMD quickly rises to astronomical levels and tone becomes nearly useless harmonic-infested "mush".

As I mentioned earlier, "soft clipping" isn't really all THAT typical tube characteristic. That's just self-perpeatuating myth, just like the one that tubes produce "musical" even order distortion and solid-state produce odd order. Oh, and with complex signals + IMD there are no "evens" or "odds".

[anotherjim]

The reason for "congested" sound of CMOS inverters is that they actually clip much, much softer than any real triode gain stage.

Yes indeed. Another quirk from that is to apparently push all the noise and harmonics into the zero crossing area, which I hear as a "bad digital" kind of sound.
For adding character, a single inverter with modest (single figure) gain, gently driven (say, after a lossy tone stack) works pretty good. Which is why I always have to add a shout for the 4007, since we can use the rest of it for a couple of class A stages as discrete mosfets.

[iainpunk]

Id like to add that the load line of a tube is not quite straight, even compared with bjt transistors. This adds to the asymmetry of tubes. Even when run 'clean' tubes will change (dare i say distort) the wave form and introduce a set of harmonics. This adds to the "tube sound" audiophiles are after. This is in my opinion the biggest difference between a symmetrically biased tube and CMOS distortion plus the tube probably clips harder, as before mentioned (this will depend on the exact device used.)
I hope this adds to the discussion.

[iainpunk]

This is a "clean" tube amplified sinewave. Its not symmetrical as you can see, this is what i meant with my previous post.

[rankot]

I have built this pedal using Russian schematic, just added output buffer and drive control. Here is Catalinbread SCOD schematic for comparison, too, cause they're similar.





And even one more, Moscato:

[rankot]

Looking at those two other schematics, it seems that Russain 2p2 bypass caps are too small. I will try with 2n2 soon.

[Fancy Lime]

Do you mean to imply that you tried with 2p2 caps already? Where did you get such small caps? What kind of caps were they? Ceramic?

Also: we really need to retire the whole "CMOS inverter clipping sounds like tubes" thing. I get the appeal of these things to marketing departments but other than that... As if all tube amps sounded alike. Or all CMOS inverter overdrives. You can *make* CMOS inverters sound quite similar to a certain amp of your choosing (tube or otherwise) if you know what you are doing. But you can do that just as well with a couple of diodes to ground. Or any of the other standard clipping techniques. Just a question of design appropriate to the sonic goal and device used. That being said, I really like the "naked" CMOS inverter sound you get from cascading two or three CD4069 stages set to unity gain and pushed just to the edge of overdrive by a preceding gain stage. Not because it sounds like tubes (it doesn't) but because it sounds good (at least to me).

Andy

[anotherjim]

2p2 caps? Maybe by the Gimmick capacitor method...
http://213.114.131.21/begin/gimmik-0.htm
...however, I fear the cut-off will be super low with 2n2 against a 2M2 resistor. 22p might work well to smooth any harshness.

[rankot]

Believe it or not, I bought 2p2 caps and used them for a build 

However, SCOD use 2n2, 10n and 22n bypass caps; Moscato use 100p, 1n and 2n2. I will try with 2n2 to see what's going on.

[rankot]

That being said, I really like the "naked" CMOS inverter sound you get from cascading two or three CD4069 stages set to unity gain and pushed just to the edge of overdrive by a preceding gain stage. Not because it sounds like tubes (it doesn't) but because it sounds good (at least to me).

Got any sound sample to share?

[dschwartz]

I love CMOS inverters. But Teemuk is tha man...
I agree with his point that is the whole circuit and not the device what makes the tone... filter and filter and filter.. that's the secret..no magic device.
Check Darkglass circuit and you'll see that it has a lot of well thought interstage filtering.. that's the magic..not the CMOS.. maybe it adds some soft clipping but you could use anything there and retweak the filters and you got the same results..
My beloved Drive Maker design uses 2 cmos stages but i chose them just for simplicity..i made an all opamp version and it sounded the same..even liked the opamp version a bit more (with leds in the FB)..
If you want nice assymetric clipping, a cmos stage with 1 led on the FB sounds pretty awesome..i get soldano-like harmonics with that...