Clipping diodes exposed

[merlinb]

Many DIYers likes to play with clipping diodes. But most engineers maintain that there is no difference in the clipping nature of silicon diodes or LEDs; the only difference is their forward votlage. Sure, LEDs will give you more headroom before clipping, but that's sonically the same result as using 1N4148s and turning the gain down a bit.

But do different diodes really clip identically? I decided to find out by measuring the total harmonic distortion plus noise (THD+N) in an archetypal clipping circuit. If the diodes are in any way different, it should show up most clearly at the very onset of clipping, long before it is audible as overdrive.
Here's the test circuit I used:

And here is the measured distortion using an Audio Precision System 1. Bandwidth 10Hz-80kHz. Test frequency 1kHz (I also checked at 250Hz; it was the same).
As you can see, there is little or no difference between LEDs, zeners (operating as forward biased diodes), or 1N4148s. The curves are the same, just shifted by their difference forward voltages. There is a small difference between the 1N4148s and 1N4006s, but I suspect not audible. The biggest difference is likely to be between using diodes, and NOT using diodes (opamp clipping). Unfortunately, I didn't have any germanium diodes to test.


Here is the same information on log/log scale:

[samhay]

Nice.
Made more sense after I multiplierd the x axis by ~ 30 (2.8 x 11) - the graphs may be improved with a second axis showing approx peak voltage across the diodes.
Care to try some of the fancier clipping options like a gate-drain shorted MOSFET ?

[merlinb]

Care to try some of the fancier clipping options like a gate-drain shorted MOSFET ?

OK, here is the result of using some MOSFETs that I had to hand:

[Seljer]

How about what happens if you wire a potentiometer in series with the diodes as a "softness" control?

[gjcamann]

This is great info, what a great way to compare clippers using THD 

I consider the Shaka Braddah to use the most correct MOSFET clipping. You need the diodes to inhibit the internal MOSFET diode from interfering.
http://www.muzique.com/schem/shaka5.gif
I would also like to see the Fulltone Fulldrive/OCD clipping stage tested, because it sounds so unique and squishey.
http://4.bp.blogspot.com/_y4AYtND8Hz8/SGs-_zgfRKI/AAAAAAAAAMA/8IMJgYSyobE/s1600-h/OCD+Scheme.jpg
http://www.madbeanpedals.com/projects/Deadringer/docs/DeadRinger.pdf

[samhay]

Care to try some of the fancier clipping options like a gate-drain shorted MOSFET ?

OK, here is the result of using some MOSFETs that I had to hand:

Wow, that was quick and that's what I had in mind.
Looks like the dual MOSFETs do clip softer but need more than a 9V supply to work properly. No great suprise, but good to see the evidence.

One thought. The impedance of your FB loop is pretty low by Tube Screamer (etc) standards. I don't think this matters unless the diodes can be reverse biased, but might be interesting to see if it behaves any differenlty with a 1M FB resistor.

The 'softness' control is the other obvious think to look at; will this just shift the curves to the right?

[merlinb]

One thought. The impedance of your FB loop is pretty low by Tube Screamer (etc) standards. I don't think this matters unless the diodes can be reverse biased, but might be interesting to see if it behaves any differenlty with a 1M FB resistor.

That shouldn't make any difference, it will just mean the clipping begins with smaller input signals (which will make it more difficult to measure distortion, since the signal-noise ratio will get worse).

The 'softness' control is the other obvious think to look at; will this just shift the curves to the right?

It won't shift them, it will reduce their slope, but otherwise it won't create any new differences between them.

[R.G.]

Pretty much as expected. I'm mildly surprised the opamp sans clipping diodes was not more abrupt, but then opamps do odd things.

Actually, now that I think about it, I guess that's not all that odd. It does rise very abruptly, but then rounds off as the input signal rises, pushing it past the early clipping where the forward gain is still enough to enforce steadily rising distortion. Once it's clipping, it gets harder to make it clip more and the amplifier gets less sensitive. Very different case from clipping diodes, where what is going on is exploring the V-I curve.

Something I've always intended to do now that I'm using a software oscilloscope pretty much entirely is run a similar test, but with the spectrum analysis function to capture the curves not just of THD, but of second, third, and further harmonics. Curves of THD capture the abruptness of the transition into distortion, but on top of that the harmonics curves would capture the change in clipping tone as well, at least as much as any visual curves can capture a sound.

How about what happens if you wire a potentiometer in series with the diodes as a "softness" control?

I suspect that it stretches the curves out laterally to the right, as what that does is to add a constant fraction of the undistorted signal to the clipped signal. This doesn't work for the "no-diodes" case, obviously, but one could cobble that up with a mixer.   Hmm. Now that I think about THAT, that's what the most recent products from my day job do. I had been looking at it a different way.

[samhay]

^That shouldn't make any difference, it will just mean the clipping begins with smaller input signals (which will make it more difficult to measure distortion, since the signal-noise ratio will get worse).

Quite right, but I was thinking along the lines of scaling both the 10k and 1k resistors to maintain the same gain.


^It won't shift them, it will reduce their slope, but otherwise it won't create any new differences between them.

Yeah, that makes sense - I probably should have pondered that a little longer.

[samhay]

^run a similar test, but with the spectrum analysis function to capture the curves not just of THD, but of second, third, and further harmonics.
That's how I look at clipping circuits on the breadboard. There is more information content, but much of it boils down to how asymmetric the clipping is.

I don't have any calibation of the input/output absolute voltages, so it is nice to see hard numbers here describing the clipping threshold, etc.

[Mark Hammer]

1)  Many thanks, Merlin.  Nice work, and helpful contribution.  Gotta love that "empirical evidence stuff".  

2) We tend to collectively forget just what's involved in a guitar signal.  First, it is highly variable in both amplitude  and harmonic content across notes, and amplitude/harmonics within a plucked note. Moreover, it is the VERY rare occasion when we feed the next device in line with a single pure tone.  That doesn't make measurements like those shown here moot.  Rather, I simply have no idea how anyone could make any sort of reliable attribution about what different sorts of diodes "sound like" in the face of all that variability.

[italianguy63]

I guess that supports the evidence of LEDs having distictively different clipping tone.... considering they are significantly skewed on the chart?

[tca]

Can you plot dTHD/dVin? The derivative of THD as a function of Vin (rms), just to look at the width of bell curved shapes?

P.S.
Don't know if you have enough data to do it (finite difference would do).
Or give the data in CSV format.

[merlinb]

Can you plot dTHD/dVin? The derivative of THD as a function of Vin (rms), just to look at the width of bell curved shapes?

Like so?

[tca]

That's it, thanks. Just wanted to understand what happens near the maximum of variation of THD divided by, say, the length of some interval around that value. It seems that the rate of growth of distortion before that maximum is diode independent, but I see some (small) differences after that maximum. It seems also that the rate of THD distortion stays more constant with the opamp after that point (check slopes).

Thinkering more about it.

[seedlings]

[rubs eyes] Am I reading right, or does all this mean that each clipping method tested (possible mosfet exception) is relatively similar, and [rubs eyes again] the straight opamp clipping may have a very slightly edgier onset?

Can't wait for the normalized sound demos!

Helpful contributions, as always, MB!

CHAD

[tca]

... or does all this mean that each clipping method tested (possible mosfet exception) is relatively similar...

That would mean that you could rescale THD and Vin simultaneously (+ translation for Vi), with different constants, such as all curves coalesce into one... hmmm, not sure about it.

P.S.
Look at the curves for the 1N4006 and the 1N4148, no translation needed on Vi in this case, I could rescale the 1N4006 to fit the 1N4148 by multiplying THD of 1N4006 by some constant (~ 1.1).

... I simply have no idea how anyone could make any sort of reliable attribution about what different sorts of diodes "sound like" in the face of all that variability.

... intermodulation distortion - the elephant on the dance floor!

[Seljer]

... or does all this mean that each clipping method tested (possible mosfet exception) is relatively similar...

That would mean that you could rescale THD and Vin simultaneously (+ translation for Vi), with different constants, such as all curves coalesce into one... hmmm, not sure about it.

I think that brings more merit to the notion of: if it ain't sounding good, mess with the EQ, not the diodes

[teemuk]

Another possibility is to curve trace the forward voltage region.

The usual generalisation is that you get plots similar to this...

...But in my experience the characteristic curves are more "model dependent" than "material dependent". The forward voltage region is usually around 0.3V for Germanium and Schottky devices and around .6V for Silicon but the sloping can vary greatly. i.e. Not all Ge are "softer" while not all Si are "harder".

Here is a "real life" curve:

Note that both 1N34A and AA112 are Germanium devices so yes, plenty of variation in "knee" exists even when the devices are built out of same material.


But, IMO, this is pretty much researched and known stuff. The real interesting question, IMO, is how audible these little differences are? When everything else in circuit behaviour is scaled to match can we even distinct a slight difference in characteristic curves and its effects to tone of clipping? There's no denial that differences exit but can we actually perceive tham and in what "amplitude"... I would think that in practice such detail is right at the bottom of the list of all things that affect tone of clipping (e.g. pre and post EQ'ing, clipping thresholds, clipping symmetry and the interactive variation of it, etc.)

[Mark Hammer]

I think that brings more merit to the notion of: if it ain't sounding good, mess with the EQ, not the diodes

Here, here!!