The cascaded-and-bypassed soft-clipping diode thing

[Mark Hammer]

Over the weekend, I gave in to temptation and picked up a Fender Champion 110 amp for peanuts at a local music store "garage sale".  Installed a new input jack and for $29 I had a nice amp with respectable output.  The dirty channel was way too bright for my tastes, so a few additional treble-trimming caps later it sounds pretty decent.

The dirty channel (see service manual:  http://bmamps.com/Schematics/fender/Champion_110_Service_Manual.pdf ) uses several sets of diodes to aim for a purported amp-like overdrive.  I'm not going to debate the authenticity of this, but the diode arrangement intrigues me and I'd like to understand it better.  I suspect others would too.

Looking at page 7 of the service manual, we see 4 diodes in the feedback loop of U2B:  a back-to-back pair of silicon diodes, along with a back-to-back pair of red LEDs.  The silicon diodes have a 220k resistor in parallel with them, and there is also a 470k feedback resistor.

So here is what I believe to be going on.  At signal levels less than around +/-1.5V, the gain of that stage is set by the 470k feedback resistor.  When the signal hits around +/-1.5V, the feedback signal bypasses the silicon diodes via the 220k, and the LEDs conduct.  Since the 220k is still providing some gain, the LEDs don't clip hard.  When the signal reaches around +/-2V, the silicon diodes AND the LEDs conduct, providing harder clipping.  In this way, the circuit provides clipping quality that changes somewhat, depending on the signal level.

The late Fred Nachbauer used a similar sort of strategy in his Dogzilla amplifier design, to provide a limiter function.  I take it that the multiple stages of cascaded diode pairs was partly because of the signal level differences between a tube amp and my cheezy little Fender; but it was also because the intent was to achieve a less harsh levelling of the signal.


So, I think it is worth understanding more about is how to deploy such cascaded diode pairs to achieve different qualities of clipping and dynamic control.  In particular, the strategic use of the bypass resistors in each diode stage, and the strategic use of diode type.   In the Champion amp, 1N4448 silicon diodes and red LEDs are used, but one could also use a series of schottky pairs with suitable bypass resistors to achieve a more gradual transition.

Discuss.

[Groovenut]

Thanks Mark. I think Fender used that overdrive circuit in most of their amps in that era. Keep in mind the opamp stage we are discussing is an inverting stage, so those diodes are essentially shunted to virtual ground (though almost the same analysis applies). There also appears to be an anti-parallel pair of 1n4448s shunting to ground prior to that stage.I would guess those are just transient protection though as the signal there is very very small. I built a circuit very similar to this into my Bass Driver pedal (http://www.diystompboxes.com/smfforum/index.php?topic=112744.0) it's a nice dynamic overdrive at low settings but can get very metal when you dial up the gain. I think I have my take on the essence of what going on technically in that thread.

I haven't opened any modern Fender solid state amps to see if they are still using this circuit or if they've moved on.

Thanks for opening the topic!

[Mark Hammer]

The first pair going to ground is for clipping.  A single pair is used there because less gain is in effect at that point.  Think of it like a less effective Distortion+.  What's left over after those diodes gets amplified again in the subsequent stage, which is how they can use LEDs for clipping (higher forward voltage).  So, while not a Big Muff by any stretch, it applies a version of double-clipping.

[ashcat_lt]

It looks kind of like that diode ladder thing.  Not exactly the same, but maybe close?  This thread gets into some talk about diode ladders and how and why they do what they do.

[teemuk]

There's so much to learn from a good Google search that uses terms such as "diode waveshaping" or perhaps "analog waveshaping". I will soon try to explain how just two additional resistors can shape "diode characteristics" of the circuit viewed as whole but overall most of this is decades old prior art. Reference material should be a plenty. BUT: Forget electric guitars, amps, effects and even music scene for a moment. There's a saying that when it comes to gear guitarists are conservative and dragging a few decades back compared to other musicians. It's funny because it's so true:
How long do you think things like analog signal generators have existed? Quite a while. It's all about waveshaping: Output signal from oscillator having certain type of waveshape is converted by analog circuitry to different waveshapes. And they often do it at very low percentage of distortion, since they are precision instruments. If guitarists are still at awe that one can actually employ more than one or two diodes to distort the signal, then perhaps they don't know so much about waveshaping that they should, considering that waveshaping has been traditional to guitar tones since days of early rock music. And perhaps you should seek that information from other sources, since in practice there should be decades worth of research and material concerning waveshaping as overall application. With reputation of guitarists being what they are, if I were you, I wouldn't restrict myself to refering only to guitar gear -related material about the subject.

Anyway...

Series resistance to diode will convert the almost horizontally flat "forward bias section" of diode's characteristic curve to generally more upwards sloping -type. If you compare characteristic curves of generic silicon diodes to specific germanium diodes, or semiconductor diodes as is to vacuum tube diodes, the trend is: The more internal resistance (consider it same as series resistance) the diode has, the more it has upwards sloping -type section of that particular area of operation. The effect naturally also increases the forward voltage of the diode circuit, thus higher amplitude input is needed to acquire similar magnitude of overdrive and clipping as without series resistance. However, transition to clipping is also smoother and more gradual due to added sloping characteristic of conductive area of operation.

Parallel resistance to diodes will "soften" the knee portion of characteristic curve where state changes between reverse and forward bias. Consider it as softening the clipping effect even further.

I guess an able mathematician could explain all this in terms of how additional fixed variables shape hyperbolic functions, such as "laws" that estimate characteristic curve of different types of semiconductor junctions. Or something...

In terms of that Fender circuit, the entire diode operation is defined by those LEDs due to their highest forward voltage to conduct. The series diodes with parallel resistor present series voltage dependent resistance to those LEDs. My hunch is that by substituting fixed resistance with impedance characteristics of a diode the linearly upwards sloping part of the characteristic curve (with plain series resistance) now turns into exponential -type of sloping towards "flatter" conduction characteristics that a diode with low-ish internal resistance has. When the silicon diode "series resistors" are not yet significantly conducting the LED has series resistance of that resistor parallel to diodes.

So, it is a complex and interactive process where everything affects everything, and usually not very linearly. But in nutshell it makes clipping "softer" by increasing the forward voltage of the diode circuit and by softening its overall "knee" characteristics. Not to mention, Fender's scheme employes diodes in the negative feedback loop of an amplifier instead of as shunt devices in the output of an amplifier, and as such they are usually subjected to lower magnitudes of current and simultaneously also "self-limit" their drive voltage. The result is yet softer clipping than that acquired from generic "shunt diode clipping" circuits.

[Groovenut]

The first pair going to ground is for clipping.  A single pair is used there because less gain is in effect at that point.  Think of it like a less effective Distortion+.  What's left over after those diodes gets amplified again in the subsequent stage, which is how they can use LEDs for clipping (higher forward voltage).  So, while not a Big Muff by any stretch, it applies a version of double-clipping.

I have to respectfully disagree here. The signal voltage at the junction of R21/Q2/C1-2 should never be more than ~30-40mV even at full gain. That point is a virtual ground because of the inverting nature of the amplifier. My vote is for transient suppresion during bypass.


The clipping characteristics are only softened by the series resistance, R23, until C3/C4 start to conduct, then it's hard clipping shunt style with the Vf threshold being about 2.4V. Remember those diodes see the inverting opamp input as ground so this is no different than shunt diode clipping.

[Mark Hammer]

But U2A has a max gain of 31x.  That's no fuzzbox or infinite sustain, but it's not nothing.  CR1/CR2 should clip at least a little.

[PRR]

{As Groovenut just posted} CR1 CR2 "never" conduct when Q2 is passing signal into the virtual-ground of U2B input. They are there to improve OFF rejection (large signals at an "off" JFET may break-through).

Replace the LEDs with three plain diodes-- no big difference.

Then we have a bilateral 4-diode clipper with 220K across 1/4 of the string and 470K across the whole thing. It will hard-limit at say 4*0.6V= 2.4V but cut gain some at 3*0.6V= 1.8V.

It is pure odd-order: everything symmetric. Wave tops will be rounded above 3/4 of maximum level. There is significant pre-EQ in front of U2A, too much for me to suss out today. No post-EQ (or same-as the Normal path), perhaps because the clipping is fairly soft.

[teemuk]

 

So, while not a Big Muff by any stretch, it applies a version of double-clipping.

Are you referring to the Fender circuit?

Because it's not a "double-clipping" -scheme at all. Everything clips in one, single gain stage - due to overdrive of that diode circuit - unless you of course manage to overdrive the driving opamp gain stage into clipping.

See, the diodes in the input terminal of the clipping stage's opamp are not conductive during normal operation: They cannot be. Under normal operation they are connected to "virtual ground" that is inverting amplifier's input and the audio signal magnitudes at such are practically equivalent to 0V. Especially when the amplifier is an opamp.

The diodes become operational when the channel switching FET turns non-conductive and channel is muted. Then they protect the opamp input and the FET from static voltage spikes and signal transients. But when the channel is in use they portray no function. There is not enough signal amplitude to turn them conductive. Funny thing is, I've even seen a commercial design that was straight out clone of this usual Fender SS guitar preamplifier architecture (no it wasn't a new Randall amp but an article of "tube emulator" in electronics magazine), and it included the input diodes although there was no channel-switching feature. Sometimes "designers" indeed do "paint-by-numbers" having no clue of what they are actually doing.

---

Somewhat off topic, but this Fender architecture is an interesting one overall: The initial filtering stages mimick EQ response of a typical Fender preamp (the component values vary some design-to-design but overall response effect is more or less same with that slight bit of variation that makes each amp unique in its own right), then there's a gain control followed by somekind of waveshaper to mimic output stage distortion taking place in single, final gain stage (early Fender pre's didn't really contribute much overdrive), and finally tone control + filter stages that have somewhat similar response to speaker cab emulation circuits. EQ naturally post-distortion where its most effective. You could call it analog modeling of a Fender amplifier. Crude one, but nevertheless.

Earliest ones of Fender amplifiers with this basic architecture, if I recall, used different clipping scheme: Diodes arranged in a bridge, directing currents of each signal half wave to LED section of an LDR, and the other half of the LDR was in the feedback loop to gain compress signals during overdrive. Nice idea. Clipping from LED that drives a compressor. I think there was a potentiometer control to dial in "limiting" too. There were few other different waveshaping arrangements that Fender experimented with using this preamp architecture as well but I can't remember exact details. I do remember that ROC series used a vacuum tube dual triode that was effectively configured to operate as anti-parallel shunt diodes. All these different circuits serve function to shape forward voltage to full clipping and "knee characteristics" of it, but overall the effects are minor in the whole. Without all that pre -and post processing EQ'ing practically all clipping arrangements would still end up sounding about the same: bland, farty buzzing.

How early SansAmp effect worked wasn't too different. It didn't even have any fancy signal clipping arrangement. It acquired distortion simply by overdriving opamps to rails. You can ponder how much of SansAmp's applauded "tubeyness" was in fact due to well-fashioned pre and post processing filtering, instead of exact way and detail the clipping and overdrive happened. Usually all we need is to hear something distort at preferred magnitude. We don't so much care about fine details of what makes up that distortion we hear. Heck, even the "chopping" -type distortion from intermittency of signal sounds about the same as more traditional clipping distortion. In the end, while waveform analysis might show differences that seem significant and important, the structure what creates those distortion harmonics isn't that different when you think of it: Abrupt signal magnitude changes and removal of information. In intermittent "buzzing" this happens at random but frequent intervals, in peak or crossover clipping at specific but frequent intervals. To our ears it still sounds about the same.

[Transmogrifox]

Spice simulation confirms teemuk's suggestion that it does soften the knee somewhat.  My hunch is the effect is subtle  -- something interesting to experiment and see if I can tell the difference.

The DogZilla configuration is much less subtle and you can see that by inspection of the relative resistor values.  It's a piece-wise ladder of limiting thresholds smoothed by the respective diode VI characteristics.

Either way the Fender circuit does measurably soften the knee.  For higher level signals there is about 0.03 dB difference in the spectrum, so if the thing is cranked it's not likely to make much audible difference.

The comparison of the spectrums got interesting at lower level signals (currents around 30 uA).  It appeared this additional resistor causes the spectrum to roll off just slightly faster than without it.  The 5th harmonic is about 1 dB lower than without the resistor and by the 11th harmonic the difference is 3 dB.

Given those numbers I hypothesize it will have a subtle effect on the amount of "fizzle" you hear as a note decays.  It will change the dynamic behavior audibly but not enough to convince most guitar players you aren't dealing snake oil and almost certainly nobody would notice if you clipped the resistor in between songs during a live performance.

Now I want to breadboard this circuit snippet and see if the effect is more audible than my simplistic analysis implies.  It seems the ears are the only instrument that can be used to quantify this.

[Mark Hammer]

Well, you'll note that the whole thing is preceded by a scoop for the overdrive channel.  I found the overdrive objectionably shrill, and threw a 150pf cap in parallel with the 470k feedback resistor, so that when I switched from clean to overdrive, the EQ settings wouldn't put a hole through my forehead.  There would be grind in abundance, but the lower-order harmonics would dominate.

I can't speak with any authority about other amps, but I imagine this is a recurring issue with many solid-state amps that share a single set of EQ controls with a clean and dirty channel.  In the absence of any compensation, switching from a bright clean sound to a dirty overdriven one is likely to get you too much treble....regardless of how you do the clipping.

[analogguru]

The first pair going to ground is for clipping......   

... but only as long as the overdrive is turned off.  When it is turned on, the diodes won´t see enough signal to conduct (clip) because the diodes (CR1, CR2) are connected to the virtual ground of U2B.

You may ask: Why are the diodes then there ?
The reason is simple: to reduce the maximum signal (across the FET) in bypass mode and by that bleedthrough.

[Mark Hammer]

Okay, now THAT makes sense.  Thanks for the explanation.

[merlinb]

You may ask: Why are the diodes then there ?
The reason is simple: to reduce the maximum signal (across the FET) in bypass mode and by that bleedthrough.

Correct. They are a cheap (and rather clever) alternative to the more common configuration using a shunt FET that operates in anti-phase to the series FET.

[Groovenut]

Well, you'll note that the whole thing is preceded by a scoop for the overdrive channel. ...

Actually, the overdrive channel is proceeded by a mid boost circuit connected to a low-ish input impedance amplifier.

Using my posted pics for comp # reference.

R3/C3 & R4/C5 form a 2 pole low pass filter (differing knees), C4 & R4-R5 form a high pass shelving filter, there is also some, though very little, low pass filtering that happens between C4 and R6.

input versus output of the filter


frequency shaping at the different stages


If there was too much treble getting through, you could also increase the R3 resistor value as well as the cap in the neg feedback loop you added.

Food for thought 

[Mark Hammer]

I looked at the schematic and a Superfuzz midscoop architecture jumped out at me (C4/C5/R4/R5).  Didn't bother to do the math.  I initially put a 15nf cap to ground in parallel with the diode pair just ahead of the switching JFET (CR1/CR2), then added a second cap in parallel with R10.  That made life bearable.

You will note that the Mid control on this amp is essentially the same as the "Contour" control found on some Marshall SS amps and is very similar to what I believed to be a midscoop formed by C4/5 and R4/5.

I'm still intrigued by the cascaded diode-pair thing and whether it can be optimized for different contexts by using diodes with other Vf, more pairs of them, and more strategic bypass-resistor values.

[stm]

Mark, I happen to own a Fender Champion 110 which I purchased in the early 90's.  I'm glad you are looking at the dual-diode feature, I think it is quite good.  Below I describe my general impressions with this amp referring to the service schematic:

1) In my case the amp clips too easily on the clean channel, producing ugly clicking sounds if you hit the chords hard (even with single coils).  This is because the combined gain of U1A + U1B is high enough so U1B hits the supply rails clipping hard, even when the NORMAL VOLUME is set to a very low value.  I replaced R9 (180k) for 22k, which attenuated the signal hitting U1B by about 5 dB and the situation improved.  If I had the chance to travel back in time I would use an even smaller value like 15k for better headroom; there is still plenty of gain/volume available for the CLEAN channel.

2) I found this amp lacking in brightness in the CLEAN channel, so I added a capacitor in parallel with R8 (22k).  This gave a modest but noticeable 6 dB boost to high frequencies only for the clean channel.  Don't remember the value I used, but I guess it might have been around 2.2n.  This mod requires the resistor change described in 1) in order to work.

Mods 1) and 2) make it more usable IMHO.

3) I do not like how the MID control behaves in the CLEAN channel.  Turning it down doesn't give the 400-500 Hz cut I prefer for clean sounds.  However, this MID control, which as you say acts more like a CONTOUR control seems to be well paired to the DRIVE channel at high gains.  I believe the intention was to have a sort of heavy metal type of sound which was popular in the early nineties, and I guess if you look at the MID control from that perspective it does what it should.  My solution is to use a buffered passive BASS/TREBLE control with the clean channel to get the clean sound I want.

4) The DRIVE channel is indeed a bit thin in the bass department if you intend to play blues or some lower gain overdrive type of music, but it is OK for higher gains IMHO.

In the next post I will address the DRIVE channel in more detail.

[stm]

About the DRIVE channel, there is one setting which I believe makes the dual-diode soft clipping circuit shine.  Try setting the GAIN pot to a very low value, 2 or 3 maximum.  The key here is that you don't get hard clipping from the previous stage (U2A)--you will recognize the ugly hard clipping on the peaks.  Use a humbucker pickup and set the three tone controls to 5, then adjust to taste (do not cut too much the MIDs).  This gives a crunchy overdrive without the fizz added by the hard opamp clipping.

In a nutshell you have four different regions in this circuit:

1) Very low signals go out linearly without any clipping; this occurs whenever the output of U2B is below the LED clipping threshold, let's say 1.5V.

2) As the signal increases and the LEDs start conducting, the slope or gain of U2B is reduced to 1/3rd (because 470k // 220k = 150K which is approximately 1/3rd of 470k). This happens essentially when the output of U2B moves between 1.5 and 1.5+0.5=2.0V (roughly speaking)

3) As the signal level at the output of U2B approaches 2.0V, the 1N4448 start conduction and you get harder clipping, but not as hard as the opamp itself would clip against the power supply rails.

4) The ultimate clipping occurs when the preceding stage (U1A) clips hard to the supply rails. This hard clipping is transferred downstream to the output of U1B.  In my opinion when this hard clipping occurs the result is nasty, unless you really want the high gain sound with all its harmonics.

Some food for thought: notice that as the LEDs conduct the gain is reduced to 1/3rd. In turn, the clipping threshold of the 1N4448 diodes is about 1/3rd of the LED clipping threshold. In other words, once you enter into LED gain reduction, the next gain reduction step is smaller in terms of voltage because you are approaching it with a reduced slope.  At least it makes sense to me.

One idea I had (but no chance to test it) was to extend this configuration to three diode steps, each diode in the sequence with a lower threshold.  Imagine LED, then silicon diode, then schottky or germanium diode.  When LEDs conduct gain is cut to 1/3rd, then when silicons conduct gain is cut to 1/10th of the original gain, and finally when the schottkys or germaniums conduct gain is set to minimum.

[Mark Hammer]

I purchased a Yamaha OD-100 for $5 at a music store garage sale some time in the last 8 months. (Cheap because it didn't have any knobs).  It's a pretty complex circuit for a simple overdrive, even including a compander chip.  There isn't much variety in the possible sounds, and the overdrive tone is nothing particularly inspiring.  But fed into the 110 dirty channel it sounds magnificent.  Superbly round, vocal and sustaining.  I was surprised.  It seems to condition the signal just right for the dirty channel.

[Quackzed]

Some food for thought: notice that as the LEDs conduct the gain is reduced to 1/3rd. In turn, the clipping threshold of the 1N4448 diodes is about 1/3rd of the LED clipping threshold. In other words, once you enter into LED gain reduction, the next gain reduction step is smaller in terms of voltage because you are approaching it with a reduced slope.  At least it makes sense to me.

yep. this is the 'nugget' of understanding required to design this type of 'diode ladder' clipping, that the amount of gain reduction of the waveform over the first threshold is the new 'relative' scale of the clipping curve... iow a 4v ptp signal thats 'half clipped' is like a 2v ptp signal above the first diode threshold 'its gain is reduced by 1/2' and thus the NEXT diode threshold needs to be 1/2 as big to stay linear...
...if you have 3 'rungs' each .6 volts apart set up for 1/2 gain reduction (half clipped half blend), the first rung gets hit with anything over .6v but the 2nd diode pair needs a signal not 1.2v but 1.8v to get hit (the first .6 plus .6x2 due to the 1/2 gain redux of the first pair 3rd stage needs that 1.8 plus 1.8x2 or 5.4v!!!...  so each step reduces gain  2x as much as the last, exponentially so you need exponentially less threshold between each consecutive step in the ladder.   

theres only 2 rungs in the fender setup but the idea 'nugget' is in there...