ampage Tube Amps / Music Electronics	For current discussions, please visit Music Electronics Forum.

ampage archive

Vintage threads from the first ten years

Soft clipping, CMOS and the Muff

4/6/1998 4:35 PM
R.G.	Soft clipping, CMOS and the Muff I'm taking a class this week, and the texts didn't show up, so I had to spend some time being roundly bored while the instructor looked for them. While this was going on, I started pondering the clipping in a CMOS inverter versus the clipping in a diode-feedback opamp, and while I was doodling, I realized that if you consider that an ideal feedback amp forces the current flowing through the input resistor to flow through the feedback element. The higher the input impedance, the more this is true. In a CMOS gate used as a feedback amp, the input impedance is astronomical. Therefore you can predict almost exactly what the current in the feedback path is, no matter what it is. The current at any point is Vin/Rin. For a nonlinear element like a diode, you can then just look at the I-V curves and you have the transfer function. More importantly, you can by limiting the input current limit the amount of the I-V curve you traverse. That means that you can arbitrarily keep (for instance) a diode below the point of hard clipping, even for silicon. At the end of the day, I ran to my simulator and tried it out. Sure enough, a feedback diode pair around a CMOS inverter gives nice softly rounded clipping for all values of input up to massive overdrives by simply adjusting the input resistor. It struck me that this is the reason for those silly 8.2K resistors in series with the inputs in the clipping stages of the Muff - they're there to limit the hardness of clipping in the diodes. The transistor stage never clips, the diodes keep it in the linear region. The hardness of clipping can be determined by knowing the input voltage and the I-V curves of the diodes and looking at where Vin/8.2K puts you. Using a CMOS gate lets you set the input resistor to anything you like. I'm working on the C-Muff...

4/6/1998 4:35 PM

R.G.

Soft clipping, CMOS and the Muff

I'm taking a class this week, and the texts didn't show up, so I had to spend some time being roundly bored while the instructor looked for them.

While this was going on, I started pondering the clipping in a CMOS inverter versus the clipping in a diode-feedback opamp, and while I was doodling, I realized that if you consider that an ideal feedback amp forces the current flowing through the input resistor to flow through the feedback element. The higher the input impedance, the more this is true.

In a CMOS gate used as a feedback amp, the input impedance is astronomical. Therefore you can predict almost exactly what the current in the feedback path is, no matter what it is. The current at any point is Vin/Rin.

For a nonlinear element like a diode, you can then just look at the I-V curves and you have the transfer function.

More importantly, you can by limiting the input current limit the amount of the I-V curve you traverse. That means that you can arbitrarily keep (for instance) a diode below the point of hard clipping, even for silicon.

At the end of the day, I ran to my simulator and tried it out. Sure enough, a feedback diode pair around a CMOS inverter gives nice softly rounded clipping for all values of input up to massive overdrives by simply adjusting the input resistor.

It struck me that this is the reason for those silly 8.2K resistors in series with the inputs in the clipping stages of the Muff - they're there to limit the hardness of clipping in the diodes. The transistor stage never clips, the diodes keep it in the linear region. The hardness of clipping can be determined by knowing the input voltage and the I-V curves of the diodes and looking at where Vin/8.2K puts you.

Using a CMOS gate lets you set the input resistor to anything you like.

I'm working on the C-Muff...

4/7/1998 7:16 AM
Jack Orman	In a CMOS gate used as a feedback amp, the input impedance is astronomical. Therefore you can predict almost exactly what the current in the feedback path is, no matter what it is. The current at any point is Vin/Rin. This sounds good. The input impedance of a JFET opamp is 10^12 ohms. Does this fit the definition of astronomical? Would your theoretical circuit work with this type of opamp? I look forward to hearing about your experiments along these lines and seeing the C-Muff circuit. regards, Jack P.S. Filed a patent on it this morning, selling the rights to Peavey this afternoon

4/7/1998 7:16 AM

Jack Orman

In a CMOS gate used as a feedback amp, the
input impedance is astronomical. Therefore
you can predict almost exactly what the
current in the feedback path is, no matter what
it is. The current at any point is Vin/Rin.

This sounds good. The input impedance of a JFET opamp is 10^12 ohms. Does this fit the definition of astronomical? Would your theoretical circuit work with this type of opamp?

I look forward to hearing about your experiments along these lines and seeing the C-Muff circuit.

regards, Jack

P.S. Filed a patent on it this morning, selling the rights to Peavey this afternoon

4/8/1998 5:01 AM
MHelin	R.G. wrote: n a CMOS gate used as a feedback amp, the input impedance is astronomical. Therefore you can predict almost exactly what the current in the feedback path is, no matter what it is. The current at any point is Vin/Rin. In standard opamp (summing) inverter, the input impedance is the same as the value of the input resistor. If you increase this resistor value, the impedance will increase and the current through it will decrease, as well as the current through the feedback resistor. Is this what you mean? So when you increase the feedback resistor, the current doesn't change, only the voltage swing at the output. This is how an amplifier works. Difference with diodes in feedback is just that their resistance is not constant, but when the output goes above 0.6V the diode is clipping. What makes CMOS inverter work differently (I think) is that it's not a linear amplifier itself, and even the clipping diodes can't keep it in linear region. -mikko

4/8/1998 5:01 AM

MHelin

R.G. wrote:

n a CMOS gate used as a feedback
amp, the input impedance is
astronomical. Therefore you can predict
almost exactly what the current in the
feedback path is, no matter what it is.
The current at any point is Vin/Rin.

In standard opamp (summing) inverter, the input impedance is the same as the value of the input resistor. If you increase this resistor value, the impedance will increase and the current through it will decrease, as well as the current through the feedback resistor. Is this what you mean? So when you increase the feedback resistor, the current doesn't change, only the voltage swing at the output. This is how an amplifier works. Difference with diodes in feedback is just that their resistance is not constant, but when the output goes above 0.6V the diode is clipping. What makes CMOS inverter work differently (I think) is that it's not a linear amplifier itself, and even the clipping diodes can't keep it in linear region.

-mikko

4/8/1998 8:01 AM
R.G.	I think I adjusted the content level on my note too low trying to explain. I do know how feedback amps and CMOS inverters work. The point was that this mechanism allows you to set the limit point in the diode V-I characteristic so your clipping may be as soft or hard as you like, and that this is likely a mechanism that the designer of the Tone Bender figured out back when dinosaurs roamed the earth which has been lost, at least to conscious design. There are quite a number of new designs that flow out of this realization. The C-Muff is one, there are several others, including the obvious single side clipping per stage to emulate tubes and the single sided clipping that typical tube stages give. CMOS is there primarily because it has a soft transfer characteristic itself if not used with high feedback, and because it offers a great combination of a lot of function in a small, economical package. But you already knew that...

4/8/1998 8:01 AM

R.G.

I think I adjusted the content level on my note too low trying to explain. I do know how feedback amps and CMOS inverters work.

The point was that this mechanism allows you to set the limit point in the diode V-I characteristic so your clipping may be as soft or hard as you like, and that this is likely a mechanism that the designer of the Tone Bender figured out back when dinosaurs roamed the earth which has been lost, at least to conscious design.

There are quite a number of new designs that flow out of this realization. The C-Muff is one, there are several others, including the obvious single side clipping per stage to emulate tubes and the single sided clipping that typical tube stages give.

CMOS is there primarily because it has a soft transfer characteristic itself if not used with high feedback, and because it offers a great combination of a lot of function in a small, economical package.

But you already knew that...

4/8/1998 6:13 AM
GFR	RG, sorry but I think you are sort of rediscovering the wheel this time. In any opamp of sufficiently high input impedance (J-FET, MOSFET or even some bipolar input like the LM11 and LM108), input currents are very small. So the current at the feedback loop is determined only by the input resistor and is proportional to the input voltage. If you put a resistor in the FB loop, this current will translate into voltage at the output. This I in, V out device is called a tranresistance amplifier, and together with the input R it makes the classic inverting opamp voltage amplifier. If we put something nonlinear in the FB loop, the voltage at output follows the VxI curve of the device. In the specific case of a pair of diodes, by varying the input R we can set the amount of current through them and make them clip (small R) or just softly saturate (large R). By adding a DC offset we can even use just one diode and bias it near the knee (this could be used to "simulate" a Ge transistor with Ge diodes, which are easier to find - I think I will do some research on an opamp FuzzFace...). A resistor in parallel or in series with the diodes will "linearize" the VxI curve a bit. Well, here comes the classic diode clipper fuzz! Log and anti-log amplifiers, by the way, are based on the same principle. If we put a resistor (or diodes, for the matter) in the "FB loop" from collector to base in a commom emitter circuit with a input resistor (like the Muff), it works somewhat like the inverting opamp except that that the transistor amp doesn't have an astronomical input impedance and gain, hence current through the FB loop will not be exactly the current trhough the input resistor, but depending on the values you can even calculate the gain as R2/R1 like in the opamp case with errors in the order of 10 to 20%. I have seen even active filters built around transistors based on this principle, to do low-pass filtering (18 dB/octave) in an octave up device (if you rectify just the fundamental - no harmonics - the octave up gets more pronounced). The CMOS gates thing in audio started being used because they are cheap, you get up to 6 devices in a single package and they don't need split supplies. Then someone found that when you let them saturate, they do this in a much nicer (softer) way than an opamp does, in a way similar to a push-pull tube power amp stage. If you put diodes in the FB loop then the gates will never saturate and you are just using a cheap alternative to opamps. A tip for those who are wishing to experiment with CMOS gates is to use the CD4xxxAE series. The CD4xxxB series have two extra inverters wired in series with the output, to add "buffering" (that's why the "B"). This buffering makes the transition between 0 and 1 sharper (more gain), what is good for digital but not if want it to clip softly. Also check the CD4007, it's a chip with pairs of CMOS transistors you can wire in a lot of ways (COMS inverters, CMOS switches, common drain, commom source, variable resistance, etc.), so you can use only one chip for amplifying and for bypassing the effect! Also from my experience, CMOS fuzzes are very sensitive to hum in the power supply and to picking up noise (specially if you stack a lot of CMOS gain stages in series - even if the gain at each stage is limted by resistors). You will have to layout your circuit board _VERY_ carefully. Make the ground track thicker and shorter as you can. I have found that the gates nearer to the ground pin are generally less noisier. Good luck with your C-Muff. GFR

4/8/1998 6:13 AM

GFR

RG, sorry but I think you are sort of rediscovering the wheel this time.

In any opamp of sufficiently high input impedance (J-FET, MOSFET or even some bipolar input like the LM11 and LM108), input currents are very small. So the current at the feedback loop is determined only by the input resistor and is proportional to the input voltage. If you put a resistor in the FB loop, this current will translate into voltage at the output. This I in, V out device is called a tranresistance amplifier, and together with the input R it makes the classic inverting opamp voltage amplifier.

If we put something nonlinear in the FB loop, the voltage at output follows the VxI curve of the device.

In the specific case of a pair of diodes, by varying the input R we can set the amount of current through them and make them clip (small R) or just softly saturate (large R). By adding a DC offset we can even use just one diode and bias it near the knee (this could be used to "simulate" a Ge transistor with Ge diodes, which are easier to find - I think I will do some research on an opamp FuzzFace...). A resistor in parallel or in series with the diodes will "linearize" the VxI curve a bit. Well, here comes the classic diode clipper fuzz!

Log and anti-log amplifiers, by the way, are based on the same principle.

If we put a resistor (or diodes, for the matter) in the "FB loop" from collector to base in a commom emitter circuit with a input resistor (like the Muff), it works somewhat like the inverting opamp except that that the transistor amp doesn't have an astronomical input impedance and gain, hence current through the FB loop will not be exactly the current trhough the input resistor, but depending on the values you can even calculate the gain as R2/R1 like in the opamp case with errors in the order of 10 to 20%. I have seen even active filters built around transistors based on this principle, to do low-pass filtering (18 dB/octave) in an octave up device (if you rectify just the fundamental - no harmonics - the octave up gets more pronounced).

The CMOS gates thing in audio started being used because they are cheap, you get up to 6 devices in a single package and they don't need split supplies. Then someone found that when you let them saturate, they do this in a much nicer (softer) way than an opamp does, in a way similar to a push-pull tube power amp stage. If you put diodes in the FB loop then the gates will never saturate and you are just using a cheap alternative to opamps.

A tip for those who are wishing to experiment with CMOS gates is to use the CD4xxxAE series. The CD4xxxB series have two extra inverters wired in series with the output, to add "buffering" (that's why the "B"). This buffering makes the transition between 0 and 1 sharper (more gain), what is good for digital but not if want it to clip softly. Also check the CD4007, it's a chip with pairs of CMOS transistors you can wire in a lot of ways (COMS inverters, CMOS switches, common drain, commom source, variable resistance, etc.), so you can use only one chip for amplifying and for bypassing the effect!

Also from my experience, CMOS fuzzes are very sensitive to hum in the power supply and to picking up noise (specially if you stack a lot of CMOS gain stages in series - even if the gain at each stage is limted by resistors). You will have to layout your circuit board *_VERY_* carefully. Make the ground track thicker and shorter as you can. I have found that the gates nearer to the ground pin are generally less noisier.

Good luck with your C-Muff.

GFR

4/8/1998 7:54 AM
R.G.	>RG, sorry but I think you are sort of >rediscovering the wheel this time. Very much so - didn't I make that clear? That was one of the points - this is a wheel that was used in the Tone Bender/Muff family that was not obvious, at least not to me. What is new here (rediscovered, for me at least, I may just be slow today) is realizing that the input resistor completely determines how far you drive the diodes into clipping. You can get soft clipping, hard clipping, or anything in between by (a) limiting the input voltage excursion and (b) setting the input resistor to pick the limits of excursion on the feedback diode V-I curve to get the amount and kind of clipping you want. Of course a CMOS amplifier is not necessary at all. You can use ANY inverting amplifier, be that CMOS, JFET input opamp, ordinary opamp, or even a single transistor - which is what the Tone Bender/Muff family uses. That was one point that I realize my post didn't address very well. What the simulation shows is that you can use some careful design of voltage excursion and current limiting to get a clipping that never gets "hard-edged" even when there is a massive amount of signal limiting. CMOS helps with this by having a natural soft compression in it's transfer function. I really am familiar with opamps, CMOS, and other feedback amps, as well as laying out circuit boards -VERY- carefully.

4/8/1998 7:54 AM

R.G.

>RG, sorry but I think you are sort of >rediscovering the wheel this time.

Very much so - didn't I make that clear? That was one of the points - this is a wheel that was used in the Tone Bender/Muff family that was not obvious, at least not to me.

What is new here (rediscovered, for me at least, I may just be slow today) is realizing that the input resistor completely determines how far you drive the diodes into clipping. You can get soft clipping, hard clipping, or anything in between by (a) limiting the input voltage excursion and (b) setting the input resistor to pick the limits of excursion on the feedback diode V-I curve to get the amount and kind of clipping you want.

Of course a CMOS amplifier is not necessary at all. You can use ANY inverting amplifier, be that CMOS, JFET input opamp, ordinary opamp, or even a single transistor - which is what the Tone Bender/Muff family uses. That was one point that I realize my post didn't address very well.

What the simulation shows is that you can use some careful design of voltage excursion and current limiting to get a clipping that never gets "hard-edged" even when there is a massive amount of signal limiting. CMOS helps with this by having a natural soft compression in it's transfer function.

I really am familiar with opamps, CMOS, and other feedback amps, as well as laying out circuit boards *-VERY-* carefully.

4/8/1998 12:26 PM
GFR	Yes, you sure has something new (or at least out of ordinary) going on, because the standart procedure in designing fuzz boxes is to vary the resistance in the FB loop to achieve different degrees of distortion by "linearizing" more or less the transfer function. This approach of varying the input resistor can lead to some interesting sounds. Also, besides having a natural soft compressor built-in, CMOS have less open loop gain than most opamps, which can also help to "soften" things. And since we will deal with low level signals (to avoid saturating the diodes) the absence of crossover distortion is also an advantage over some opamps. Besides "a) limiting the input voltage excursion and (b) setting the input resistor to pick the limits of excursion on the feedback diode V-I curve to get the amount and kind of clipping you want.", I would sugggest (c) setting a DC bias in the input voltage so that you can chose the operating point on the feedback diode V-I curve. Also you can put diodes, bipolar transistors, FETS, CMOS gates, OTAs, even tubes (like in the Butler tube DI) in the feedback loop to achieve different sounds. I once put a PWM modulator in the loop of an opamp to make a compressor. Since we're looking for flexible designs, there's also the "wave shaping" amps used in signal generators to convert triangle waves to sines. Basically you have various diode chains in parallel with resistors at the feedback loop of an opamp so that the actual gain is determined by which diode chains are conducting. This approximates a "S" shaped transfer function by small straight lines (linear by parts). With enough diodes and resistors, you can approximate any shape you want, even assymetric, and by using pots instead of fixed resistors... Well, it would be very flexible, but very hard to tune. Or you could have a rotary switch that would chose preset resistors and set emulations of different devices. GFR PS RG, I don't have any doubt about your familiarity with electronics, in fact I keep learning new things with every post of yours. I just tend to go a little basic (excuse me, obvious) in my replies in such things as feedback amps, gyrators, boards layout, wave shaping, etc. cause I note that a lot of AMPAGErs don't have formal education in electronics and think they can benefit a little from this, even if the reply is to someone with enough knowledge to find these things basic. So please excuse me if sometimes I'm to boring or sound like I'm teaching the priest how to pray.

4/8/1998 12:26 PM

GFR

Yes, you sure has something new (or at least out of ordinary) going on, because the standart procedure in designing fuzz boxes is to vary the resistance in the FB loop to achieve different degrees of distortion by "linearizing" more or less the transfer function. This approach of varying the input resistor can lead to some interesting sounds. Also, besides having a natural soft compressor built-in, CMOS have less open loop gain than most opamps, which can also help to "soften" things. And since we will deal with low level signals (to avoid saturating the diodes) the absence of crossover distortion is also an advantage over some opamps.

Besides
"a) limiting the input voltage excursion and (b) setting the input resistor to pick the limits of excursion on the feedback diode V-I curve to get the amount and kind of clipping you want.",
I would sugggest
(c) setting a DC bias in the input voltage so that you can chose the operating point on the feedback diode V-I curve.

Also you can put diodes, bipolar transistors, FETS, CMOS gates, OTAs, even tubes (like in the Butler tube DI) in the feedback loop to achieve different sounds. I once put a PWM modulator in the loop of an opamp to make a compressor.

Since we're looking for flexible designs, there's also the "wave shaping" amps used in signal generators to convert triangle waves to sines. Basically you have various diode chains in parallel with resistors at the feedback loop of an opamp so that the actual gain is determined by which diode chains are conducting. This approximates a "S" shaped transfer function by small straight lines (linear by parts). With enough diodes and resistors, you can approximate any shape you want, even assymetric, and by using pots instead of fixed resistors... Well, it would be very flexible, but very hard to tune. Or you could have a rotary switch that would chose preset resistors and set emulations of different devices.

GFR

PS RG, I don't have any doubt about your familiarity with electronics, in fact I keep learning new things with every post of yours. I just tend to go a little basic (excuse me, obvious) in my replies in such things as feedback amps, gyrators, boards layout, wave shaping, etc. cause I note that a lot of AMPAGErs don't have formal education in electronics and think they can benefit a little from this, even if the reply is to someone with enough knowledge to find these things basic. So please excuse me if sometimes I'm to boring or sound like I'm teaching the priest how to pray.

Page 1 of 3

Next>

Last Page>>