Haven't breadboarded this yet, still cipherin component values...
Gonna post to tubes, but figured they'd just throw a hissyfit
I decided not to use a real tube this go round. Anyways, respin
of a very old idea by Schrade (RCA 1938)
He suggested a plate feedback method for emulating a Triode
abusing a 6L6 beam power tube. Which just so happens the
6L6 had the same sort of square law behavior as do modern
JFET devices.
He suggested that negative feedback in the Cathode was not
the same as negative feedback from Plate to Grid. At least not
as the load sees it... And the distinction does seem critical.
Only one JFET "SQLW" is working as a non-linear device with
light Drain Feedback, just as Schrade's Beam Power Tube did.
The rest rely on heavy linearization from Source Feedback, as
Source Followers and Constant Current Sinks and Sources.
Many emulations become unrealistic when they mimic a single
curved transfer function into a fixed resistance, and do not
consider real misbehaving loadlines over an entire operating
area... No big deal in a pre-amp, potentially very important if
driving a loudspeaker.
Anyways, I present a general purpose triode emulation.
Not intended to rigidly mimic any one specific device.
--------------- My opinions, not presented as fact ------------
A truly great amplifier should faithfully reproduce music
without distortion, but there is absolutely no reason why
such an amplifier should faithfully reproduce clipping!
I remind readers of some nice lessons that Triodes teach us
that should not be the exclusive domain of hollow state!
Transistors can mimic all of these things if we are clever.
One extreme is softly clamped to the Grid diode, and even if
transgressing slightly into A2, clipping on this end is all but
impossible.
The other extreme is softly crushed into an ever increrasing
high impedance, that disconnects the load to go do its own
mechanical thing (whatever that is), rather than clamp it to
a low impedance power rail. Blindly driving the coil out of the
gap and cooking it....
The one sided approach to clipping generates harmonics of
the lowest possible order (more 2nd than 3rd). And the least
objectionable intermodulation products.
Pretending an amplifier will not clip, is like taking the shocks
off your truck for a "smoother ride". And many "fix" this by
simply building a bigger truck with more travel? I don't think
so. You will always find an even deeper pothole to bottom.
At the extremes of approach to clipping, "damping factor" is
not damping anything, but making a problem far worse.
Sometimes we forget that clipping can be caused by either
the program material, or the loadspeaker load. We are not
safe just because the source has been compressed within
expected limits.
-----------------------------------------------------------------------------
I present elsewhere (SE Diode, posted in Tubes) another
emulation that may hold certain advantages. If you wish,
string together Schottkys and Resistors to approximate
an I=E^1.5 curve without the vacuum diode reference.
------------------------------More opinion--------------------------
I also hold the opinoin that any deliberately non-linear amp
should be non-linear in one and only one stage, unless the
non-linear stages act together as one, without smearing of
phase inbetween. Twice modulated products are probably
not well related to the original music.
Since the loudspeaker is surely a source of distortion that
we cannot live without. Our amplifier stage best suited to
provide helpful clip supressing distortion is that final one
which connects most directly to it.
--------------------------------------------------------------------------
Unless you make some changes, the feedback scheme will
fail around 150V Plate... A first draft that may not work at
all, so don't go running out to build one just yet.
Gonna post to tubes, but figured they'd just throw a hissyfit
I decided not to use a real tube this go round. Anyways, respin
of a very old idea by Schrade (RCA 1938)
He suggested a plate feedback method for emulating a Triode
abusing a 6L6 beam power tube. Which just so happens the
6L6 had the same sort of square law behavior as do modern
JFET devices.
He suggested that negative feedback in the Cathode was not
the same as negative feedback from Plate to Grid. At least not
as the load sees it... And the distinction does seem critical.
Only one JFET "SQLW" is working as a non-linear device with
light Drain Feedback, just as Schrade's Beam Power Tube did.
The rest rely on heavy linearization from Source Feedback, as
Source Followers and Constant Current Sinks and Sources.
Many emulations become unrealistic when they mimic a single
curved transfer function into a fixed resistance, and do not
consider real misbehaving loadlines over an entire operating
area... No big deal in a pre-amp, potentially very important if
driving a loudspeaker.
Anyways, I present a general purpose triode emulation.
Not intended to rigidly mimic any one specific device.
--------------- My opinions, not presented as fact ------------
A truly great amplifier should faithfully reproduce music
without distortion, but there is absolutely no reason why
such an amplifier should faithfully reproduce clipping!
I remind readers of some nice lessons that Triodes teach us
that should not be the exclusive domain of hollow state!
Transistors can mimic all of these things if we are clever.
One extreme is softly clamped to the Grid diode, and even if
transgressing slightly into A2, clipping on this end is all but
impossible.
The other extreme is softly crushed into an ever increrasing
high impedance, that disconnects the load to go do its own
mechanical thing (whatever that is), rather than clamp it to
a low impedance power rail. Blindly driving the coil out of the
gap and cooking it....
The one sided approach to clipping generates harmonics of
the lowest possible order (more 2nd than 3rd). And the least
objectionable intermodulation products.
Pretending an amplifier will not clip, is like taking the shocks
off your truck for a "smoother ride". And many "fix" this by
simply building a bigger truck with more travel? I don't think
so. You will always find an even deeper pothole to bottom.
At the extremes of approach to clipping, "damping factor" is
not damping anything, but making a problem far worse.
Sometimes we forget that clipping can be caused by either
the program material, or the loadspeaker load. We are not
safe just because the source has been compressed within
expected limits.
-----------------------------------------------------------------------------
I present elsewhere (SE Diode, posted in Tubes) another
emulation that may hold certain advantages. If you wish,
string together Schottkys and Resistors to approximate
an I=E^1.5 curve without the vacuum diode reference.
------------------------------More opinion--------------------------
I also hold the opinoin that any deliberately non-linear amp
should be non-linear in one and only one stage, unless the
non-linear stages act together as one, without smearing of
phase inbetween. Twice modulated products are probably
not well related to the original music.
Since the loudspeaker is surely a source of distortion that
we cannot live without. Our amplifier stage best suited to
provide helpful clip supressing distortion is that final one
which connects most directly to it.
--------------------------------------------------------------------------
Unless you make some changes, the feedback scheme will
fail around 150V Plate... A first draft that may not work at
all, so don't go running out to build one just yet.
Attachments
Explaination
The cascode stack on the right multiples drain voltage by 12,
or divides VPlate voltage by 12. Same thing, only different...
CC2, CC3, and the Zenier simply set a fixed voltage drop from
the drain to the gate for bias. CC3 Sink may be omitted when
IPlate can tolerate the added DC.
CC1, Source Follower 1, and the LED set a small voltage drop
to compensate for VGS rise (is depletion bias an anti-drop?).
Anyways, you might do away with one drop diode or the other
if you tolerate some DC through the mixer.
The pair of 47K's in the feedback mixer (to the gate of SQLW)
average the input signal against the drain feedback signal.
This point is not expected to zero out, but move around the
triode and square law region of the JFET's operating area.
SQLW sees 1/24th of the VPlate (suggesting a Mu of 24!),
but also only 1/2 of VGrid (knocking Mu back down to 12).
J203 currents in the schematic above, refer to part binning
by wide open VGS=0 current. Not currents one is expected
to measure in the running circuit.
And I misspelled "O.H. Schade" not "Schrade" . Whoops!!
Graphs to show square law similarity between 6L6 and J203,
especially in the constant current region. I think if the scales
were presented a bit different, they would look even more akin.
No reason the 1938 trick wouldn't work just as well today!
Remember, those 6L6 Triode curves are not "Triode Strapped",
but Triode emulation based on Schade's feedback scheme.
The cascode stack on the right multiples drain voltage by 12,
or divides VPlate voltage by 12. Same thing, only different...
CC2, CC3, and the Zenier simply set a fixed voltage drop from
the drain to the gate for bias. CC3 Sink may be omitted when
IPlate can tolerate the added DC.
CC1, Source Follower 1, and the LED set a small voltage drop
to compensate for VGS rise (is depletion bias an anti-drop?).
Anyways, you might do away with one drop diode or the other
if you tolerate some DC through the mixer.
The pair of 47K's in the feedback mixer (to the gate of SQLW)
average the input signal against the drain feedback signal.
This point is not expected to zero out, but move around the
triode and square law region of the JFET's operating area.
SQLW sees 1/24th of the VPlate (suggesting a Mu of 24!),
but also only 1/2 of VGrid (knocking Mu back down to 12).
J203 currents in the schematic above, refer to part binning
by wide open VGS=0 current. Not currents one is expected
to measure in the running circuit.
And I misspelled "O.H. Schade" not "Schrade" . Whoops!!
Graphs to show square law similarity between 6L6 and J203,
especially in the constant current region. I think if the scales
were presented a bit different, they would look even more akin.
No reason the 1938 trick wouldn't work just as well today!
Remember, those 6L6 Triode curves are not "Triode Strapped",
but Triode emulation based on Schade's feedback scheme.
Attachments
Took out most of the overzealous "optional parts".
Hopefully easier now to grasp the basic concepts.
If SQLW was a MOSFET with higher gate voltage,
might do away with the zenier next? Its a thought.
Schade's original RCA pages can be found here.
http://www.one-electron.com/Misc_Docs.html
Hopefully easier now to grasp the basic concepts.
If SQLW was a MOSFET with higher gate voltage,
might do away with the zenier next? Its a thought.
Schade's original RCA pages can be found here.
http://www.one-electron.com/Misc_Docs.html
Attachments
Hah, without knowing I seem to have "invented" that, too (although I have the book the Schade article is taken from).
If the goal is only to get triode/UL-like curves but not replacing actual tubes in circuits, I have a simple JFET version (sligthly simpler that yours) that works pretty well, ie as preamp stage. Everything is just scaled down in level.
The basic idea is exactly the same as Schade's: feed part of the plate voltage back to the grid, he did it with passive AC divider and then letting the gate voltage ride on top of that (pg. 41, fig.33c). If you have an input xformer this scheme as is would work equally well with any FET.
I divide down the drain-to-input voltage which then feeds the gate, after being offsetted with a current source. This looses some gain, though not much, as the resistor ratio can be quite high. The gate is isolated with buffer to retain high input impedance. One can/must use a buffer (bjt is OK) an the drain if the divider needs to be isolated from the drain load.
As usual with FETs, some measures must be taken to stabilize the operation point thermally and it needs to be trimmable too, to compensate device-to-device changes (or you select your FETs very well). Some modulated cascode option are doable to linearize the output characteristics very well and shift then a bit further to the left (UL-style curves), but then biasing gets tricky. As I was after that typical triode curvature's sound and supply sag gain compression character (not a HiFi-application, but electronic instrument amp) I settled with this simple arrangement.
- Klaus
If the goal is only to get triode/UL-like curves but not replacing actual tubes in circuits, I have a simple JFET version (sligthly simpler that yours) that works pretty well, ie as preamp stage. Everything is just scaled down in level.
The basic idea is exactly the same as Schade's: feed part of the plate voltage back to the grid, he did it with passive AC divider and then letting the gate voltage ride on top of that (pg. 41, fig.33c). If you have an input xformer this scheme as is would work equally well with any FET.
I divide down the drain-to-input voltage which then feeds the gate, after being offsetted with a current source. This looses some gain, though not much, as the resistor ratio can be quite high. The gate is isolated with buffer to retain high input impedance. One can/must use a buffer (bjt is OK) an the drain if the divider needs to be isolated from the drain load.
As usual with FETs, some measures must be taken to stabilize the operation point thermally and it needs to be trimmable too, to compensate device-to-device changes (or you select your FETs very well). Some modulated cascode option are doable to linearize the output characteristics very well and shift then a bit further to the left (UL-style curves), but then biasing gets tricky. As I was after that typical triode curvature's sound and supply sag gain compression character (not a HiFi-application, but electronic instrument amp) I settled with this simple arrangement.
- Klaus
Attachments
Dimitri's paper:
http://www.aes.org/e-lib/browse.cfm?elib=12627
Anybody have it and willing to share the essentials (maybe Dimitri himself)?
http://www.aes.org/e-lib/browse.cfm?elib=12627
Anybody have it and willing to share the essentials (maybe Dimitri himself)?
Take a look at the ZV9, where the cascode is modulated
so as to get the loadline cancellation that you might be
looking for.
so as to get the loadline cancellation that you might be
looking for.
Thanks Nelson.
ZV9 indeed was the thing that got me going, although my goals were different, the main one being to get that gain change vs supply voltage behaviour. De-cascoding of sorts...
As I remembered now (um, getting old, it seems), Mr Nikitin (x-pro) has an article about a FET triode on his website:
http://www.ant-audio.co.uk/index.php?cat=post&qry=philosophy
and a product als well:
http://www.ant-audio.co.uk/Theory/ANT Amber Headphone Amplifier.pdf
(which also references ZV9 as an inspiration).
Finally, I've found a circuit design based on Dimitri's AES article:
http://www.freestompboxes.org/viewtopic.php?f=13&t=902
Klaus
ZV9 indeed was the thing that got me going, although my goals were different, the main one being to get that gain change vs supply voltage behaviour. De-cascoding of sorts...
As I remembered now (um, getting old, it seems), Mr Nikitin (x-pro) has an article about a FET triode on his website:
http://www.ant-audio.co.uk/index.php?cat=post&qry=philosophy
and a product als well:
http://www.ant-audio.co.uk/Theory/ANT Amber Headphone Amplifier.pdf
(which also references ZV9 as an inspiration).
Finally, I've found a circuit design based on Dimitri's AES article:
http://www.freestompboxes.org/viewtopic.php?f=13&t=902
Klaus
KSTR said:
Klaus,
my circuit and my article pre-dates ZV8 and 9 🙂 . However there is an old patent (I've got a reference at home in email from dimitri and will post it later) describing exactly the same circuit idea, as usual, re-discovered much later.
Alex
I never claimed to have invented it, clearly this existed in 1938.
Well before I was born, and even Schade refers to earlier work...
I was merely disappointed with "Fetzer Valve" implemetation
I had seen recently on the web. Nothing but a source resistor...
Did not (in my opinion) behave at the drain anything like a real
triode. The output impedance far too high, and of no particular
curvature as seen by the load.
Nor any of those cascodified below the knee schemes that
hide the real plate load from participating in the equation.
Though I will almost certainly want to abuse that trick for
linear stages in a pre-amp that only drive high impedance.
What is needed is a smooth plate resistance curve between
x=z*y^1.5 and x=z*y^2. And a slight "lean to the right" with
increasing voltage and/or decreasing current....
I wish to drive a loudspeaker, and the load impedance will
be dynamic and highly variable. All load lines must behave
with reasonable similarity, and smooth crush toward cutoff.
---------------------------------------------------------------------------------
Whats the cleanest square law you can buy in a power MOSFET?
Does anything solid (and simple) aproximate a 3/2 power law
over a useful range? Other than a stack of diodes and bypass
resistors?
Well before I was born, and even Schade refers to earlier work...
I was merely disappointed with "Fetzer Valve" implemetation
I had seen recently on the web. Nothing but a source resistor...
Did not (in my opinion) behave at the drain anything like a real
triode. The output impedance far too high, and of no particular
curvature as seen by the load.
Nor any of those cascodified below the knee schemes that
hide the real plate load from participating in the equation.
Though I will almost certainly want to abuse that trick for
linear stages in a pre-amp that only drive high impedance.
What is needed is a smooth plate resistance curve between
x=z*y^1.5 and x=z*y^2. And a slight "lean to the right" with
increasing voltage and/or decreasing current....
I wish to drive a loudspeaker, and the load impedance will
be dynamic and highly variable. All load lines must behave
with reasonable similarity, and smooth crush toward cutoff.
---------------------------------------------------------------------------------
Whats the cleanest square law you can buy in a power MOSFET?
Does anything solid (and simple) aproximate a 3/2 power law
over a useful range? Other than a stack of diodes and bypass
resistors?
The references are of two different types.
Simply cascoding the JFET at a low voltage creates the triode curve,
and choosing the variation of Vds of the gain device (which I call
cascode modulation) creates the load-line cancellation.
Simply cascoding the JFET at a low voltage creates the triode curve,
and choosing the variation of Vds of the gain device (which I call
cascode modulation) creates the load-line cancellation.
A triode is a set of curves, not restricted to any single load line.
The cascode linearizing trick for JFETs is neat, and I'll probably
even make use of it some day (into a high impedance, or with
a heavy duty follower). But definitely not a general purpose
emulation of any Triode working into any other load.
The cascode linearizing trick for JFETs is neat, and I'll probably
even make use of it some day (into a high impedance, or with
a heavy duty follower). But definitely not a general purpose
emulation of any Triode working into any other load.
There are some Mosfet like devices called Static Induction Transistors that have very short channel length. They have triode like curves, without the need for the drain to gate resistive feedback, since the drain potential affects the field in the channel. Unfortunately, they don't seem to be commercially available, just research devices. I did hear some while back however that someone had built an amplifier with them that was commercially available at very high $$$.
The construction of the SIT devices even looks like a triode, but embedded in silicon. The gate is just a linear grid like array of dots.
Don
The construction of the SIT devices even looks like a triode, but embedded in silicon. The gate is just a linear grid like array of dots.
Don
I have no idea which SIT these curves were measured from?
Found this graph in a paper "2003_EPE_Final_Paper.pdf" found
while searching "Static Induction Transistor".
Sorta Triodish, very A2... I'm sure than when/if real devices fall
into our hands, we can then cherry pick a better set of curves.
Anyways, it all the same thing only completely different.
The main difference between otherwise identical Triode
and Pentode is screening out the plate feedback. Higher
voltage gain and impedance at the loss of good linearity
and intrinsic damping factor (low impedance).
Non-Linearity was eventually tweaked to a square law
in beam power tubes, so it could be easily cancelled in
push pull or differential. This carries on to audio grade
FETs even today.
The linearizing feedback can be replaced externally in
Tetrode, Pentode, and solid state devices with Pentode
compatible curves. But works best in devices that were
close to linear to begin with, still very good but perhaps
not the absolute best with square-lawish devices....
I got my eye on three NPT IGBTs that seem very linear,
but only above 2A. Can't seem to find one small enough
to bring the knee in that curve down where I would like
to see it. And of those that were linear, only these few
seem to have reasonable low capacitance..
IRGB4B60KD1, SGP6N60UFD, ISL9V2040D3S
You could build a heck of an OTL into 2 ohms or less...
Emulated Triode vs Constant Current...
Or deliberately abuse them (linearly) at 25% duty cycle
at some high rate of pulse them to keep the current up
without too low a supply voltage to match speakers.
Linear/OFF/Linear/OFF/Linear/OFF, not the same as
pulse width modulation...
Found this graph in a paper "2003_EPE_Final_Paper.pdf" found
while searching "Static Induction Transistor".
Sorta Triodish, very A2... I'm sure than when/if real devices fall
into our hands, we can then cherry pick a better set of curves.
Anyways, it all the same thing only completely different.
The main difference between otherwise identical Triode
and Pentode is screening out the plate feedback. Higher
voltage gain and impedance at the loss of good linearity
and intrinsic damping factor (low impedance).
Non-Linearity was eventually tweaked to a square law
in beam power tubes, so it could be easily cancelled in
push pull or differential. This carries on to audio grade
FETs even today.
The linearizing feedback can be replaced externally in
Tetrode, Pentode, and solid state devices with Pentode
compatible curves. But works best in devices that were
close to linear to begin with, still very good but perhaps
not the absolute best with square-lawish devices....
I got my eye on three NPT IGBTs that seem very linear,
but only above 2A. Can't seem to find one small enough
to bring the knee in that curve down where I would like
to see it. And of those that were linear, only these few
seem to have reasonable low capacitance..
IRGB4B60KD1, SGP6N60UFD, ISL9V2040D3S
You could build a heck of an OTL into 2 ohms or less...
Emulated Triode vs Constant Current...
Or deliberately abuse them (linearly) at 25% duty cycle
at some high rate of pulse them to keep the current up
without too low a supply voltage to match speakers.
Linear/OFF/Linear/OFF/Linear/OFF, not the same as
pulse width modulation...
Attachments
There seem to be several SIT family devices in the works...
Gate-Controlled Punch Through Transistor
www.eng.auburn.edu/~wilambm/pap/2003/punchthrough_Boise.pdf
Low current definately doesn't tell the whole story.
At very high currents is more like Pentode/MOSFET,
you won't see that in this picture... Read the PDF!
Gate-Controlled Punch Through Transistor
www.eng.auburn.edu/~wilambm/pap/2003/punchthrough_Boise.pdf
Low current definately doesn't tell the whole story.
At very high currents is more like Pentode/MOSFET,
you won't see that in this picture... Read the PDF!
Attachments
I think what you are looking for is a soft limiter as soppposed to a hard limiter as per transistors.
I built one in the 1980's from a cct out of a magazine.
I have tried to rebuild the cct from memory but it is clearly wrong as the bias is modulated by the second stage.
It needs another bias pair of resistors.
If I remember correctly you tweak the pot until you get the sound you want.
I enclose the cct digram anyway.
I built one in the 1980's from a cct out of a magazine.
I have tried to rebuild the cct from memory but it is clearly wrong as the bias is modulated by the second stage.
It needs another bias pair of resistors.
If I remember correctly you tweak the pot until you get the sound you want.
I enclose the cct digram anyway.
Attachments
Now, let me ask you, why do you need triode output characteristics (plate current/plate voltage)? Is your intent to connect speaker to the output?
- Status
- Not open for further replies.