Hi,
I just thought I'd like to draw attention to the fact that crossover distortion, an important subject for solid state power amplifiers (and the subject of much discussion) seems to be rarely mentioned when it comes to tube amplifiers.
It seems to me that, for tubes, the subject of output stage biasing mostly comes down to quiescent power dissipation. Fact is, I suppose, that the standing current requirements for low crossover distortion are higher for tube amplifiers, and would therefore impact power dissipation more than with solid state (particularly bipolar output stages).
But the same principles should apply. We need a relatively constant gm through the crossover region. OK, tubes will tend to have a less abrupt discontinuity, but this will be partly negated by lower loop gain.
I did dig this article up:
Amplifier auto bias circuits: Class-A, Class-AB!
As the article explains, part of the problem is the fact that tubes do age, and this complicates matters. But it does seem to me that, mostly, choice of bias level is somewhat arbitrary, and that greater efforts ought to be made to try to establish and maintain optimum bias.
Or maybe just rely on high loop gain to iron out (hopefully fairly soft) discontinuities?
I just thought I'd like to draw attention to the fact that crossover distortion, an important subject for solid state power amplifiers (and the subject of much discussion) seems to be rarely mentioned when it comes to tube amplifiers.
It seems to me that, for tubes, the subject of output stage biasing mostly comes down to quiescent power dissipation. Fact is, I suppose, that the standing current requirements for low crossover distortion are higher for tube amplifiers, and would therefore impact power dissipation more than with solid state (particularly bipolar output stages).
But the same principles should apply. We need a relatively constant gm through the crossover region. OK, tubes will tend to have a less abrupt discontinuity, but this will be partly negated by lower loop gain.
I did dig this article up:
Amplifier auto bias circuits: Class-A, Class-AB!
As the article explains, part of the problem is the fact that tubes do age, and this complicates matters. But it does seem to me that, mostly, choice of bias level is somewhat arbitrary, and that greater efforts ought to be made to try to establish and maintain optimum bias.
Or maybe just rely on high loop gain to iron out (hopefully fairly soft) discontinuities?
After reading a couple of solid-state power amplifier design books I set out to quantify crossover distortion in a tube amp I built. I had assumed that there would be a significant amount of distortion generated at the gm discontinuity when one tube cut off and the other had to completely take over.
I took detailed measurements in a tube amp I had built with no feedback around the output transformer and 30% local feedback in the output stage. I measured distortion at various power levels with 50mA, 60mA, and 70mA idle current in the output tubes and found negligible difference in distortion (although there were definite benefits of lower Zout at higher idle currents).
I got distortion pretty low in that open-loop design but gm-doubling distortion was still undetectable. I suspect that most tube amps suffer from the same issue; that other sources of distortion will mask gm-doubling distortion. It seems that you have to take extraordinary pains to minimize other distortions before this will be the dominant problem, like it is in solid-state amps.
I have developed a much lower distortion high-swing driver stage that I will be using in future experiments to quantify crossover distortion from gm-doubling in the crossover region on tube amps, but I haven't gotten there yet.
I took detailed measurements in a tube amp I had built with no feedback around the output transformer and 30% local feedback in the output stage. I measured distortion at various power levels with 50mA, 60mA, and 70mA idle current in the output tubes and found negligible difference in distortion (although there were definite benefits of lower Zout at higher idle currents).
I got distortion pretty low in that open-loop design but gm-doubling distortion was still undetectable. I suspect that most tube amps suffer from the same issue; that other sources of distortion will mask gm-doubling distortion. It seems that you have to take extraordinary pains to minimize other distortions before this will be the dominant problem, like it is in solid-state amps.
I have developed a much lower distortion high-swing driver stage that I will be using in future experiments to quantify crossover distortion from gm-doubling in the crossover region on tube amps, but I haven't gotten there yet.
To me, an amplifier with bias set-up by cathode resistors is Class A or A/B, for example a Telefunken 10W 2xEL86 movie sound amplifier, while one with negative control grid voltage is Class B, for example a Dynacord 2x80W PublicAdress amplifier with 2x2xEL34. I read, that cathode circuit push-pull power tubes amplifiers may show very different harmonic distortion at lower levels depending on load norm and phase.
when one tube cuts out in a pp w/ct opt two things happen simultaneously:
gm drops to 1/2 and load impedance as seen by the tube also drops as there is only 1/2 of the primary active ---> could it be that both effects cancel out ?
Taking an EL84 as an example, the Mullard data sheet shows that with Va = Vg2 = 250V the gm value drops from 16 mA/V at Vg1=0 down to 6 mA/V at Vg1=-11V. The gm drops to zero when Vg1 = -17.5V. This is the context for gm doubling in Class AB, for this valve.
Unfortunately, I don't think they cancel out. There is some interesting discussion here:
Inherent distortion in class AB
If you compare the output level where the switching is happening respect to rated power most class AB solid state amps would be considered pure class B amps in the tube world!
The other problem of solid state amps is their very low Zout! Yes. Having low Zout they have no control on the current circulating into the coil of a driver. For low frequency drivers this is a problem! They control the voltage for maximum damping. But unfortunately it is the current the real source of magnetic field and if there is no control on the current it will be distorted beacuse the driver is not really linear (due to Bxl and compliance). A non-linear issue from the amplifier thrown into a non linear load can get really nasty. That's easily a second order effect with a good tube amp driving the same driver.
In the end the fact that people like the tube sound more might well be because they actually distort less when driving dynamic speakers. That's not always the case but more common than one might think. And also explains why solid state amps with tiny distortion levels on dummy loads can sound different....
Just look at the tube's gm curves to see what happens in the crossover region.
Most tubes are incapable of producing gm doubling. If you look at the gm curves for the E55L below (typical tube gm curve, labelled S), you will notice the S shape of the gm curve. When flipped around and overlapped for the P-P case you would get a flat bottomed V shaped gm sum, with rounded corners at the bottom edges. So gm sag in the overlap region is the problem. As one increases the overlap region (going toward class A) the bottom of the flatted V comes up like a glass being filled, until, with complete overlap, you get nearly constant gm.
Sagging gm in the center will produce odd harmonics, especially 3rd.
Most tubes are incapable of producing gm doubling. If you look at the gm curves for the E55L below (typical tube gm curve, labelled S), you will notice the S shape of the gm curve. When flipped around and overlapped for the P-P case you would get a flat bottomed V shaped gm sum, with rounded corners at the bottom edges. So gm sag in the overlap region is the problem. As one increases the overlap region (going toward class A) the bottom of the flatted V comes up like a glass being filled, until, with complete overlap, you get nearly constant gm.
Sagging gm in the center will produce odd harmonics, especially 3rd.
Attachments
They sort of cancel out. The cumulative AB pentode Gm is not exactly constant withing plate current range, but it is close to constant. Under certain circumstances outlined in data sheets for a few pentodes like 807, an AB pentode stage may be quite linear, even more linear than Class A 2A3 PP.
Last edited:
The plots I saw in Douglas Self's book showed a similar V-shaped gm sums for BJTs and Mosfets, only as you increased overlap you would get a spike that would come up out of the center of the "V" so that you would eventually get a spike sticking up above everything else in the overlap section with troughs next to it. If you further increased overlap, you would get a smooth mound in the overlap region.
There was a point that Self called "optimal class-B" where the spike wouldn't be too tall but the troughs wouldn't be to deep either and overall distortion was minimized. Mosfets were worse than BJTs.
One of these days for curiosity's sake I'm going to take my Unity-Coupled amp and keep decreasing the bias on the output tubes until distortion starts to rise quickly. I'd just like to know where the "optimal class-B" point is. I don't think I would set the amp there, since I want to have the Zout nice and low but I just wonder if we are talking 5mA or 35mA or what.
Self's problem is, that "optimum" for class B highly depends on temperature, and since his "blameless" transistors are biased by voltage depending on temperature of the heatsink that is far from temperatures of transistor dies, it is never optimal, and transistor amps designed by "blameless" OpAmp topology as the result sound worse than tube amps (one of reasons).
In order to overcome this problem Douglas built an amp with a PCB full of 5532 chips in parallel. ;-)
It may look like one tube where each opamp represents a gap between wires of the control grid. 🙂
In order to overcome this problem Douglas built an amp with a PCB full of 5532 chips in parallel. ;-)
It may look like one tube where each opamp represents a gap between wires of the control grid. 🙂
sser2,
What did you mean by:
"..and also an argument for no-NFB pentodes being superior to triodes in an output stage".
Is that: pentodes in push pull that are without negative feedback, Versus
triodes in push pull that are without negative feedback?
Or is that: pentodes in push pull that are with negative feedback, Versus
triodes in push pull that are without feedback?
Or what?
Can you point to the literature, please?
What did you mean by:
"..and also an argument for no-NFB pentodes being superior to triodes in an output stage".
Is that: pentodes in push pull that are without negative feedback, Versus
triodes in push pull that are without negative feedback?
Or is that: pentodes in push pull that are with negative feedback, Versus
triodes in push pull that are without feedback?
Or what?
Can you point to the literature, please?
Because thay have high plate resistance and could get low damping factor and so control the current rather than the voltage. However because of the output transformer it's never a true constant current source, still a near constant power source, but you do get DF<1 easy.
The other thing is very likely bass alignment of most commercial speakers won't work well. Needs a change\re-tunining or make dedicated speakers. Not a big problem.....
Exactly. Pentode amplifier with no NFB does not need to reproduce bass. That should be a task for a separate amplifier and bass speaker, both optimized for the purpose.
Otherwise, no-NFB pentode power stage can be used with a special low Qms full range speakers that do not rely on electric damping. Such speakers were made during 20s - 50s, before the curse of closed and vented box designs, which lead to another curse of global NFB required to drive such speakers. For almost two decades after introduction of power pentodes, global NFB was not used in pentode amplifiers, including high end radios of the time, and there were no complaints about quality of bass.
Properly done pentode power stage does not need global or local NFB to reduce distortion because it has low intrinsic open loop distortion. THD of a pentode PP stage employing 807 or a pair of those magnificent TV sweep tubes may be an order of magnitude less than that of a comparable PP 2A3 triode stage. No revelation here, this is all in tube data sheets.
Driving speakers with high output impedance source, such as a no-feedback pentode amplifier, reduces speaker distortion compared to driving them with low output impedance source.
Otherwise, no-NFB pentode power stage can be used with a special low Qms full range speakers that do not rely on electric damping. Such speakers were made during 20s - 50s, before the curse of closed and vented box designs, which lead to another curse of global NFB required to drive such speakers. For almost two decades after introduction of power pentodes, global NFB was not used in pentode amplifiers, including high end radios of the time, and there were no complaints about quality of bass.
Properly done pentode power stage does not need global or local NFB to reduce distortion because it has low intrinsic open loop distortion. THD of a pentode PP stage employing 807 or a pair of those magnificent TV sweep tubes may be an order of magnitude less than that of a comparable PP 2A3 triode stage. No revelation here, this is all in tube data sheets.
Driving speakers with high output impedance source, such as a no-feedback pentode amplifier, reduces speaker distortion compared to driving them with low output impedance source.
I disagree. Curent amp is the in theory the best for that task. In practice there are compromises so it's the usual business to get the best balance among all requirements.....
No need to go back to the fifities. There are modern drivers that are suitable. It's just that it is not the industry standard for mass production because contant voltage source is chaeper....
Actually modern quality drivers can be surely better regarding suspensions that can be fairly linear within limits. That is of course an advantage regardless of amp technology.
We seem to drifting off the subject matter, which is a pity.
Firstly, the thread isn't supposed to be a transistor vs tube debate. We all know that transistor crossover distortion is much more abrupt than in tubes, that care has to be taken with temperature tracking, and that transistor class AB amps tend to have much higher loop gain. And that steps can be taken to mitigate, such as use of parallel high-frequency transistors (to minimise thermal effects and ensure decent loop gain at hf).
Secondly, the thread is certainly NOT anything to do with a voltage-drive vs current-drive debate. I personally think the latter is severely flawed on several counts, but that would be the subject of a different thread.
Now surely someone has something constructive to say concerning tube crossover distortion?
Firstly, the thread isn't supposed to be a transistor vs tube debate. We all know that transistor crossover distortion is much more abrupt than in tubes, that care has to be taken with temperature tracking, and that transistor class AB amps tend to have much higher loop gain. And that steps can be taken to mitigate, such as use of parallel high-frequency transistors (to minimise thermal effects and ensure decent loop gain at hf).
Secondly, the thread is certainly NOT anything to do with a voltage-drive vs current-drive debate. I personally think the latter is severely flawed on several counts, but that would be the subject of a different thread.
Now surely someone has something constructive to say concerning tube crossover distortion?
A strictly square law device (I = kVgVg) gives a linear ramping gm versus grid/gate drive voltage. Two such devices in P-P complementary conduction sum to a constant gm (much desired). Deviations from there produce all the problems in P-P operation.
Looking at the typical S shaped tube gm curve, at low current (left side), the gm is curving upward with further V drive. This portion is operating with greater than square law, and when P-P summed with a similar upward curving complementary device produces a gm sum that sags in the center but curves upward away from the center.
On the right side of the S curve, gm is bending over (downward from a linear ramp at least) from cathode saturation. This is operation with less than square law. Two of these curved sections in P-P will produce a bumped gm sum. And two completely linear devices (1.0 law) with constant gm will sum to a rectangular bump, or doubled gm in the overlap region.
Fortunately tube grid1 operation mostly centers around square law (central region of the S curve). Grid2 operation is typically nearer to 3/2 power law.
One would have to under bias (extreme idle current) class A to get the bending over parts of gm to overlap and cause gm "doubling". Class A, correctly arranged, just has the two S curves overlapped to null out gm sum variation everywhere. (ends of the S parts compensating each other)
Bipolar transistors have linear gm when current mode operated. Which most VAS stages are. And so would badly gm double in crossover if left alone. Only by limiting the Voltage offset between the N and P type follower's Bases to the threshold turn-on region can the gm bump be reduced within reason.
Mosfets start out conduction with a small region of curved over gm before becoming square law, so they also have a gm bump at low current crossover.
Higher idle current increases the width (but not height) of the bump, once formed.
Tubes pretty much always have curving up gm near turn on via grid 1 drive, so they can't gm double, but will always gm droop in the center with low idle current. Higher idle current will overlap the regions of square law. (the linear ramping gm regions) Those will produce a flat gm sum section. So sufficient idle current will give a broad constant gm region (in the class A region). Surrounded by the V shaped increasing gm of the class B regions up until saturation. Causing 3rd and some other odd harmonic distortions with sufficient signal level above the class A region.
Grid 2 (3/2 power law) drive however could theoretically produce some gm humping up at cross over. But fortunately grid 2 also starts out curving upward near turn-on, then transitions to near 3/2 power law through most of its range. Since it passes through square law briefly in that transition region, it has an optimum bias point to overlap the two square law sections to give constant gm across some crossover region width without humping up. But then outside that cross-over region it has the usual V shaped gm sum, but a curving over/outward type V rather than a straight ramp sided V. So somewhat different type odd harmonics above low level than grid 1 drive.
Crazy or Twin Drive has constant effective gm throughout most of the individual tube operating region. This would cause gm doubling at crossover if it were not for the tube reverting briefly to grid1 only operation effectively near cut-off. This is because grid 2 gm is dropping toward zero while the current limiting effects of the grid 1 resistor are dropping out to give full grid 1 authority briefly near cut-off. Grid1 has much higher gm than grid2, so it takes over near cut-off.
Not sure yet what exactly the crossover gm sum looks like then, but there is no doubt some optimum biasing for least gm variation across the crossover. Outside the crossover region, the class B regions are constant gm. This is about as close to ideal as one can hope for, for P-P operation. No V shaped gm sum here. So no odd harmonics either.
OH, and triodes in P-P, being near linear law, could badly gm double in P-P. Fortunately triode internal N Fdbk is voltage derived, so will smooth the transition over for class AB.
.. Have to go outside and suck up some more wind, wheww...
Looking at the typical S shaped tube gm curve, at low current (left side), the gm is curving upward with further V drive. This portion is operating with greater than square law, and when P-P summed with a similar upward curving complementary device produces a gm sum that sags in the center but curves upward away from the center.
On the right side of the S curve, gm is bending over (downward from a linear ramp at least) from cathode saturation. This is operation with less than square law. Two of these curved sections in P-P will produce a bumped gm sum. And two completely linear devices (1.0 law) with constant gm will sum to a rectangular bump, or doubled gm in the overlap region.
Fortunately tube grid1 operation mostly centers around square law (central region of the S curve). Grid2 operation is typically nearer to 3/2 power law.
One would have to under bias (extreme idle current) class A to get the bending over parts of gm to overlap and cause gm "doubling". Class A, correctly arranged, just has the two S curves overlapped to null out gm sum variation everywhere. (ends of the S parts compensating each other)
Bipolar transistors have linear gm when current mode operated. Which most VAS stages are. And so would badly gm double in crossover if left alone. Only by limiting the Voltage offset between the N and P type follower's Bases to the threshold turn-on region can the gm bump be reduced within reason.
Mosfets start out conduction with a small region of curved over gm before becoming square law, so they also have a gm bump at low current crossover.
Higher idle current increases the width (but not height) of the bump, once formed.
Tubes pretty much always have curving up gm near turn on via grid 1 drive, so they can't gm double, but will always gm droop in the center with low idle current. Higher idle current will overlap the regions of square law. (the linear ramping gm regions) Those will produce a flat gm sum section. So sufficient idle current will give a broad constant gm region (in the class A region). Surrounded by the V shaped increasing gm of the class B regions up until saturation. Causing 3rd and some other odd harmonic distortions with sufficient signal level above the class A region.
Grid 2 (3/2 power law) drive however could theoretically produce some gm humping up at cross over. But fortunately grid 2 also starts out curving upward near turn-on, then transitions to near 3/2 power law through most of its range. Since it passes through square law briefly in that transition region, it has an optimum bias point to overlap the two square law sections to give constant gm across some crossover region width without humping up. But then outside that cross-over region it has the usual V shaped gm sum, but a curving over/outward type V rather than a straight ramp sided V. So somewhat different type odd harmonics above low level than grid 1 drive.
Crazy or Twin Drive has constant effective gm throughout most of the individual tube operating region. This would cause gm doubling at crossover if it were not for the tube reverting briefly to grid1 only operation effectively near cut-off. This is because grid 2 gm is dropping toward zero while the current limiting effects of the grid 1 resistor are dropping out to give full grid 1 authority briefly near cut-off. Grid1 has much higher gm than grid2, so it takes over near cut-off.
Not sure yet what exactly the crossover gm sum looks like then, but there is no doubt some optimum biasing for least gm variation across the crossover. Outside the crossover region, the class B regions are constant gm. This is about as close to ideal as one can hope for, for P-P operation. No V shaped gm sum here. So no odd harmonics either.
OH, and triodes in P-P, being near linear law, could badly gm double in P-P. Fortunately triode internal N Fdbk is voltage derived, so will smooth the transition over for class AB.
.. Have to go outside and suck up some more wind, wheww...
Last edited:
- Status
- Not open for further replies.