What he meant was the following:
Self took EF output stage configurations with emitter resistors of 0.47, 0.33, 0.22 and 0.1 Ohms. For each configuration, he determined the optimum bias to minimize crossover distortion. He recorded the voltage drop across the emitter resistors and the collector current. His results showed that the voltage drop in the optimum condition varied from 42.6 mV to 52.8 mV for the above range of resistors. OTOH, the current varied from 59 mA to 215 mA. Clearly, the voltage is varying by only a small amount, while the current varies much more. That's why he reached the conclusion that the voltage was the more important quantity.
There's also an analysis by Leach that parallels the HP article that John Curl quoted in this regard a while back. It is at http://users.ece.gatech.edu/~mleach/papers/classab.pdf
Edit: Changed several things due to a bit of confusion on my part
Self took EF output stage configurations with emitter resistors of 0.47, 0.33, 0.22 and 0.1 Ohms. For each configuration, he determined the optimum bias to minimize crossover distortion. He recorded the voltage drop across the emitter resistors and the collector current. His results showed that the voltage drop in the optimum condition varied from 42.6 mV to 52.8 mV for the above range of resistors. OTOH, the current varied from 59 mA to 215 mA. Clearly, the voltage is varying by only a small amount, while the current varies much more. That's why he reached the conclusion that the voltage was the more important quantity.
There's also an analysis by Leach that parallels the HP article that John Curl quoted in this regard a while back. It is at http://users.ece.gatech.edu/~mleach/papers/classab.pdf
Edit: Changed several things due to a bit of confusion on my part
FWIW-About a year ago I wrote Leach and Self about the issue of gm-doubling, because they both argued opposite sides of the issue. I tried to stir up a conversation between them, but didn't get anywhere. I hesitate to publish what he wrote in a private letter, but if it could help further the understanding of the issue, here is what Self said about Leach's paper referenced by Andy C:
"I've only read the paper briefly, (I hope to give it a closer look tomorrow) but I think the problem is nomenclature. Leach, like a lot of people calls "Class-AB" what I would call "optimal Class-B". Therefore he has no name to use for my AB which is Class-B but with the bias turned up to give a Class-A region.
He demonstrates that gm-doubling does not exist with optimal biasing- but I'm not sure anyone ever said it did. I certainly didn't. Turn up the bias and it's there all right. I have seen it too often to have any doubts.
His practice of equating small-signal voltage gain with "normalised transconductance" worries me, because they are quite different things with different dimensions."
Please note that these comments were made from a cusory inspection of Leach's paper, and should not be held against him if invalid. Hopefully, though, it can spur further inspection and understanding as to why there is an apparent disagreement, here.
"I've only read the paper briefly, (I hope to give it a closer look tomorrow) but I think the problem is nomenclature. Leach, like a lot of people calls "Class-AB" what I would call "optimal Class-B". Therefore he has no name to use for my AB which is Class-B but with the bias turned up to give a Class-A region.
He demonstrates that gm-doubling does not exist with optimal biasing- but I'm not sure anyone ever said it did. I certainly didn't. Turn up the bias and it's there all right. I have seen it too often to have any doubts.
His practice of equating small-signal voltage gain with "normalised transconductance" worries me, because they are quite different things with different dimensions."
Please note that these comments were made from a cusory inspection of Leach's paper, and should not be held against him if invalid. Hopefully, though, it can spur further inspection and understanding as to why there is an apparent disagreement, here.
Interesting. And here I thought the discrepancy was that some people were referring to the theoretical transconductance into a short-circuit load while Leach was referring to the transconductance in the presence of a load.
I think Self has created a lot of confusion by referring to what most people call class AB as class B. He's talking about output stage bias currents greater than 200 mA, so for a low enough output power, the conduction will approach a full cycle, not a half-cycle as class B is generally thought of.
I think Self has created a lot of confusion by referring to what most people call class AB as class B. He's talking about output stage bias currents greater than 200 mA, so for a low enough output power, the conduction will approach a full cycle, not a half-cycle as class B is generally thought of.
I agree. Self's nomenclature seems to make no sense. It seems he should be calling his optimum class B "optimum class-AB" if biasing is above pure B.
Be that as it may, another issue is whether or not higher biasing is better for a given situation. For example, if you bias higher in order to get, say, a watt into class A, and your horn-loaded loudspeakers coast at a few hundred milliwatts at a fairly loud listening level, and maybe up to a watt at really, really loud levels, what's the point of "optimum" biasing?
Be that as it may, another issue is whether or not higher biasing is better for a given situation. For example, if you bias higher in order to get, say, a watt into class A, and your horn-loaded loudspeakers coast at a few hundred milliwatts at a fairly loud listening level, and maybe up to a watt at really, really loud levels, what's the point of "optimum" biasing?
quote:
Note: quiescent current doesn't matter. Vq is the vital quantity.
unquote
Workhorse/Kanwar said:
Since Vq inherently depends upon Iq therefore its independent effective role cannot justifies the statement.......
Kanwar,
I think we ALL understand that there is a relationship between Iq and Vq, one causes the other. Anyone understanding the subject woiuld easily capture from the context that Vq is the determining factor, to be set, for minimum non-linearity. From that, the Iq follows automagically.
Are we now deliberately ignoring semantics in an attempt to score a point?
Jan Didden
Note: quiescent current doesn't matter. Vq is the vital quantity.
unquote
Workhorse/Kanwar said:
Since Vq inherently depends upon Iq therefore its independent effective role cannot justifies the statement.......
Kanwar,
I think we ALL understand that there is a relationship between Iq and Vq, one causes the other. Anyone understanding the subject woiuld easily capture from the context that Vq is the determining factor, to be set, for minimum non-linearity. From that, the Iq follows automagically.
Are we now deliberately ignoring semantics in an attempt to score a point?
Jan Didden
I meant drop on Re which is pretty 'natural' phenomenon. But feedback loop corrects it increasing damping factor. I think it is bad, because the feedack corrects something which isn't so essential to correct and has nothing to do with XOdist. Just drop on higher Re is higher when output currect is an ampere or two.
regards
regards
darkfenriz said:
Hi,
I'm sure I follow you. One would use optimum bias to try to linearize the output stage. Even with optimum bias there will still be some xover distortion. It is impossible to get it down to zero, only to minimise it. So if you use feedback, you can still lower it further. I'm not saying it is necessary, that depends on your own choices. But I don't think you can say, ahh, I have optimum bias, THEREFORE I don't need (global) feedback.
Optimum bias and feedback are two tools the engineer has in his toolkit to make a linear amp. Which one (or both) he uses, or even other ones, that's up to him.
Jan Didden
Some sense at last
"Optimum bias and feedback are two tools the engineer has in his toolkit to make a linear amp. Which one (or both) he uses, or even other ones, that's up to him."
A good amplifier design only requires reasonable output stage transconductance to proceed with global NFB correction of all the inherent non-linearities within it's loop! Any glaringly obtuse characteristics are corrected by initial design. Discontinuities are poor design.
"Optimum bias and feedback are two tools the engineer has in his toolkit to make a linear amp. Which one (or both) he uses, or even other ones, that's up to him."
A good amplifier design only requires reasonable output stage transconductance to proceed with global NFB correction of all the inherent non-linearities within it's loop! Any glaringly obtuse characteristics are corrected by initial design. Discontinuities are poor design.
Completely true. The output stage transconductance variations are not the main problem for the nonlinearities in the output stage. A BJT stage loads the VAS and the nonlinearities of this loading are magnitudes larger than the transconductans deviations. As a result underbiasing causes crossover distortion because of discontinuities and overbiasing...
... nearly nothig!
Theoretical and idle current of Id = (Vt / Re*) is ideal,
where Re* = Re + Re'
Re is the emitter resistance of the BJT power stage and Re is the intrinsic emitter of the output devices.
But theoretical values are derived from much too simple BJT models.
amplifierguru said:
Well, the problem is that if you ask 3 people what they understand by 'reasonable output stage transconductance', you get 4 different answers. If your remaning non-linearity is X, and you apply 20dB feedback, to a first approximation you get X/10 non-linearity. Is that low enough? Again, depends on your choices. There is no absolute answer to this (and that is also why we have so much wildly different types of amplifiers on the market).
I thought this discussion was about how to manipulate the bias setting to improve linearity in certain topologies, without having an absolute goal that is nirwana.
Jan Didden
andy_c said:
I have always followed the same terminology as Self used: that classes are defined by the phase of conduction, not biasing - in Class A, a device is conducting during 360 degrees of the signal. Class B, a device is conducting just 180 degrees and Class C < 180 degrees.
Once you agree with that, it is fairly logical and clear to follow his reasoning.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.