I have (and will) keep working on a class A ccs-loaded Sziklai output stage, and doing so i found that the nonlinear capacitance of the gate worked in my favour as it modified the ammount of compensation, making it higher when the transconductance was higher and lowering it when the transconductance (and so OLG) was lower.
The idea i got from this, and that will seem grotesque to everybody not actually doing the maths, is that a cascoded vas could be made extraordinarily linear by using a varicap diode as compensation capacitor.
It would work the following way:
As the input current grows the transconductance of the VAS grows, so the amount of voltage increase needed at Cdom to produce a certain current increase becomes lower (i'm assuming that this happens after the pole and that the vas is working as an integrator). Assuming that the varicap behaves exponentially, the current needed to produce a voltage increase at the vas would go higher the same way the transconductance does, so the two effects would cancel and the whole system will be closer to a perfect integrator.
As the terms appear in exponential form, a clever cascoding that made the voltage at the varicap change by the correct amount would allow to compensate for the effect squared, and this will fulminate also the distortion due to the exponential transconductance at the input stage, giving "mega-linearity".
Has anyone tryed something like this, or knows a technical reason that discourages it? Please avoid the Cdom must be linear because you want things linear because it's easy to prove on paper that the nonlinear cap makes the overall system more linear.
The idea i got from this, and that will seem grotesque to everybody not actually doing the maths, is that a cascoded vas could be made extraordinarily linear by using a varicap diode as compensation capacitor.
It would work the following way:
As the input current grows the transconductance of the VAS grows, so the amount of voltage increase needed at Cdom to produce a certain current increase becomes lower (i'm assuming that this happens after the pole and that the vas is working as an integrator). Assuming that the varicap behaves exponentially, the current needed to produce a voltage increase at the vas would go higher the same way the transconductance does, so the two effects would cancel and the whole system will be closer to a perfect integrator.
As the terms appear in exponential form, a clever cascoding that made the voltage at the varicap change by the correct amount would allow to compensate for the effect squared, and this will fulminate also the distortion due to the exponential transconductance at the input stage, giving "mega-linearity".
Has anyone tryed something like this, or knows a technical reason that discourages it? Please avoid the Cdom must be linear because you want things linear because it's easy to prove on paper that the nonlinear cap makes the overall system more linear.
Interresting enough.. I think my brain have to process this for a while then I'll go back and see .. hmm actually I got an idea while reading this as it do contain some good info.. :-D thx..
(hmm but spontaneously I'd like to make the problem go away in some other way)
(hmm but spontaneously I'd like to make the problem go away in some other way)
nikwal said:
Error cancelation is refered by Douglas Shelf as attacking the symptoms instead of the problem. While this is true up to a certain point, the transconductance of a bjt is exponential by nature and there is little to do about it. Biasing it high enough to make it look linear is trying to hide the problem, not solving it.
Thanks for your interest and i'm happy my post has somehow been useful.
once you've chosen your amplifying device Dr Ed Cherry has spelled out several times 3 ways to linearize a gain stage:
Increase the bias to signal ratio - use less of the nonlinear gain curve
Cancellation – really only useful with diff pair and push-pull emitter followers, and even order distortions
Negative Feedback – local feedback as in degeneration, emitter/source followers or larger feedback loops including more gain stages
what is being discussed here appears to be a form of cancellation by using a additional nonlinear device, cancellation includes "predistortion" as well as even order cancellations of diff pairs or complementary stages - and only works as well as the nonlinear parts are well known and fully predictable/stable over the operating range and with unmeasured environmental or loading changes
using ideas from Cherry's list isn't cheating since as far as I know they encompass all the ways we know that work
but please feel free to add new concepts to Cherry's list, it would be a considerable service to electronic engineering
Increase the bias to signal ratio - use less of the nonlinear gain curve
Cancellation – really only useful with diff pair and push-pull emitter followers, and even order distortions
Negative Feedback – local feedback as in degeneration, emitter/source followers or larger feedback loops including more gain stages
what is being discussed here appears to be a form of cancellation by using a additional nonlinear device, cancellation includes "predistortion" as well as even order cancellations of diff pairs or complementary stages - and only works as well as the nonlinear parts are well known and fully predictable/stable over the operating range and with unmeasured environmental or loading changes
using ideas from Cherry's list isn't cheating since as far as I know they encompass all the ways we know that work
but please feel free to add new concepts to Cherry's list, it would be a considerable service to electronic engineering
jcx said:
No, not really. Check this: http://perso.orange.fr/francis.audio2/AmpHegglun.doc
I wouldn't build such an amp, though.
OK "emitter" is too specific for you?
we can add source followers to the list of distortion cancellation schemes - or even just "complementary" stages
but complementary stages are more limited in that different carrier mobility and/or N vs P doped Si conductivity make it impossible to make completely complementary devices
Cherry goes on to point out that distortion cancellation schemes usually see a practical limit of <30 dB cancellation while at audio frequencies feedback can achieve many more orders of magnitude improvement
as an example jfet diff pairs with nominal "square law" gm should be the most linear SS stage - until you compare them to BJT diff pairs with emitter degeneration to give the same gm as the fets - then bipolar diff pairs are more linear
we can add source followers to the list of distortion cancellation schemes - or even just "complementary" stages
but complementary stages are more limited in that different carrier mobility and/or N vs P doped Si conductivity make it impossible to make completely complementary devices
Cherry goes on to point out that distortion cancellation schemes usually see a practical limit of <30 dB cancellation while at audio frequencies feedback can achieve many more orders of magnitude improvement
as an example jfet diff pairs with nominal "square law" gm should be the most linear SS stage - until you compare them to BJT diff pairs with emitter degeneration to give the same gm as the fets - then bipolar diff pairs are more linear
I don't know where this falls, but if you vary the tail current of an LTP in just the right relationship to the differential input you can generate a flat region around zero and preserve low noise, which is difficult with Schmook's offset diff-pairs. We have a patent (AD8099). I suppose one could quibble but I don't really want to get into the "is it feedback" argument.
So you want a reverse biased varicap diode? How about zener voltage of a typical varicap?
Regards
Adam
Regards
Adam
Scott,
I haven't see that ADI circuit but as a guess varying tail current modulates gm so this would constitute a form of "gain scheduling" which I would mostly classify as distortion cancellation
in control theory I think gain scheduling is equivalent to a "pre-distortion"/cancellation scheme where you have prior knowledge of the nonlinear device's transfer function/distortion and create a fixed compensating gain schedule ( preferably a smooth curve) - where the gains are switched/varied by another measured variable
I agree that sometimes the influence of the “scheduling” variable may seem to include some aspects of negative feedback, we would have to be very careful about using the same assumptions/definitions/semantics to draw sharp distinctions
nor is there any need to draw such distinctions to have a useful circuit - a useful circuit may well include both pre-distortion/cancellation aspects and negative feedback
I haven't see that ADI circuit but as a guess varying tail current modulates gm so this would constitute a form of "gain scheduling" which I would mostly classify as distortion cancellation
in control theory I think gain scheduling is equivalent to a "pre-distortion"/cancellation scheme where you have prior knowledge of the nonlinear device's transfer function/distortion and create a fixed compensating gain schedule ( preferably a smooth curve) - where the gains are switched/varied by another measured variable
I agree that sometimes the influence of the “scheduling” variable may seem to include some aspects of negative feedback, we would have to be very careful about using the same assumptions/definitions/semantics to draw sharp distinctions
nor is there any need to draw such distinctions to have a useful circuit - a useful circuit may well include both pre-distortion/cancellation aspects and negative feedback
I havn't seen a single reason why this should interfere with negative feedback, the amp will still have nfb and stability should be equal or even better. Compensation has to be made for the worst case transconductance, so a more linear design will be also more stable and leave more margin for feedback.
- Status
- Not open for further replies.