"Okay, the VAS's purpose in life is to amplify the error signal subtracted by the differential. It's job is to attenuate the distorted drive current of the output stage. So when you decrease VAS gain, you increase the error signal. Even if it were twice as linear it now has half the gain, so the result is a doubling of distortion assuming the VAS distorts much less than the output stage, which is usually true."
As Jan said, you cannot look at these things in isolation.
The Miller cap linearizes the TIS, lowers the TIS output impedance at HF, and importantly stabilizes the amplifier.
Sure, there other more advanced comp methods, but don't dis a well tried and simple solution that is underpinned by some solid math.
My e-Amp can be jumpered for MC, TMC and the feedback loop 'wide-banded' for some added experimentation and I can tell you that straight MC sounds remarkably good and measures pretty ok as well.
As Jan said, you cannot look at these things in isolation.
The Miller cap linearizes the TIS, lowers the TIS output impedance at HF, and importantly stabilizes the amplifier.
Sure, there other more advanced comp methods, but don't dis a well tried and simple solution that is underpinned by some solid math.
My e-Amp can be jumpered for MC, TMC and the feedback loop 'wide-banded' for some added experimentation and I can tell you that straight MC sounds remarkably good and measures pretty ok as well.
"Considering the popularity of Doug Self's articles and books, VAS seems like a more accepted term today. Cordell adopted it, so how bad can it be? Maybe it falls into the bag with all the other compromises adopted by the use of economic British designers?"
I think the term VAS crept into use over many years, probably a lot to due with Self. I was looking at some old technical papers (from the 1980's IIRC) and they referred to the 2nd stage as the TIS transimpeadance stage. Bruno Putzeys recently also commented on the subject: the correct engineering term is TIS.
I have recently joined Edmond in calling it the TIS.
I think the term VAS crept into use over many years, probably a lot to due with Self. I was looking at some old technical papers (from the 1980's IIRC) and they referred to the 2nd stage as the TIS transimpeadance stage. Bruno Putzeys recently also commented on the subject: the correct engineering term is TIS.
I have recently joined Edmond in calling it the TIS.
This can't be true. If local feedback at the TIS were a good thing, TMC, TPC, MIC and many other compensations would be redundant. Local feedback at the VAS does NOT reduce distortion, it never has, except for low-OLG amps which rely on OL linearity. We only use local MC because we must for stability. There is no other reason I know of.
What is improved by lower TIS output impedance? Nothing, because the lowered impedance just conducts more distortion from the output stage through the VAS and through the input stage. We are seeking to reject distortion through proper impedance techniques, and a low VAS output impedance is the wrong direction.
Yes, MC stabilizes the amp. Yes, it's tried and true. Of course it has solid math, is there anything else? Solidly, local TIS feedback has no purpose at audio frequencies. That's why TMC and TPC get rid of it at audio frequencies. If what you are saying is true then plain MC would be better than both of them.
I am not arguing anything else but that the "local feedback linearized TIS" is meaningless at audio frequencies.
I have designed and built 4 or 5 amps using RC shuntcomp and no Miller to speak of. I have never had problems with local oscillation, except in one case where I used a collector-driven output stage. RC shunt compensation applied correctly is fine. I think it is more forgiving of supply and trace inductances due to higher PSRR and higher TIS output impedance.
RC shunt compensation shouldn't be compared to MC. It is closer to TPC. I think the two might be equivalent.
I guess that makes sense, if a V/I converter is considered an amplifier. But a resistor is a V/I converter, and by calling it a transimpedance stage we are treating it as a positive impedance.
An inverting transimpedance amplifier would have negative Iin/Vout, so if treating it as a passive component maybe it makes sense to see it backwards. But what about a noninverting TIS?
I see this as a discrepancy.
I dont want to get into a comp method 'slinging match' here keantoken, but will try to answer your points 1 by 1
This can't be true. If local feedback at the TIS were a good thing, TMC, TPC, MIC and many other compensations would be redundant. No, there are often many ways to solve a problem and tradeoffs to make. Just because MIC and TMC may offer certain advantages, does not make MC redundant either. Its a valid compensation method - and not just in audio
Local feedback at the VAS does NOT reduce distortion, it never has, except for low-OLG amps which rely on OL linearity. We only use local MC because we must for stability. There is no other reason I know of.The MC cap linearizes the TIS stage - even more so when the TIS local loop gain (e.g. by using a beta enhancer) is high; if you are talking about overall amplifier distortion, then perhaps . . . but without some form of comp it would likely be unstable.
What is improved by lower TIS output impedance? Nothing, because the lowered impedance just conducts more distortion from the output stage through the VAS and through the input stage. We are seeking to reject distortion through proper impedance techniques, and a low VAS output impedance is the wrong direction.TIS distortion increases with load - local feedback through Cdom mitigates that.
Yes, MC stabilizes the amp. Yes, it's tried and true. Of course it has solid math, is there anything else? Solidly, local TIS feedback has no purpose at audio frequencies. That's why TMC and TPC get rid of it at audio frequencies. If what you are saying is true then plain MC would be better than both of them.I am not saying that. I am saying its a valid comp method and should be seen as one of the tools in an amplifier designers arsenal
I am not arguing anything else but that the "local feedback linearized TIS" is meaningless at audio frequencies.and at 20 kHz?
I have designed and built 4 or 5 amps using RC shuntcomp and no Miller to speak of. I have never had problems with local oscillation, except in one case where I used a collector-driven output stage. RC shunt compensation applied correctly is fine. I think it is more forgiving of supply and trace inductances due to higher PSRR and higher TIS output impedance.I assume here you are talking about loading the TIS output to ground or do you mean across the diff amp collector loads? I've seen both methods used. You sometimes have to comp the MIC loop using a RC network across the LTP loads - see Cordell (chapter 4 IIRC)
RC shunt compensation shouldn't be compared to MC. It is closer to TPC. I think the two might be equivalent. Best way to compare these different comp methods is to do a series of loop gain/phase plots.
This can't be true. If local feedback at the TIS were a good thing, TMC, TPC, MIC and many other compensations would be redundant. No, there are often many ways to solve a problem and tradeoffs to make. Just because MIC and TMC may offer certain advantages, does not make MC redundant either. Its a valid compensation method - and not just in audio
Local feedback at the VAS does NOT reduce distortion, it never has, except for low-OLG amps which rely on OL linearity. We only use local MC because we must for stability. There is no other reason I know of.The MC cap linearizes the TIS stage - even more so when the TIS local loop gain (e.g. by using a beta enhancer) is high; if you are talking about overall amplifier distortion, then perhaps . . . but without some form of comp it would likely be unstable.
What is improved by lower TIS output impedance? Nothing, because the lowered impedance just conducts more distortion from the output stage through the VAS and through the input stage. We are seeking to reject distortion through proper impedance techniques, and a low VAS output impedance is the wrong direction.TIS distortion increases with load - local feedback through Cdom mitigates that.
Yes, MC stabilizes the amp. Yes, it's tried and true. Of course it has solid math, is there anything else? Solidly, local TIS feedback has no purpose at audio frequencies. That's why TMC and TPC get rid of it at audio frequencies. If what you are saying is true then plain MC would be better than both of them.I am not saying that. I am saying its a valid comp method and should be seen as one of the tools in an amplifier designers arsenal
I am not arguing anything else but that the "local feedback linearized TIS" is meaningless at audio frequencies.and at 20 kHz?
I have designed and built 4 or 5 amps using RC shuntcomp and no Miller to speak of. I have never had problems with local oscillation, except in one case where I used a collector-driven output stage. RC shunt compensation applied correctly is fine. I think it is more forgiving of supply and trace inductances due to higher PSRR and higher TIS output impedance.I assume here you are talking about loading the TIS output to ground or do you mean across the diff amp collector loads? I've seen both methods used. You sometimes have to comp the MIC loop using a RC network across the LTP loads - see Cordell (chapter 4 IIRC)
RC shunt compensation shouldn't be compared to MC. It is closer to TPC. I think the two might be equivalent. Best way to compare these different comp methods is to do a series of loop gain/phase plots.
Last edited:
Hi Bonsai,
Do you usually do loop gain measurement? Do you use current-injection approach?
I have used HP3577A with an active differential probe for open-loop. I will try current-injection for loop-gain.
Glad to see this thread alive again with some technical discussion, thanks.
As is often the case for me, I'm not 100% sure I am visualizing the intended circuit or implementation, specifically the above mentioned RC shunt compensation.
Since this is Bob's thread would it be possible to reference a specific figure(s) in his book? I would think that Bob would encourage and benefit from such a discussion.
Keantoken, your electronics insight is amazing for someone any age, but at your age its truly a gift.
Thanks
-Antonio
Hi Guys
In Bob's book, see figures 4-11/12
7-15, 7-16 (front-end of his mosfet amp)
9-4/5 for TPC, BTC compared to miller
9-6 for input comp
9-7 for MIC
9-8 for "inclusive miller" wrapping output stage and VAS
9-9 for TMC
25-15 for a non-global-FB amp with similar front-end to 7-16
Also look at Self's notes about VAS gain reduction using an R in parallel with the miller cap. In his 5th-ed PA book, fig.5-16b and 5-17 graph. The latter shows low-freq gain reduction without altering high-freq performance.
Have fun
Kevin O'Connor
In Bob's book, see figures 4-11/12
7-15, 7-16 (front-end of his mosfet amp)
9-4/5 for TPC, BTC compared to miller
9-6 for input comp
9-7 for MIC
9-8 for "inclusive miller" wrapping output stage and VAS
9-9 for TMC
25-15 for a non-global-FB amp with similar front-end to 7-16
Also look at Self's notes about VAS gain reduction using an R in parallel with the miller cap. In his 5th-ed PA book, fig.5-16b and 5-17 graph. The latter shows low-freq gain reduction without altering high-freq performance.
Have fun
Kevin O'Connor
Magnoman, where is it you read about my age? It was probably a very old post. I'm older now.
I never said MC wasn't a valid compensation technique. It does what it is intended to do. I am only arguing that local feedback MC reducing distortion of the amplifier at audio frequencies is a myth, which is why we use TMC or TPC to cut out local feedback at audio frequencies... And that we already know this. At RF of course, it is still there.
I never said MC wasn't a valid compensation technique. It does what it is intended to do. I am only arguing that local feedback MC reducing distortion of the amplifier at audio frequencies is a myth, which is why we use TMC or TPC to cut out local feedback at audio frequencies... And that we already know this. At RF of course, it is still there.
I never said that I preferred no compensation to MC. What do you think I'm saying? I know that local feedback linearizes the VAS, but it doesn't linearize the amplifier. My apologies if I did not get this across. It might be my mistake, but it sounds strangely as if you are agreeing with me.
Local feedback just decreases gain and increases the load on the LTP. A low impedance VAS sees the distorted Gm distortion of the output stage and conducts it through the VAS and the feedback loop, increasing distortion.
I never argued this point.
Not according to a simulation I just did. But if you can show me a simulation where 20KHz THD is improved by plain MC over TPC or TMC I'll accept your point.
I use RC shunt across the TIS input. I have not used it across the LTP so far, since I'm wary of input impedance effects, and I'd probably have to use plenty of degeneration. I have never used MIC, it usually worsens AC behavior with this kind of compensation.
I just compared RC shunt and TPC in simulation on my last amp design. The result was very similar. TPC lowered distortion by about 1-2db, and caused some troublesome resonance peaks.
Last edited:
Magnoman, where is it you read about my age? It was probably a very old post. I'm older now.
...
I use RC shunt across the TIS input.
....
Keantoken,
Indeed it was an old post (dont remember where) but you were 17 then !!!
Sorry but I dont understand your reference to RC shunt across the TIS input, basically the loss of pole splitting?
For more complex high order compensation, I know its not a problem in the audio spectrum but, should we at least verfy that the effects on settling truly can be ignored (and it might be fun and instructive), similarly with unconditional stability.
Thanks
-Antonio
I'm 19 now, I think. So I guess not that long ago.
RC shuntcomp is improved by significant TIS degeneration, and in this case degeneration doesn't decrease VAS gain at audio. Otherwise, varying VAS Gm will cause wandering poles...
I'm not quite sure what you mean by settling. If you mean energy stored in the RC time constant, then this can be problematic if the frontend is repeatedly saturated, but recovery is usually more than fast enough. For fast amps the RC time constant can be quite small.
I don't get unconditional stability so much. An unconditionally stable amp can still ring like crazy and oscillate into an output cap, based on the "unconditionally stable" amps I've seen. I don't have much faith in unconditional stability.
RC shuntcomp is improved by significant TIS degeneration, and in this case degeneration doesn't decrease VAS gain at audio. Otherwise, varying VAS Gm will cause wandering poles...
I'm not quite sure what you mean by settling. If you mean energy stored in the RC time constant, then this can be problematic if the frontend is repeatedly saturated, but recovery is usually more than fast enough. For fast amps the RC time constant can be quite small.
I don't get unconditional stability so much. An unconditionally stable amp can still ring like crazy and oscillate into an output cap, based on the "unconditionally stable" amps I've seen. I don't have much faith in unconditional stability.
So 17-2006+2012 = 23 years? (still pretty young, btw, I'm almost 70)
Some calculations results will indeed be "age of the capitain" dependent....
Keantoken
Maybe you can post a sketch I suspect I'm not understanding your circuit.
How is it that adding degeneration doesn't decrease gain under 20Khz, unless the next stage has a very high input impedance, and doesn't miller compensation move the stages next pole further away.
By settling with higher order compensation I meant the effects you will see from having multiple rc times, essentially going from hf to lf, which would be evident in step responses (again at very low level, still fast and likely not an audio problem but worth seeing).
I should have said conditional stability, as with higher order compensation there is more chance of having 180 phase shift within the loop cross-over frequency, and the potential for having a gain reduction initiate oscillation such as during start up or approaching clipping.
19, geez, still in college thats a little depressing,
Are you near SanAntonio, if so you should look at working summers at SwRI.
Thanks
-Antonio
Maybe you can post a sketch I suspect I'm not understanding your circuit.
How is it that adding degeneration doesn't decrease gain under 20Khz, unless the next stage has a very high input impedance, and doesn't miller compensation move the stages next pole further away.
By settling with higher order compensation I meant the effects you will see from having multiple rc times, essentially going from hf to lf, which would be evident in step responses (again at very low level, still fast and likely not an audio problem but worth seeing).
I should have said conditional stability, as with higher order compensation there is more chance of having 180 phase shift within the loop cross-over frequency, and the potential for having a gain reduction initiate oscillation such as during start up or approaching clipping.
19, geez, still in college thats a little depressing,
Are you near SanAntonio, if so you should look at working summers at SwRI.
Thanks
-Antonio
Closer to Dallas actually. BTW, feel free to talk to me if there's something I can do for money.
I think you mean gain margin. I have to admit I don't find myself needing OLG graphs much. I can look at differential output current and output impedance and that tells me most of what I need to know. Whenever I've checked gain margin with OLG it was good enough.
There is some overshoot with this type of compensation. Overshoot increases with a smaller capacitor, and you don't necessarily need the smaller capacitor for lower audio distortion. In my latest schematic I increased the cap from 560pF to 10nF, reducing overshoot to 1% and THD changed only by about 4db. I was surprised. So it seems this circuit's limitation at the moment is not the compensation.
Here are AC response, OLG and FFT plots from a "generic-kean" amp I threw together in the simulator. FFT is taken at 33V into 6R (90W). The circuit is a generic Blameless with 3 tricks, one of which is the shunt compensation. It seems my compensation techniques confuse the OLG probe, so there is a phase offset. I built my last amp which used this compensation scheme and aforementioned tricks, so I know it works.
The red trace is with compensation set to 2.5% overshoot, the orange trace is for 20% overshoot.
I think you mean gain margin. I have to admit I don't find myself needing OLG graphs much. I can look at differential output current and output impedance and that tells me most of what I need to know. Whenever I've checked gain margin with OLG it was good enough.
There is some overshoot with this type of compensation. Overshoot increases with a smaller capacitor, and you don't necessarily need the smaller capacitor for lower audio distortion. In my latest schematic I increased the cap from 560pF to 10nF, reducing overshoot to 1% and THD changed only by about 4db. I was surprised. So it seems this circuit's limitation at the moment is not the compensation.
Here are AC response, OLG and FFT plots from a "generic-kean" amp I threw together in the simulator. FFT is taken at 33V into 6R (90W). The circuit is a generic Blameless with 3 tricks, one of which is the shunt compensation. It seems my compensation techniques confuse the OLG probe, so there is a phase offset. I built my last amp which used this compensation scheme and aforementioned tricks, so I know it works.
The red trace is with compensation set to 2.5% overshoot, the orange trace is for 20% overshoot.
