From the naming of the thing, and taking into account the earlier posts of a.o Syn08, one would conclude that a miller compensation is a compensation technique relying on the miller effect. That is, the 'miller cap' is hung between two circuit points with opposing signal phases, and a non-zero source impedance.
Whether there are 1 or x transistors in between those circuit points doesn't appear to be significant.
And, whether it results in a dominant pole depends on the component values. One could use miller compensation to create a dominant pole, but that doesn't mean that by definition miller comp = dominant pole compensation.
Jan Didden
Whether there are 1 or x transistors in between those circuit points doesn't appear to be significant.
And, whether it results in a dominant pole depends on the component values. One could use miller compensation to create a dominant pole, but that doesn't mean that by definition miller comp = dominant pole compensation.
Jan Didden
janneman said:
Yup, that's it. Bob's compensation is still using the Miller effect. However, it's not necessary a dominant pole compensation. Still call it Miller compensation if you like, it's only semantics.
It's interesting that the lag compensation as described by Self and others is in fact not essentially different from the so-called "Miller compensation" above. The capacitor from the VAS output to the ground can be viewed as the effect of loading the base amp by a (smaller) Miller cap. That's because a parallel-parallel feedback network loads the output of the base amp with a reactance of gain*feedback reactance, while the input is loaded by the feedback reactance only. From this perspective, lag compensation is about equivalent to some Miller compensation (if we can neglect the non-ideal base amp impact).
Although essentially incorrect, I think "Miller compensation" was though mostly used to identify a Cdom type schema.
: the one describing or using the miller effect.
Miller compensation
Hi Jan,
I agree with that. However, in case of Bob's EC amp for example, a duality will arise. Calling it Miller compensation implies it's lag compensation. But, as I have mentioned before, the signal at the VAS output is (almost) identical to the OPS output (due to HEC). IOW, the compensation cap sees a voltage that is equal to the output voltage. If this cap was directly tied to the output (which should theoretically make no difference), than it is suddenly lead compensation. Funny isn't it?
Cheers,
Edmond.
janneman said:
Hi Jan,
I agree with that. However, in case of Bob's EC amp for example, a duality will arise. Calling it Miller compensation implies it's lag compensation. But, as I have mentioned before, the signal at the VAS output is (almost) identical to the OPS output (due to HEC). IOW, the compensation cap sees a voltage that is equal to the output voltage. If this cap was directly tied to the output (which should theoretically make no difference), than it is suddenly lead compensation. Funny isn't it?
Cheers,
Edmond.
Re: Re: Miller compensation
Therefore I said "theoretically". In practice, of course it makes a difference. So what? You missed my point (as usual). I'm trying to find out why people like Mike K.S. have called it lead compensation.
PS: You really like splitting hairs, don't you?
syn08 said:BZZZT...wrong guess
The output impedances of the VAS and the OPS are significantly different. Therefore, the impact of a Miller compensation cap is different. Your statement would hold for ideal base amp stages, which is not the case.
Therefore I said "theoretically". In practice, of course it makes a difference. So what? You missed my point (as usual). I'm trying to find out why people like Mike K.S. have called it lead compensation.
PS: You really like splitting hairs, don't you?
jacco vermeulen said:
Jacco,
Did you know John Miller pioneered the combined use of pos and neg feedback (in tube amps)?
"Combining postitive and negative feedback", John M Miller, Electronics, March 1950.
http://www.linearaudio.nl/Documents/miller combined fb.pdf
Jan Didden
Charles Hansen said:
Hi Charles,
I agree with much of what you say here. Simulation and measurement give us insight and help us to get there, but how the amplifier sounds is of ultimate importance. Those of us who simulate and measure do also listen a lot. Moreover I, and probably quite a few others, agree that there is much we do not know and understand in audio. That is why I always allow for what I call the "X" factor.
I think that one area that demonstrates what we don't know is passive components. If you take three identical amplifiers with, say, a 5 uF input coupling capacitor, and equip each one with a different type of capacitor, many would say that there will be an audible difference. Assume that all three capacitors are of course carefully matched in value to within, say, 1%.
Equip the first amplifier with a $2 mylar capacitor. Equip the second amplifier with a $5 polypropylene capacitor. Equip the third amplifier with a $30 boutique polypropylene capacitor. I don't know of any external measurement that one can make on those three amplifiers that will reveal which capacitor is in which amplifier. Yes, you will be able to see a difference in DA between the mylar capacitor and the polys, but this is a direct measurement of the capacitor, and the difference in DA does not necessarily show up in a black box measurement of the amplifier, whether it be a linear or non-linear measurement.
BTW, although DA and ESR are often cited as sonic influencers in capacitors, I have measured carefully these parameters in $5 and $30 polypropylenes and seen no difference. There is something else going on if those capacitors sound different. I don't know what it is.
Cheers,
Bob
Re: Miller compensation
Hi Edmond,
These are very interesting points. I have never thought of it as lead compensation because it does not enclose the output stage, but I can see how one would think of it as lead compensation. I think of it as input Miller compensation.
Conventional input compensation is when you just put a pole-zero series RC network across the bases of the input differential pair. It is lag compensation that has the advantage of not degrading slew rate. However, such traditional input compensation can degrade S/N and maybe some other things. The input Miller compensation I use in my amplifier is like input compensation that is achieved by negative feedback, so that it does not have the same disadvantages as ordinary input compensation.
It is easy to see that it is Miller compensation, because a Miller integrator is formed in the forward path of the amplifier (obviously, a Miller integrator need not be limited to feedback around a single transistor, as we build Miller integrators around op amps all the time). Just as with an op amp, however, the multi-stage amplifier around which the Miller capacitor is placed must be compensated. The good news is that this fast group of small-signal transistors can be very lightly compensated to a closed loop bandwidth often exceeding 20 MHz. The final nice thing is that the input stage itself has now been wrapped in the NFB of the input Miller compensation.
I had never seen this type of compensation applied to an audio amplifier before, but I'm guessing that someone else must have come up with it before me. If you know of a reference I'd be interested to see it.
Cheers,
Bob
Edmond Stuart said:
Hi Edmond,
These are very interesting points. I have never thought of it as lead compensation because it does not enclose the output stage, but I can see how one would think of it as lead compensation. I think of it as input Miller compensation.
Conventional input compensation is when you just put a pole-zero series RC network across the bases of the input differential pair. It is lag compensation that has the advantage of not degrading slew rate. However, such traditional input compensation can degrade S/N and maybe some other things. The input Miller compensation I use in my amplifier is like input compensation that is achieved by negative feedback, so that it does not have the same disadvantages as ordinary input compensation.
It is easy to see that it is Miller compensation, because a Miller integrator is formed in the forward path of the amplifier (obviously, a Miller integrator need not be limited to feedback around a single transistor, as we build Miller integrators around op amps all the time). Just as with an op amp, however, the multi-stage amplifier around which the Miller capacitor is placed must be compensated. The good news is that this fast group of small-signal transistors can be very lightly compensated to a closed loop bandwidth often exceeding 20 MHz. The final nice thing is that the input stage itself has now been wrapped in the NFB of the input Miller compensation.
I had never seen this type of compensation applied to an audio amplifier before, but I'm guessing that someone else must have come up with it before me. If you know of a reference I'd be interested to see it.
Cheers,
Bob
Re: Re: Miller compensation
One caveat here. You can use my compensation scheme even if you take the capacitor from the first emitter follower following the VAS, rather than from the high-impedance VAS output node. Indeed, sometimes I do this so as not to load the VAS with the compensation network. For example, in my scheme, you can replace the compensation capacitor with a two-capacitor "T" network, obtaining "T" compensation (what we called it at Bell Labs in the 1970's) or what others call two pole compensation. Taking the compensation from the first EF then eliminates the loading effect of the T network on the VAS.
TMC can also be implemented with my compensation scheme.
Cheers,
Bob
syn08 said:
BZZZT...wrong guess
The output impedances of the VAS and the OPS are significantly different. Therefore, the impact of a Miller compensation cap is different. Your statement would hold for ideal base amp stages, which is not the case.
One caveat here. You can use my compensation scheme even if you take the capacitor from the first emitter follower following the VAS, rather than from the high-impedance VAS output node. Indeed, sometimes I do this so as not to load the VAS with the compensation network. For example, in my scheme, you can replace the compensation capacitor with a two-capacitor "T" network, obtaining "T" compensation (what we called it at Bell Labs in the 1970's) or what others call two pole compensation. Taking the compensation from the first EF then eliminates the loading effect of the T network on the VAS.
TMC can also be implemented with my compensation scheme.
Cheers,
Bob
Re: Re: Re: Miller compensation
Regardless of the type of compensation scheme used, shifting the compensation from the VAS output to the output of the first emitter follower is not necessarily a recipe of lower HF distortion.
Removing the compensation capacitor from the VAS output gives up the low HF VAS output impedance afforded by the compensation feedback loop and turns the VAS into a current source.
The non-linear input capacitance of the first emitter follower then factors into the overall distortion equation to a much greater degree and (depending on the properties of the first emitter follower) the HF THD may very well be higher instead.
Bob Cordell said:
Regardless of the type of compensation scheme used, shifting the compensation from the VAS output to the output of the first emitter follower is not necessarily a recipe of lower HF distortion.
Removing the compensation capacitor from the VAS output gives up the low HF VAS output impedance afforded by the compensation feedback loop and turns the VAS into a current source.
The non-linear input capacitance of the first emitter follower then factors into the overall distortion equation to a much greater degree and (depending on the properties of the first emitter follower) the HF THD may very well be higher instead.
Re: Re: Re: Re: Miller compensation
Hi Glen,
Yes and No. Yes, because it does turn the VAS into a current source, with all its consequences.
No, because you have forgotten one thing: If the emitter follower is also part of the Miller loop, the gain of that loop will be much higher, which in turn will decrease the distortion.
If we go one step further, i.e. tying the comp. cap. directly to the output, as Cherry did (and pray the amp don't start oscillating), the distortion will be even much lower (about one order).
Cheers,
Edmond.
G.Kleinschmidt said:
Hi Glen,
Yes and No. Yes, because it does turn the VAS into a current source, with all its consequences.
No, because you have forgotten one thing: If the emitter follower is also part of the Miller loop, the gain of that loop will be much higher, which in turn will decrease the distortion.
If we go one step further, i.e. tying the comp. cap. directly to the output, as Cherry did (and pray the amp don't start oscillating), the distortion will be even much lower (about one order).
Cheers,
Edmond.
Re: Re: Miller compensation
Hi Bob,
Of course it's lag compensation, but this guys has a different opinion (as usual):
http://www.diyaudio.com/forums/showthread.php?postid=1122186#post1122186
Me too.
I also have never seen it before. The first time when I saw your schematic, I realized that this would probably be the best of all possible compensation schemes.
Cheers,
Edmond.
Bob Cordell said:
Hi Bob,
Of course it's lag compensation, but this guys has a different opinion (as usual):
http://www.diyaudio.com/forums/showthread.php?postid=1122186#post1122186
Me too.
I also have never seen it before. The first time when I saw your schematic, I realized that this would probably be the best of all possible compensation schemes.
Cheers,
Edmond.
Re: Re: Re: Re: Re: Miller compensation
Hi Edmond
Yes, but if the consequences of turning the VAS into a current source are great enough, the extra gain of the miller loop afforded by the emitter follower may still not make up for the increased (can be huge) HF distortion caused by (mostly) the non-linear input capacitance of the first EF.
I do remember you, a while ago, confirming by email that you have also found that shifting the comp. cap from the VAS isn't always beneficial and can actually raise the HF THD in some instances, after I told you about the problem via email.
Cheers,
Glen
Edmond Stuart said:
Hi Edmond
Yes, but if the consequences of turning the VAS into a current source are great enough, the extra gain of the miller loop afforded by the emitter follower may still not make up for the increased (can be huge) HF distortion caused by (mostly) the non-linear input capacitance of the first EF.
I do remember you, a while ago, confirming by email that you have also found that shifting the comp. cap from the VAS isn't always beneficial and can actually raise the HF THD in some instances, after I told you about the problem via email.
Cheers,
Glen