I red again (I do that time to time) old articles about amplifiers.
Ed Cherry writes EW July 1997 "Ironing out distortion" about the miller compensation. He recommended to include the power stage into miller loop. This should lower the cross over distortion.
However D. Self wrote in EW Feb 1994 that this kind of solution will lead easily in to parasitic oscillation and he also wrote that the benefit is minimal if the bias in the power stage is set right.
No doubt that both gentlemen have their point. However the idea was so interesting that I started to figure how it could work with Fet power stage. The Fets generate more distortion, and the cheap "hexfets" or similar has big differences (p vs. n) in conductance over the cross over region.
The Fets are much faster than bipolar which could make them better in the millering what Ed Cherry recommends.
Further... what about if the miller compensation is cut in to parts. One including the power stage and one part as the traditional including only Vas. It could solve the oscillation problem but still probably lower the Fet cross over region distortion and to make open loop distortion lower on higher freqs.
Anyone tested or any opinions ?
Ed Cherry writes EW July 1997 "Ironing out distortion" about the miller compensation. He recommended to include the power stage into miller loop. This should lower the cross over distortion.
However D. Self wrote in EW Feb 1994 that this kind of solution will lead easily in to parasitic oscillation and he also wrote that the benefit is minimal if the bias in the power stage is set right.
No doubt that both gentlemen have their point. However the idea was so interesting that I started to figure how it could work with Fet power stage. The Fets generate more distortion, and the cheap "hexfets" or similar has big differences (p vs. n) in conductance over the cross over region.
The Fets are much faster than bipolar which could make them better in the millering what Ed Cherry recommends.
Further... what about if the miller compensation is cut in to parts. One including the power stage and one part as the traditional including only Vas. It could solve the oscillation problem but still probably lower the Fet cross over region distortion and to make open loop distortion lower on higher freqs.
Anyone tested or any opinions ?
In a standard amplifier all the voltage gain comes from the VAS.
Cdom is usually connected between the VAS output and input.
The Miller effect makes this capacitor appear to be much larger
and it forms the dominant pole compensation of the amplifier.
Note that the Miller effect only takes place when the loop
has voltage gain, the more gain, more Miller effect.
Cdom also forms a local feedback loop around the VAS, as
it reduces gain it also reduces the output impedance of the
VAS improving its capability to drive the output stage.
However a possibility is to connect Cdom between the VAS
input and the output of the amplifier, including the output
stage within the "Miller Loop", the loop that is subject to
the feedback through Cdom.
sreten.
Cdom is usually connected between the VAS output and input.
The Miller effect makes this capacitor appear to be much larger
and it forms the dominant pole compensation of the amplifier.
Note that the Miller effect only takes place when the loop
has voltage gain, the more gain, more Miller effect.
Cdom also forms a local feedback loop around the VAS, as
it reduces gain it also reduces the output impedance of the
VAS improving its capability to drive the output stage.
However a possibility is to connect Cdom between the VAS
input and the output of the amplifier, including the output
stage within the "Miller Loop", the loop that is subject to
the feedback through Cdom.
sreten.
Yes, that's Miller's theorem.sreten said:
I know this one. But that's the 1st time I read about a Miller Loop
sreten said:
However a possibility is to connect Cdom between the VAS
input and the output of the amplifier, including the output
stage within the "Miller Loop", the loop that is subject to
the feedback through Cdom.
how?
In a real amplifier Cdom is a real capacitor that takes Cbc of
the VAS transistor out of the equation, usually 47pF to 200pF.
You have a point, if Cdom is connected to the output, then Cbc
of the VAS transistor would have some effect, and not particularly
predictable either.
Perhaps its this that causes " intractable HF oscillation with
Inclusive Miller-compensation " according to D.Self.
Perhaps you'd need to cascode the VAS (to prevent local Miller
effect) for it to have a chance of working without oscillation.
sreten.
the VAS transistor out of the equation, usually 47pF to 200pF.
You have a point, if Cdom is connected to the output, then Cbc
of the VAS transistor would have some effect, and not particularly
predictable either.
Perhaps its this that causes " intractable HF oscillation with
Inclusive Miller-compensation " according to D.Self.
Perhaps you'd need to cascode the VAS (to prevent local Miller
effect) for it to have a chance of working without oscillation.
sreten.
This topic becomes more and more interesting
I'd like to begin in designing my own amplifiers, and I'm currently looking for informations about this.
Tell me if I'm off topic, so I'll maybe start a dedicated topic.
-what's the benefict of adding a Miller compensation (no matter which one (local or loop)) ?
-And what's the advantage of a Miller compensation, compared to a "global" compensation?
By global, I mean placing a cap across the feedback resistor (from output to input). In this scheme, the total speed of the amp is reduced, but all stages are running locally at "full speed"
-"In a standard amplifier all the voltage gain comes from the VAS."
Nothing from the differential?
I'd like to begin in designing my own amplifiers, and I'm currently looking for informations about this.
Tell me if I'm off topic, so I'll maybe start a dedicated topic.
-what's the benefict of adding a Miller compensation (no matter which one (local or loop)) ?
-And what's the advantage of a Miller compensation, compared to a "global" compensation?
By global, I mean placing a cap across the feedback resistor (from output to input). In this scheme, the total speed of the amp is reduced, but all stages are running locally at "full speed"
-"In a standard amplifier all the voltage gain comes from the VAS."
Nothing from the differential?
Sreten, the idea in the consept Ed Cherry presented in his article is not to have miller capacitor only over the power stage but over Vas and power stage.
You write that the miller feedback loop lowers the output impedance of the Vas which is right, however that is not so important in case Fet power stage (IMO) because of two reasons;
1) the Fets has high input impedance so they can deal with higher driving impedance as well.
2) the Fets are not linear (not even close to bibolar) so they need feedback loop more than bipolar.
You write that the miller feedback loop lowers the output impedance of the Vas which is right, however that is not so important in case Fet power stage (IMO) because of two reasons;
1) the Fets has high input impedance so they can deal with higher driving impedance as well.
2) the Fets are not linear (not even close to bibolar) so they need feedback loop more than bipolar.
Hmmmm.....
I've been thinking about this more and can now see its possible
to have two compensation capacitors, one the normal value to
the output and another smaller value to the output of the VAS,
the ratio determining the amount of loop feedback applied to
the output stage.
how well it would work I've no idea.
sreten.
I've been thinking about this more and can now see its possible
to have two compensation capacitors, one the normal value to
the output and another smaller value to the output of the VAS,
the ratio determining the amount of loop feedback applied to
the output stage.
how well it would work I've no idea.
sreten.I think Electron was talking about connecting the Miller capacitor to the output instead of the VAS collector and thereby including the output stage. That's what Cherry proposed. He wouldn't make any Miller loop around the output stage alone.
Electron's second proposal was to split the Miller C into two parts in that one is connected the usual way from VAS collecor to it's base and the second one form the output to the VAS base.
This could indeed give another possibility for trading performance against stability.
Regards
Charles
Electron's second proposal was to split the Miller C into two parts in that one is connected the usual way from VAS collecor to it's base and the second one form the output to the VAS base.
This could indeed give another possibility for trading performance against stability.
Regards
Charles
phase_accurate said:
Hi Charles,
This indeed is virtually mandatory for a BJT power stage.....
Nevertheless, in practice i have found that the split needs to be less than 1/2 to the output stage, with the rest to the TIS collector,...if local stability within the loop is to be gauranteed...
Unfortunately, THD now becomes virtually indistinguishable from TIS collector-refered compensation.......
A MOSFET output stage on the other hand does not suffer from the limited bandwidth of BJT power devices....and can be enclosed by a single cap. , subject to the usuall precautions...eg gate stopper resistors...etc...
This topic goes to the core of successful audio amplifier design and is glossed over with sweeping gestures again and again.....
Stability is difficult to predict, and not well handled by simulations. The best option, particularly given the diversity of speaker loads in the wild, is to experiment empirically without the dubious advantage of PSpice....... Flying line of sight is passe, however, for those committed to reading their instruments......
I cannot see the virtue in a Miller loop Cdom regimes, because as Sreten comments, the Cdom is only effective in a voltage gain mode, the purpose of which is to bring open loop gain down below unity by the pole frequency, so that at the point where NFB becomes PFB, the open loop gain is insufficient to sustain destructive oscillation.
By incorporating the output stage into the loop, we succeed only in adding the phase shift of the outputs to the mix, very likely further destabilising the amplifier. However, it might be possible to make the amplifier more tolerant of capacitive loads, but I suspect the Cdom would need to be smaller for stability. Until this is tested empirically, it is impossible to comment on the sonic effects.
The usual way to achieve stability with capacitive loads, however, is to add a smaller capacitor from the collector of the VAS to the feedback node. This works well, and is extremely useful to improved stability into real world loads, particularly electrostatic speakers.
Cheers,
Hugh
Stability is difficult to predict, and not well handled by simulations. The best option, particularly given the diversity of speaker loads in the wild, is to experiment empirically without the dubious advantage of PSpice....... Flying line of sight is passe, however, for those committed to reading their instruments......
I cannot see the virtue in a Miller loop Cdom regimes, because as Sreten comments, the Cdom is only effective in a voltage gain mode, the purpose of which is to bring open loop gain down below unity by the pole frequency, so that at the point where NFB becomes PFB, the open loop gain is insufficient to sustain destructive oscillation.
By incorporating the output stage into the loop, we succeed only in adding the phase shift of the outputs to the mix, very likely further destabilising the amplifier. However, it might be possible to make the amplifier more tolerant of capacitive loads, but I suspect the Cdom would need to be smaller for stability. Until this is tested empirically, it is impossible to comment on the sonic effects.
The usual way to achieve stability with capacitive loads, however, is to add a smaller capacitor from the collector of the VAS to the feedback node. This works well, and is extremely useful to improved stability into real world loads, particularly electrostatic speakers.
Cheers,
Hugh
Cdom compromise
How about this as a way to go halfway? Q5 and Q6 are the bottom half of a symmetrical cascoded VAS stage. Cdom picks off from a point which is proportional to the output by making the cascode bias variable level, but should still have HF components direct from the VAS drive too. Could use two pole compensation too.
How about this as a way to go halfway? Q5 and Q6 are the bottom half of a symmetrical cascoded VAS stage. Cdom picks off from a point which is proportional to the output by making the cascode bias variable level, but should still have HF components direct from the VAS drive too. Could use two pole compensation too.
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.