to Miller compensate or not to Miller compensate

Status
Not open for further replies.
This is a post by ppl, quoting AKSA and adding comments of his own. It is taken from an old thead on the Halcro amp:

Originally posted by AKSA
I would like to affirm John's observations, and add a few of my
------------------------------------------------------------------------------------
1. Excessive negative feedback does indeed bring back single tone distortion measurements, but creates a
myriad of high-order, subliminal, and often odd-order artefacts which are highly objectionable to the human
ear. For example (I read this somewhere years ago, but cannot remember the source!): A trumpet played
hard and loud has an additional 0.05% of H5, H7 and H9 over a quietly played trumpet, yet if adjusted for
amplitude and heard from some distance, sounds very different despite a H2/H3/H4 spectral composition
essentially the same. This very clearly draws attention to the spectral distribution of the distortion,
something not given much credence in anything but tube circles.

2. Because of the near infinite impedance presented to a voltage amplifying device by a current source,
tubes and SS, the device is thus able to offer its full voltage amplification. In a tube, this is mu, in a
transistor, it is the ratio of the collector to the emitter impedance, with the influence of beta thrown in, and
is typically 60dB. The stage gain of a transistor with a near-infinite collector load is very different to a finite
load. For a given overall gain, this naturally increases feedback factor, which, beyond a certain point as John
points out, is undesirable for sound quality.

3. A current mirror or source is also very fast, and furthermore makes the gain of the stage extremely
sensitive to impedance changes in the following, driven stage. If the output stage is push pull, the variation
in impedance of this load with signal is quite radical, yet this is rarely discussed in light of the uniformly high
impedance presented by the current source supplying current to the amplifying device. Global negative
feedback is expected to 'fix' this problem, and yet the impedance changes, like a tube grid moving into
positive bias, is quite sharp at the crossover transition.

4. We need to give more attention to the voltage amplifying device itself, since we need to pull its open loop
gain back to below unity at the pole frequency by adding lag compensation across its input/output
(base/collector). Rather more lag compensation is required with a current mirror load; this is because we
must pull the OLG back to below unity by the pole frequency to avoid instability. Because the usual
6dB/octave single pole compensation is contending with more OLG to begin with, this compensation is more
savage than it otherwise might be. Lag compensation is bad because it slows the amplifier and traditionally
the voltage amplifier is the slowest stage of a global nfb amp. The amp must be nimble. If it were possible to
pull back OLG by using a finite load which rapidly increased its loading at higher frequencies for other,
unrelated reasons, then such savage compensation might not be necessary.

5. In closing, I would say that all amplifiers sound worse as one increases the lag compensation. Too much is
leaden and flat; too little is fuzzy - and risky for tweeters - as the amp lapses into short term instability on
transients. The trick is to arrange it so that this compensation is both optimal and minimal - a tall order, but
one with sonic rewards.
------------------------------------------------------------------------------------

Hugh has made some importent observattions regarding Audio Amp Design. I for the most part agree with the
Above statments and totaly agree with respect to Lag Compensation. Years ago when i would do Mod's on
Audio Amps,One of the first thing to get removed or Reduced was the Phase lag Network. this greatly
improved the Midrange and High End by reducing the harshness and glare of the Amp. To maintain Stability I
would then reduce the Openloop gain by eather using Emmiter resistors on the Input stage diff Amp and or
Puting a load resistor on the Output of the Second Vas stage. Sometimes using a higher value of emmiter
resistor than was used on this stage was done to futher reduce the Open Loop gain if needed. The Load
resistor would also somewhat offset the Dramatic impedance changes seen by the Second vas stage from
the Output Stages input Impedance Changes. The result was alot Cleaner and more stable Amp.

When I did my own Amp designs i would then not fall into the trap of using all sorts of Compensation
methods to stabilize the Amp Circuit. Phase Lead compensation is also bad as it slowes the Amp down I like
to have stability come naturaly by using the exsisting capacitence of the devices and selecting the open
loop gain so as the have the Unity gain crossing frequency happen prior to the first pole. This will produce
higher THD numbers and in DC coupled designs require care with DC offset, however correction methods are
available to take care of this.

I also like to have the open loop gain be constant across the Audio bandwidth as this produces a Consistent
THD number with frequency and not have the rising THD vs Frequency typical of Circuits using high open
loop gain and limited open loop bandwidth.

Current sources & Mirrors can be quite usefull if properly used and used in the right places. using a current
source to supply current to the Emmiter's of the Input Diff amp is a good thing as it improves both the
comon mode as well as the Power supply rejection ratio of the Circuit. Current Mirror's used on a folded
cascode Voltage gain stage can also give good results, however on the more conventional cascode or comon
Emmitter stage thay as Hugh pointed out can create problems unless the operating current is set so high so
as to be able to drive the Output stage's non linear impedance at it's worst case.

Using Feedback around just the output stages is not IMHO a good thing as it reduces the Stability and speed
of the output stage and this stage must by nature be alot faster than the Vas stages to avoid instability.
 
AKSA and ppl don't seem to like Miller compensation, but I am really curious about the reasoning behind this is.

I can understand their argument to a degree in that it is probably not a nice thing to have the open loop gain starting to roll off within the audio band. Also, the input stage has to be able to deliver the charging current for the Miller cap, otherwise slew rate limiting will occur.

Assuming that DC open loop gain is limited by emitter degeneration of the input long tailed pair and maybe the VAS, so that the open loop rolloff point can be placed at say 25 kHz, and assuming that it can deliver ample current, would one try to do the frequency compensation on the input stage or by adding a Miller capacitance on the VAS transistor?

I suspect it may be better to go for the Miller compensation, because it will:
a) linearize the parasitic inherent Miller capacitance of the VAS transistor
b) be more efficient than rolling off the input stage gain because of the pole slitting action, i.e. greater overall bandwidth can be achieved
c) reduce distortion caused by the nonlinear input impedance of the output transistors because it lowers the output impedance of the VAS stage because of its feedback action

Comments? Flames?
 
Nothing can be better then Miller compensation

This is a quote from http://www.normankoren.com/audio/

In the original PAS, the feedback loop is stabilized by 33 pF capacitor CLFB in shunt with feedback resistor RLFB. In the present modification, we eliminate CLFB, replacing it with Miller capacitor C3M connected between input stage grid 3G and plate 3P. We also add C3C connected between input stage cathode 3C and ground to shunt RF interference from the output cable. R3GS and C3M provides the dominant pole that controls open-loop rolloff and assures stability. The validity of this technique was confirmed by a recent article in the Journal of the Audio Engineering Society, which used a highly mathematical analysis to determine that .feedback with Miller compensation is a superior approach to error-correcting amplifier design
 
Electronics World Article

Hi all,

in this month's Electronics World (Mar. 2003) John Ellis makes a case for the phase lead, input lag (PLIL) technique as an alternative to Miller capacitor. He uses a number of amplifiers as a test beds including Self's 'Blameless' design. I was wondering if any of the posters had read this?

On a related note he observed low level background oscillation with some designs at around 1MHz and attributed this to a combination of marginal stability and inductance in the output stage emitter resistors. I have noticed a background oscillation of around 500 kHz in a recently completed example of Self's load-invariant amp. design and am about to change the emitter resistors to non-inductive Vishay thick film types. Haven't had a chance yet - they arrived about 15 minutes ago! Has anyone previously noticed this problem or tried this cure?

James
 
The one and only
Joined 2001
Paid Member
You can go round and round on this, but sometimes a
few pF lag is just what you need, and as long as it
isn't excessive, doesn't tend to create problems.

Myself, I find that nice simple amplifiers don't tend to
need lag compensation, but they often like a couple pF
across the feedback resistor to perfect the square wave.
 
"... am about to change the emitter resistors to non-inductive Vishay thick film types"

My prediction: it won't cure the 500kHz noise. :Popworm:

Just a word about stability in general FWIW. Obviously if a circuit is oscillating out of control it won't sound too good and may well cook your speaker, but what about some minor instability? What should the phase margin be? Why does it matter whether an amps response is totally stable or not? Provided the voltage gain is relatively flat to 20kHz does it matter that it, say, doubles at 200kHz? So what if a square-wave has an HF ringing on it? You won't hear it!

Honestly, we cannot hear over 20kHz.

There are indirect reasons why phase margin and compensators may cause audible effects. Figure out what these are and you can make better choices.
 
Correct

Hi Traderbam,

Your prediction was correct. It didn't change the low level oscillations - just moved them in frequency from approx. 500 kHz to around 320 kHz.

I fully realise that this is not audible but its always there - a small sine wave even with the input shorted to ground. I might not be able to hear it but I'm going to fix before I start to use this amp in earnest.

Thanks,

James
 
Ex-Moderator
Joined 2002
Nelson Pass said:
You can go round and round on this, but sometimes a
few pF lag is just what you need, and as long as it
isn't excessive, doesn't tend to create problems.

Myself, I find that nice simple amplifiers don't tend to
need lag compensation, but they often like a couple pF
across the feedback resistor to perfect the square wave.

So, for instance, as you can't get much simpler than a gainclone, ( Sorry Nelson!), would a small cap accross the feedback resistor be advantageous in that application as well?
 
capslock said:

I suspect it may be better to go for the Miller compensation, because it will:
a) linearize the parasitic inherent Miller capacitance of the VAS transistor
b) be more efficient than rolling off the input stage gain because of the pole slitting action, i.e. greater overall bandwidth can be achieved
c) reduce distortion caused by the nonlinear input impedance of the output transistors because it lowers the output impedance of the VAS stage because of its feedback action
Comments? Flames?

Let me ask a question. Will linearizing the miller capacitance of the VAS stage matter if the feedback through the output stage back into the long tailed pair is fast enough to let the long tailed pair do the compensation? If the slowest portion of the amplifier is the long tailed pair, shouldn't many of the non linearities be corrected for? (Assuming that no 'gross' levels of distortion exist in other stages)


In my experience, it is always better to improve the speed of other stages in the amplifier before you try to compensate via added miller capacitance. Increasing bias on some stages can help, depending on circuit configuration. Only add miller capacitance as a last resort. (Again, as Nelson said, nothing excessive) Even then, to prevent phase shifts at higher frequencies it may be worth trying some sort of pole splitting. I have at times found a resistor in series with the added miller cap will compensate the amplifier and still keep phase shifts relativley low at higher frequencies. However at lower levels of gain, these options may not be avaliable.



traderbam said:
What should the phase margin be? Why does it matter whether an amps response is totally stable or not? Provided the voltage gain is relatively flat to 20kHz does it matter that it, say, doubles at 200kHz? So what if a square-wave has an HF ringing on it? You won't hear it!

There are indirect reasons why phase margin and compensators may cause audible effects. Figure out what these are and you can make better choices.

That is a good question, what should be pahse margin be? I guess it would depend on your load. If running a purely resistive load, you could probably get away with a few degrees. However if you start adding any complex loads, things may change. I would say phase margin needed would depend on the load impedance. If your speakers don't cause the amp to oscillate, then you're fine. However, if you don't know what the impedance will be, you may be better to err on the side of caution.

Let me ask another question...
What if the gain is flat to 20Khz and the phase shift was around 180 degrees? Would that affect the sound of the amplifier?

I believe that a bigger factor to sound quality is the phase shift verses frequency, not amplitude vs frequency. Think of an instrument, and shift each of the harmonics. Will the sound still be the same? My guess is no. The pressure waves that will hit your ear will not the same as from an amplifier that has no phase shift.

For example, I have inclosed an image, which shows four traces with identical spectral amplitudes. Each trace has the same fundimental frequency with a second 'harmonic' frequency of 20% of the fundimental frequency, differing only by phase shift. Could the human ear detect the difference? Maybe, maybe not, I have not done any audio testing of these waveforms yet. The point is, that phase shift changing with frequency can affect a complex waveforms amplitude vs. time trace.

-Dan
 

Attachments

  • disto.jpg
    disto.jpg
    50.3 KB · Views: 1,306
AKSA said:
Hi Nemestra,

Try inserting base stoppers on the outputs of 10R and on the drivers of 100R, if they are not already there.

Cheers,

Hugh

It is a good advice..... Like written in another thread, the BJT can act as a "inductor" so adding a base resistor will lower the Q of the "inductor" and remove the tendency to oscillation in the transistor.

ad a give capacitive load this tendency to oscillation will rise. This will give you a oscillating current on the base of up to 1 - 3amp in peak.... This will affect all the way back to the VAS stage..

With mosfets (BUZ900DP/BUZ905DP) in the outputs this current on the gate would be 1/10 the size of the BJT (MJL3281A/MJL1302A).

Sonny
 
James/Hugh/Sonny:
I am not sure that this is a local oscillation of the output transistors because in my experience this occurs at much higher frequency, usually at half or even around the f_T of the transistor. Then of course, I don't know what kind of transistors James is using. A local oscillation will usually occur with capacitve loading, and it will start on one polarity. A ferrite bead can cure the problem.

James:
Could you summarize the points of the article, especially what the topology of this PLIL technique is and why it would be superior? Or might I even ask you for scans?

traderbam:
I like to keep the amp as fast as I can but be stable with all kinds of loads. The reasoning behing this is:
- If there is instability, it is usually due to some positive feedback. This means it can grow as the load changes slightly or even the temperature of some components changes, maybe even snowball and fry something.
- It's true we cannot hear beyond 20 kHz, but the input junctions are essentially rectifiers, so they will generate a DC or LF component from this. And assuming the HF varies nonlinearly with the audio signal, we get our audio band signal modulated. There was an Analog Devices app note on HF rectification in metrology amps.

dkemppai:
You may be right that the input stage can linearize the rest of the system. But intuitively, this feels like trying to balance something heavy and unwieldy with a stick made of rubber. Sorry, just a feeling, no solid argument.

I am not sure we have the same definition of pole splitting. In my understanding, a Miller cap without the resistor does the pole splitting. The argument is as follows: Usually the output stage is the slowest stage, so it dictates how the other stages must be rolled off. The Miller compensation makes for a lower impedance drive to the output stage, so it becomes effectively faster. This means that the overall gain rolloff need no longer be as severe as it would have been estimated from looking at the uncompensated circuit. In other words, you can choose a higher open loop rolloff point with VAS Miller compensation compared to input stage rolloff.

Regards,

Eric
 
capslock said:
James/Hugh/Sonny:
I am not sure that this is a local oscillation of the output transistors because in my experience this occurs at much higher frequency, usually at half or even around the f_T of the transistor. Then of course, I don't know what kind of transistors James is using. A local oscillation will usually occur with capacitve loading, and it will start on one polarity. A ferrite bead can cure the problem.

Regards,

Eric

Yes you are right, the oscillations also tends to die out. and it is mostly into a capacitive load.

Sonny
 
Hey, I've just bought this CD by Marie Frank. Never heard of her. It's really good!

dkemppai wrote: "Will linearizing the miller capacitance of the VAS stage matter if the feedback through the output stage back into the long tailed pair is fast enough?"
In theory, provided the circuit is stable the more feedback the less this non-linearity will matter, yes. IME it is very difficult to make the vas stage stable without a compensation cap to slow down the collector voltage (or really killing its gain). I've tried this many times: various difficult-to-track-down parasitic feedback paths tend to terrorize (topical or what?) the performance. So I'd say you need several 10s of pF minimum and a sensible pcb layout to avoid parasitic instability. The latter cannot be fixed at the LTP stage. With a vas BJT with no emitter resistor this creates a pole at some several kHz - and a 90 deg phase shift. Then the rest of the amp is normally designed to add no more than 45deg up to the point where loop gain is unity.
 
Namestra,
I am not familiar with the Self circuit, can you post it? The oscillation you are seeing could be caused by several things.

For a start I would check the basic things first - like your psu and grounding arrangements. Make sure you are using separate ground wires appropriately and your are star earthing (there is a whole big thread on this in here somewhere) and that psu cables are tightly bound together to minimize loop inductance.

Then we need to check the design.

BAM
 
There are at least three ways to tame the VAS stage:

a) emitter degeneration: brings down gain pretty independent of frequency, so constant loop gain in the audio band can sometimes be achieved, linearizes VAS stage, increases input impedance, does little to change output impedance

b) Miller compensation cap between C and B: frequency dependent feedback, lowers input and output impedance as frequency increases, may sometimes be the only compensation needed.

c) resistor between C and B: never tried it, would also bring down gain independent of frequency same as a), would decrease input and output impedance similar to b), but independent of frequency;
relaxes demands on overall compensation same as a), but compensation must still be done on some other stage


comments?
 
Dear capslock,

first pole time constant is (VAS input resistance)*(Cbc VAS)*(VAS gain)

second pole time constant is (output stage input resistance)*(C associated with VAS output node)

in your cases a) and c) the second pole time constant will remain unchanged and you should make much higher first pole time constant to move second pole below unity loop gain. Only in case b) you will get the second pole time constant lower (then original one) due to NFB via Miller capacitor. With higher frequency second pole you can use not so low the dominant one.

another nonlinearity is associated with VAS output node - nonlinear input current of the output stage. To makes this nonlinearity lower you should keep VAS output resistance low - again by Miller compensation. Here the advantage over c) will be that the VAS output resistance will be lower with frequency, that helps to keep associated disto at low level (please keep in mind that the overall loop gain is lower with frequency)

so for the discussed topology LTP-VAS-output stage b) is superior
 
dimitri said:

second pole time constant is (output stage input resistance)*(C associated with VAS output node)

May be this is too simple. Suppose you put a class A emitter follower buffer (which is made of very fast low or medium power transistors) in front of the output stage. It will increase input impedance by say a factor of 200, namely the AC current gain. Input capacitance will not be brought down significantly (or actually not at all if original predrivers and the additional buffer stage are both using the Sanyo video transistors discussed above) + there is nothing we can do about the output capacitance of the VAS. So will an additional buffer bring down the second pole by a factor of up to 200? My guess is it will not cause a change by more than 10%.
 
capslock wrote---------
I also like to have the open loop gain be constant across the Audio bandwidth as this produces a Consistent
THD number with frequency and not have the rising THD vs Frequency typical of Circuits using high open
loop gain and limited open loop bandwidth
---------------------------------
Then you can use another topology with several gain stages and individual frequency compensation in the each stage, like in 29 years old Otala amp: http://home.online.no/~tsandstr/OtalaStory.htm

capslock wrote---------
I suspect it may be better to go for the Miller compensation, because it will:
a) linearize the parasitic inherent Miller capacitance of the VAS transistor
b) be more efficient than rolling off the input stage gain because of the pole slitting action, i.e. greater overall bandwidth can be achieved
c) reduce distortion caused by the nonlinear input impedance of the output transistors because it lowers the output impedance of the VAS stage because of its feedback action
----------------------------
a) - false statement. The sensitivity to changes in nonlinear Cbc will remains unchanged with addition of the external capacitor. The solutions are cascode or emitter follower before VAS
b), c) - true
 
Status
Not open for further replies.