Miller compensation - pros/cons?

[Joel]

This type of loop stabilizing compensation cap is seen all the time inside opamps, and SS audio amps, but very rarely in tube amps - and almost never in vintage gear. This has me wondering if there's a reason for that.

An externally hosted image should be here but it was not working when we last tested it.

It seems to be a good technique. It works. This example is from the amp I've got on the bench right now, and the 47pF cap was able to stabilize the amp with 20dB of NFB, while still retaining a closed-loop bandwidth of -3dB at 100kHz. Whereas it took a much larger cap (330pF) from plate to ground of the first stage to get similar stability... but the amp was then -3dB at <60kHz.

I didn't measure distortion at 20kHz with both methods, but I would bet that it was MUCH lower with the Miller cap.

Subjectively, after an all-too-brief listening session, I have to say that it sounded better with Miller compensation as well.

So, what, if anything, am I missing??

Joel

 

[tubetvr]

This has me wondering if there's a reason for that.

Yes, there are several reasons.

First is that you only need phase compensation if you use feedback, most classic audio amplifier use limited feedback and open loop response is determined by the output transformer, using Miller compensation in such an amplifier would seriously limit the bandwidth. For preamplifiers it was not so common to use heavy feedback so compensation was not needed. In a single stage amplifier or a multistage amplifier with a limited number of stages stability is achieved without compensation as long as the feedback is reasonable.

2nd reason, Miller compensation waste gain. Early it was discovered that this method for loop compensation was not optimal, there are much better methods available where you can keep the open loop gain higher and therefore achieve better closed loop performance but still have the same stability as with Miller compensation. In order to do this you need to design networks that is specific for an amplifier and have the effect of compensating phase at critical frequencies so as to keep stability. It is possible to show that using this kind of networks is always better than using Miller compensation. For good examples of this method see Valley and Wallman, Vacuum tube amplifiers chapter 9 "Low frequency amplifiers with stabilised gain" I use this method in my OTL and achieve very wide bandwidth with good stability in spite of using heavy feedback.

Phase compensating networks have been used in some classic audio amplifiers like the Williamson

Summary: There are better methods available than Miller compensation if stabilisation is needed and in many classical circuits very little compensation is needed or used.

Regards Hans

 

[Tom Bavis]

What's wrong with it? It isn't TRADITIONAL... Norman Crowhurst never mentioned it to my knowledge. I assume you've read Norman Koren's page? http://www.normankoren.com/Audio/ esp. "Feedback and Fidelity".

One thing I'd suggest... use a NPO ceramic capacitor - if you use mica, it shouldn't have DC across it - a friend told me it was a design rule at Harris Radio years ago. Silver + DC = silver migration. It may take years, but leakage develops across mica caps.

 

[Joel]

Hans,

The step networks, like in the Williamson, also shunt down the load resistance - ie. throw away the gain. So, why is this superior?

And, I was under the impression that -3dB at 100kHz was pretty good with an output transformer in the loop! Apparently you think otherwise?

Phase compensation will stabilize the amp (150pF across the 15k feedback resistor), but the output has large overshoot and ringing on square waves. The whole point of dominant pole compensation is to make the gain of the amplifier significantly down at the point where the phase shift equals 360 degrees. So, then it's necessary to make the open loop bandwidth "poor".

But, hey, feedback is a complex subject, and I am NO master of it, by any means.

regards,
Joel

 

[EC8010]

It strikes me that you are achieving much the same thing by adding a capacitor from anode to grid as you are by adding it from anode to AC ground. Yes, your capacitor is smaller, but that's because its value is being multiplied by the gain of the stage - Miller effect.

However, it also seems to me that Miller compensation will be dependent on the resistance of the source driving the amplifier and will cause the entire amplifier to have a higher input capacitance than if the capacitor had simply been placed across the anode load.

I like to use an RC network across the anode load and another RC network across the feedback resistor and juggle them for the best result. It would help if I had four hands, but two are surprisingly effective. My best amplifier manages -3dB at 200kHz, but I've got a silvered mica capacitor in the anode step network...

That's the first I've heard of silver migration in silvered mica capacitors - any more detail?

 

[kevinkr]

It works, but one of the reasons why it was not commonly used is that I suspect it raises the input capacitance in exactly the same way as the miller capacitance in a triode does. Using a high mu triode with miller compensation could result in very large effective input capacitances .

 

[tubetvr]

The step networks, like in the Williamson, also shunt down the load resistance - ie. throw away the gain. So, why is this superior?

It throw away less gain and the gain is only reduced above a certain point, it is a fact that this kind of compensation is vastly superior to Miller compensation, pls read Bodes work where he explains about optimal frequency/gain behavior.

And, I was under the impression that -3dB at 100kHz was pretty good with an output transformer in the loop! Apparently you think otherwise?

No, it is quite good but I know that it is possible to achieve better performance with other methods.

Phase compensation will stabilize the amp (150pF across the 15k feedback resistor), but the output has large overshoot and ringing on square waves.

This is wrong, it doesn't have to be like that, it only depends on how you design the phase compensation network, you can achieve any degree of overshoot you want or no overshoot.

The whole point of dominant pole compensation is to make the gain of the amplifier significantly down at the point where the phase shift equals 360 degrees. So, then it's necessary to make the open loop bandwidth "poor".

Yes but that is what is bad, with dominant pole compensation you will get a gain that drops 6dB per octave but with phase compensating networks ala Bode you can achieve the optimum 9dB per octave asymptotic resulting in higher open loop bandwidth.

The key question is why Miller is used in most opamps and in many SS audio amplifiers when other type of compensation is superior?, there is a simple answer to this riddle even though there are quite many opamps that allow advanced compensation schemes and where it is also described in data sheets.

 

[Joel]

It strikes me that you are achieving much the same thing by adding a capacitor from anode to grid as you are by adding it from anode to AC ground.

Well, not really though, because with the Miller cap there is negative feedback. That has to linearize things to some extent. With shunt capacitors, and/or step networks, there is no NFB.

As for the input capacitance, let's see:
with the 12AX7 as drawn, I figure a total input C of about 137pF including strays and the 47pF cap. With the typical Zo of a good preamp, -3dB would be somewhere near 900kHz. Isn't that too high to even be a factor?

Joel

 

[Joel]

This is wrong, it doesn't have to be like that, it only depends on how you design the phase compensation network, you can achieve any degree of overshoot you want or no overshoot.

Hans,

My understanding is that, if you solely relied on 'phase compensation', you would be sending quite large voltage spikes back to the input of the amp, since the amp has not been bandwidth limited to below the point where the feedback loop has become positive feedback - and that this will not only cause momentary overloads of the input, but also slew-rate limiting.

Joel

 

[aletheian]

This type of loop stabilizing compensation cap is seen all the time inside opamps, and SS audio amps, but very rarely in tube amps - and almost never in vintage gear. This has me wondering if there's a reason for that.

An externally hosted image should be here but it was not working when we last tested it.

It seems to be a good technique. It works. This example is from the amp I've got on the bench right now, and the 47pF cap was able to stabilize the amp with 20dB of NFB, while still retaining a closed-loop bandwidth of -3dB at 100kHz. Whereas it took a much larger cap (330pF) from plate to ground of the first stage to get similar stability... but the amp was then -3dB at <60kHz.

I didn't measure distortion at 20kHz with both methods, but I would bet that it was MUCH lower with the Miller cap.

Subjectively, after an all-too-brief listening session, I have to say that it sounded better with Miller compensation as well.

So, what, if anything, am I missing??

Joel

I'm no expert on any of this by any means... usually I'll go by ear and scope, and just call it a day when I like what I hear, but I have used that Miller compensation scheme several times, as well as adding resistance in series with the cap, or RC plate snubbing OR grid to ground caps OR Williamson style compensation. I'll just try them all and see what works best.

But as a result, I probably don't use the 'absolute best' method possible, but the cap feedback loops has worked wonders on occasion where other methods sounded dull. (I guess if my designs were better, then I might not HAVE to use so many tricks on occasion... LOL)

my $ .02

 

[tubetvr]

and that this will not only cause momentary overloads of the input, but also slew-rate limiting.

That is only of interest if the input signal is not bandwidth limited which it should be in any good amplifier.

However a phase compensating network doesn't have to be in the feedback loop, in my OTL I have 2, one for low frequencies and one for high but both are in the amplifier path.

I have designed low frequency compensating networks for the Williamson amplifier also see here http://www.tubetvr.com/Williamson.pdf
for original with compensation only for high frequencies and here with additional compensation for low frequencies http://www.tubetvr.com/Williamson_comp2.pdf
note that neither of these have compensation in the feedback loop.

Regards Hans

 

[Joel]

Hans,

Those are not "phase compensation". Those are dominant pole compensation.

Phase compensation "speeds up" the feedback loop at high frequencies so it can "catch" the input signal with the highly phase-shifted amplifier output signal.

Joel

ps. If you adjust your low frequency step network so that the .022uF cap is across the 1.5M resistor only, then it can be changed to a 100V rated unit. The way you have it now, both the caps need to be 400V rated. The network will have the exact same response. FYI.

 

[tubetvr]

Those are not "phase compensation". Those are dominant pole compensation.

Then you are using your own definition on what a dominant pole compensation network is. These networks will not have a noticeable effect until a certain limit where they will affect the phase, the total response is a sum of this response and the rest of the circuit elements. A dominant pole compensation is where you have one single pole that determine the response of the whole amplifier. Pls read a text about how to stabilise amplifiers using phase compensating networks, the text I referred to by Valley/Wallman is quite good if you don't want to read the original Bode text.

Regards Hans

 

[Joel]

Then you are using your own definition...

Maybe I am. I apologize if I've got it wrong.
I own Valley and Wallman. I'll look through it again tonight.

Thanks!
Joel

 

[ilimzn]

It works, but one of the reasons why it was not commonly used is that I suspect it raises the input capacitance in exactly the same way as the miller capacitance in a triode does.

Precisely - since the exact place where the cap is put, is where the Miller capacitance occurs. It gets multiplied by gain, which is the reason you can use small values. That being said, this is not nearly as important as the fact that the time constant depends on the source impedance (which could include inductance and the possibility of resonance!). This is just fine if the previous stage is set (i.e. constant), which in the case of amp input, it is not. Then there is also the PSRR issue, since output depends on supply rail, input gets some of this as feedback via miller).
With plate to AC gorund, the input capacitance is not disturbed, and is smaller, and normally should not be a signifficant factor regarding overal closed loop response.

 

[Joel]

The high frequency roll off due to this cap is not caused by the source impedance.

Don't believe me?
On my bench I'm driving the amp from 600 ohm generator, plus the 1k ohm grid stopper. Total input capacitance is ~137pF.

That gives us a -3dB point of 726,440Hz. I assure you that the ampifier itself has ZERO output at 700kHz. So, this would not be able to 'compensate' it at all.

 

[EC8010]

How did you measure your input capacitance?

 

[kevinkr]

Umm, this circuit is also known as a miller integrator and it's performance is affected by both the source impedance of the prior stage as well as the generator impedance of the stage driving it. (for modeling purposes you can usually lump the two together.)

1.6K is a pretty small source impedance, try the thevenin equivalent source of a 100K volume pot for example which worst case would present a source impedance of 25K to the input of that first stage.

Also unless I have got a screw loose the effective miller capacitance is theoretically mu * capacitance from grid to plate, although in this case you should substitute the effective stage gain for mu..

I think you will find that it is actually a very large capacitance present at the input relative to ground depending on effective mu 3000pF or more and theorectically you will start to roll off around 33kHz if it were actually this bad. (Which I am sure it actually isn't.)

JMO..

 

[Joel]

EC8010,

It's calculated, not measured. Could be a bit higher in real life, although I did add 4pF for strays.

* * *

So, I think the question actually did get answered, in a round-about way. Let me take a shot at a summation:

This is used to good effect in opamp circuits, etc, because the frequency at which instability occurs is so high that a small value cap can be used to compensate. Whereas in the typical tube amp, especially one with an output transformer that has poor high frequency response, the cap needs to be so large (in order to pull the response down below the offending frequency) that the input capacitance to the amplifier becomes impractically large. However, in the case where the transformer has wide bandwidth, a more reasonable sized cap can be used, and the technique is quite effective.

How's that sound?

My Magnequest B18 seems to have pretty good HF extension, so I'm able to get away with 47pF, which isn't horrible. Especially because I use a preamp with an output transformer. 270 ohms output impedance.

Thanks so much to all who responded.

Joel

 

[Johan Potgieter]

Perhaps just a few general points; most has already been said.

1. In a good design the open loop gain of an amplifier should ideally be flat over the audio spectrum, i.e. phase corrective measures should really only be necessary outside the band (in this case > 20 KHz). In that case linearity, such as Miller compensation will improve, is not of consequence.

2. Miller dominant pole compensation is more prevalent in semiconductor circuits, where current feed to the next stage is present. In tube circuits the "next stage" is simply a resistor which can mostly be quite high, i.e. the load is simple. The usual R.C compansation network over the anode load resistor of the 1st stage is simple and effective enough in most cases and isolated in its influence to that point. Such a network is often necessary to off-set phase shift elsewhere, but ....

3. As said, this is a lagging phase shift. While simple and easy it causes the fed-back signal to always "arrive late", in a manner of speaking, thus giving rise to transient overload problems. Stability is more effectively achieved by combining this with a lead capacitor (or RC) over the feedback resistor. In many cases the latter is all that is necessary. Correct compensation can be achieved comparatively easily by viewing a (say) 5 KHz square wave and correcting for overshoot. (A little experience is needed to interpret the result in order to get the right C and serie R. A complete frequency run into the MHz region is more informative.)

4. Especially in semiconductor design (which is outside the scope of this thread) the matter can be thorny. I have suggested elsehwere an informative article by Dr John Ellis (Electronics World, March 2001 - year under correction) suggesting the alternative of PLIL (phase-lead-input-lag) compensation. It contains several gain/phase diagrams to illustrate the point, and is also valuable for general knowledge, although problems with tube designs are not as complex as with semiconductors.

For a full understanding basic Nyquist/Bode theory needs to be studied as suggested above.

Regards