Miller Capacitance

[tapehead ted]

Hello all,
It's come to my attention that high frequencies can bleed from a triode's anode to it's grid via the miller capacitance. This would seem to form a kind of anode follower where high frequencies are fed back more or less full strength through the grid. It's negative feedback, and of the kind that would make the input resistance at those frequencies approximately equal to the grid stopper plus the output resistance of the previous stage, like any other anode follower as far as I can see.

So the low pass filter formed by the grid stopper, Zo of the previous stage, and the miller capacitance is flattened out by this high frequency negative feedback, and bandwidth is extended anyway.

Just negative feedback so far, so not inviting oscillation.

But when there are two triode gain stages, the first feeding the second, the high frequency bleed from the second triode is present at the anode of the first triode, although it has to make it's way through the gird stopper of the second stage to get there. It may also have to go through a coupling cap, but in a directly coupled (DC coupled w/o a potential divider) situation, all there is in the way is a grid stopper.

So the bleed from the second stage anode to the second stage grid flows through the second stage grid stopper to the anode of the first stage, where it again bleeds through, via the first stage miller capacitance, to the grid of the first stage, where it is now positive feedback as it reappears at the second stage anode after passing through two inverting gain stages, and the loop gain at high enough frequencies could be in the hundreds or thousands, almost the entire combined gain of the two stages. Anode resistances and anode resistors form a bit of a potential divider, but not nearly enough to reduce the loop gain below 1.

This looks to me to be a disastrous situation, and I can't see how to prevent it. Can that second stage grid stopper somehow be sized to manage this? It's worth noting that if both triodes are in the same envelope, the second stage grid stopper is bypassed by the series combination of the miller capacitance and the anode to anode capacitance.

I've found the odd NOS data sheet that lists Ca-a, and it's the two or three pf you might hope for, but an Electro-Harmonix 12AX7 datasheets lists 520pf! I'm not seeing any other new tubes even listing Ca-a, so thank you EH for the warning.

Thoughts?

 

[Johan Potgieter]

..... but an Electro-Harmonix 12AX7 datasheets lists 520pf! I'm not seeing any other new tubes even listing Ca-a, so thank you EH for the warning.

You are indeed correct; this capacitance provides negative feedback, but too little to be of concern, except as Miller capacitance in conjunction with a high enough input resistance. As such in conjunction with a high input resistance it may begin to attenuate high audio frequencies.

Not usually given (as you found), the EH value of 520pF astounds. Intuitively one would expect something of the order of 1 pF judging by other inter-electrode capacitances. I found two other data sources listing Ca-a as 0,75pF; much more in line with expectations. Oscillation in audio amplifiers could occur where the reactance of Ca-g causes enough phase shift in conjunction with other circuit values, to turn the feedback angle positive.

This changes your logic somewhat.

 

[trobbins]

perhaps the EH data has a typo - p instead of f.

 

[gorgon53]

520pf no way, more like 520 femtofarad

 

[tapehead ted]

I called EH and talked to somebody about not that value but the heater to cathode capacitance given as "5nf (nominal)". The person I spoke to did not dispute the figure or that it was a thousand times greater than you might expect. Their defense was that ROHS regulations prohibited using the right materials. The filament is tungsten and the insulation is aluminum oxide, so that doesn't sound too convincing.

JJ just said small capacitances are hard to measure but it was OK.

I think it quite possible that they manufacture to certain specs and just let others fly...

 

[DF96]

Yes, 520fF sounds more plausible for Ca-a. Note that 'pf' is not a capacitance, as the symbol for farad is 'F' not 'f'.

It's come to my attention that high frequencies can bleed from a triode's anode to it's grid via the miller capacitance.

I'm not certain that "bleed" is a useful word to use here. It may lead you astray. Note that Miller capacitance is what the input source sees; maybe what you mean by "Miller capacitance" is anode-grid capacitance?

It's negative feedback

That depends on what impedance is attached to the grid.

and of the kind that would make the input resistance at those frequencies approximately equal to the grid stopper plus the output resistance of the previous stage, like any other anode follower as far as I can see.

So the low pass filter formed by the grid stopper, Zo of the previous stage, and the miller capacitance is flattened out by this high frequency negative feedback, and bandwidth is extended anyway.

Sorry, you lost me there. Miller capacitance usually restricts bandwidth.

But when there are two triode gain stages, the first feeding the second, the high frequency bleed from the second triode is present at the anode of the first triode, although it has to make it's way through the gird stopper of the second stage to get there. It may also have to go through a coupling cap, but in a directly coupled (DC coupled w/o a potential divider) situation, all there is in the way is a grid stopper.

So the bleed from the second stage anode to the second stage grid flows through the second stage grid stopper to the anode of the first stage, where it again bleeds through, via the first stage miller capacitance, to the grid of the first stage, where it is now positive feedback as it reappears at the second stage anode after passing through two inverting gain stages, and the loop gain at high enough frequencies could be in the hundreds or thousands, almost the entire combined gain of the two stages. Anode resistances and anode resistors form a bit of a potential divider, but not nearly enough to reduce the loop gain below 1.

This looks to me to be a disastrous situation, and I can't see how to prevent it.

As this is not disastrous, maybe your analysis is incorrect?

 

[tapehead ted]

I was at least partially incorrect. I was using the miller capacitance in the feedback equations, when it is the feedback that creates the miller effect. It all worked out when I just used Cg-a. I used the capacitive reactance of Cg-a in the feedback equation, and got something very much like an RC filter using Cg-a*gain (the miller capacitance).

So I was counting the miller effect twice. One of the effects of this kind of feedback (for instance in an anode follower, where additional capacitance is usually connected between grid and anode) is that the input impedance is reduced by feedback and becomes closer to the value of the grid stopper (and whatever else adds to the series resistance).

I'm wondering if the input resistance is thus decreased at the effected frequencies in addition to the effect of the multiplied (miller) capacitance, or if I would be again counting the miller effect more than once. Writings on the anode follower are infrequent and somewhat incomplete, but this does happen with other anode followers as best I can make out- not only is gain reduced by feedback, but input resistance is reduced to a value determined not only by the grid leak resistance but the grid stop resistance as well.

(From now on I'll refer to pF, not pf, thank you!)

 

[trobbins]

JJ just said small capacitances are hard to measure but it was OK.

Many modern cheap multimeters have a capacitance range that would easily confirm (or not) a heater-cathode capacitance value above about 400pF, and some cheap meters go down to single digit pF measurement.

That capacitance won't change with heater being powered or not.

You could also go the more laboratory path, and use a signal generator to drive a series LCR circuit and check for the signal maxima across the external sense R. I haven't done that yet, but may just set it up soon.

 

[DF96]

I'm wondering if the input resistance is thus decreased at the effected frequencies in addition to the effect of the multiplied (miller) capacitance,

Yes, input resistance can be affected too. You would need to consider the frequency-dependent behaviour of the feedback network, and the frequency-dependent behaviour of the forward gain provided by the valve and its anode load. In some cases the result can actually be an inductive input impedance.

 

[tapehead ted]

Many modern cheap multimeters have a capacitance range that would easily confirm (or not) a heater-cathode capacitance value above about 400pF, and some cheap meters go down to single digit pF measurement.

That capacitance won't change with heater being powered or not.

You could also go the more laboratory path, and use a signal generator to drive a series LCR circuit and check for the signal maxima across the external sense R. I haven't done that yet, but may just set it up soon.

I'd love to see some measurements along this line- I'm not experienced taking such measurements. JJ has told me some things in the past along the lines of "that's all taken from old datasheets"1 I did see something on googel where someone took some measurements of a new Mullard vs. an NOS- I wonder what all they measured.

DF96- I don't know how I would recognize an inductive input impedance- maybe by the curve?

Signals can appear at the anode from various places other than the grid. From the power supply, for instance. And there's a way, via Cg-a, to get from there to the grid. What I see now is that using the feedback equation the "Rin" is XCg-a for any such signals, plus whatever output resistance they came from. So when they are fed back through the feedback impedance is also XCg-a, so they have a gain of around one. Not the oodles of gain I anticipated.

Still, a gain of around potentially problematic.

 

[Miles Prower]

Hello all,
It's come to my attention that high frequencies can bleed from a triode's anode to it's grid via the miller capacitance. This would seem to form a kind of anode follower where high frequencies are fed back more or less full strength through the grid. It's negative feedback, and of the kind that would make the input resistance at those frequencies approximately equal to the grid stopper plus the output resistance of the previous stage, like any other anode follower as far as I can see.

That's not right. Miller Effect appears at the grid as an AC impedance that is smaller than it would otherwise appear. It's reverse bootstrapping, and why common cathode has the poorest high frequency performance. In common cathode topologies, the feedback is through the reverse transfer capacitance.

If you reference the signal at the plate to AC ground, the output and input are anti-phase. If referenced to the DC rail, these signals are in phase. It's this in phase voltage that C_rt sees, and therefore, the feedback is positive, not negative. This is how the "Miller oscillators" -- such as the tuned plate/tuned grid oscillator -- works even though there seems to be no connection between the plate and grid. Positive feedback through C_rt drives the oscillation. It's also the reason why triode RF amps require neutralization to operate with stability: you need to feedback enough anti-phase signal to swamp out the inherent +FB. (Pentode RF amps often require neutralization as well even though C_rt is much smaller, the voltage gain is also much larger.)

Having a source of +FB in an audio stage can sometimes cause high frequency oscillation by forming an accidental Colpitts oscillator with stray impedances. Grid stoppers prevent this by loading the stray LC tuners, lowering their Q to the point where oscillation won't occur.

But when there are two triode gain stages, the first feeding the second, the high frequency bleed from the second triode is present at the anode of the first triode, although it has to make it's way through the gird stopper of the second stage to get there. It may also have to go through a coupling cap, but in a directly coupled (DC coupled w/o a potential divider) situation, all there is in the way is a grid stopper.

So the bleed from the second stage anode to the second stage grid flows through the second stage grid stopper to the anode of the first stage, where it again bleeds through, via the first stage miller capacitance, to the grid of the first stage, where it is now positive feedback as it reappears at the second stage anode after passing through two inverting gain stages, and the loop gain at high enough frequencies could be in the hundreds or thousands, almost the entire combined gain of the two stages. Anode resistances and anode resistors form a bit of a potential divider, but not nearly enough to reduce the loop gain below 1.

This looks to me to be a disastrous situation, and I can't see how to prevent it. Can that second stage grid stopper somehow be sized to manage this? It's worth noting that if both triodes are in the same envelope, the second stage grid stopper is bypassed by the series combination of the miller capacitance and the anode to anode capacitance.

I've found the odd NOS data sheet that lists Ca-a, and it's the two or three pf you might hope for, but an Electro-Harmonix 12AX7 datasheets lists 520pf! I'm not seeing any other new tubes even listing Ca-a, so thank you EH for the warning.

Thoughts?

Feedback through more than one stage is via the DC rail, and causes low frequency instability. It can be a possible problem with RF stages, and the cure for that kind of instability is baffle shields between stages to prevent radiation and magnetic coupling between LC tuners that are tuned to the same frequency. With RF, you can also have a potential feedback mechanism across multiple stages if the chassis forms an accidental waveguide. The fix for this is also baffle shields, and/or building the circuit in a chassis that's long and thin (seen often in o-scope vertical deflection amps) if you anticipate such a problem.

A value of 520pF for anode-anode capacitance is ridiculous. I have multi-plate variable capacitors that don't have that much capacitance. It's either a typo or a dropped decimal point.

 

[DF96]

I don't know how I would recognize an inductive input impedance- maybe by the curve?

Frequency response of amplitude and phase of impedance.

Signals can appear at the anode from various places other than the grid. From the power supply, for instance. And there's a way, via Cg-a, to get from there to the grid. What I see now is that using the feedback equation the "Rin" is XCg-a for any such signals, plus whatever output resistance they came from. So when they are fed back through the feedback impedance is also XCg-a, so they have a gain of around one. Not the oodles of gain I anticipated.

It may be simpler to calculate the impedance at the anode, taking account of feedback capacitance etc.

 

[tapehead ted]

Thank you Miles- the visual of the big air variable cap somehow fitting into the 12zx7 envelope it hilarious and gets the point across to me very memorably! I did some primitive measurements and much prefer it to talking to supposedly knowledgable people at the US offices of the distributors of Eastern European tube manufacturers... Thanks again, never thought of referencing phase to the DC rail.

Actually I did see that whatever came in on the DC rail was in phase at all anodes it appeared at. At some point I lost track of that. So the diminishing impedance at the grid due to feedback IS the Miller effect- again I was counting it twice. Is the reverse transfer capacitance the same value as Ca-g?

 

[hpeter]

A value of 520pF for anode-anode capacitance is ridiculous. I have multi-plate variable capacitors that don't have that much capacitance. It's either a typo or a dropped decimal point.

the huge a** stc4212 has only Cag 18,8µµF

 

[tapehead ted]

I'm also seeing a path along the ground rail, bleed (or unwanted signal anyway) crawling up cathodes and grid leaks. Depending on the circuit this could be more significant than unwanted signal on the HT.

I've been mapping out circuits with an eye to the fact that signal has no idea where I intend it to go- it just goes wherever it can go.

 

[DF96]

Poor grounding is a well known source of problems.

 

[rmb]

That's not right. Miller Effect appears at the grid as an AC impedance that is smaller than it would otherwise appear. It's reverse bootstrapping, and why common cathode has the poorest high frequency performance. In common cathode topologies, the feedback is through the reverse transfer capacitance.

A value of 520pF for anode-anode capacitance is ridiculous. I have multi-plate variable capacitors that don't have that much capacitance. It's either a typo or a dropped decimal point.

520 pF anode to anode capacitance is ridiculous. Measure the capacitance between the two anodes in a 12**7 and this will be a few pF and unaffected by Miller effects.

520 pF Miller capacitance IS believeable. Another reason to keep input leads short and minimize stray capacitance. 5.2 pF on the grid input lead, Miller that by 100 (u for 12AX7) and you have effectively 520 pF at the input. Assume 1 megohm from grid to ground, time constant is in audio range....

I don't like the 12AX7. Most are microphonic and noisy. The 7025 was better but still had inherent 12AX7 problems. I like the 12AY7, 6072, or D3a (triode connection) for preamp work...

"Stopper" resistors ARE a good idea and one RARE use for carbon resistors, especially for triodes such as the 6DJ8, 6922, 6BQ7 types... stoppers are REQUIRED for the 417A, 5842, D3a, and other high transconductance types... "Stopper" resistors are generally in the 300 to 1000 ohm range so the carbon resistor doesn't introduce excessive noise...

 

[tapehead ted]

Thanks for moving this thread.

I'm looking at noise/signal from ground entering a tube from the cathode, where since the output resistance of the ground is very small, there's a potential divider made of say, in a typical common cathode stage, the cathode resistor, maybe bypassed to be even a smaller part of the potential divider, and rk of the valve, which I'm reading is (Ra+ra)/(u+1). Not much attenuation there.

Then the path up the grid leak, where the potential divider could be made up of the grid leak resistor and the anode resistance of the previous stage.

I'd like to believe that when ground noise arrives on the cathode it tends to cancel with the ground noise arriving on the grid, and if somehow they came out the same amplitude they would cancel completely.

 

[tapehead ted]

So on the topic of miller capacitance, I am now concluding that there are two distinct effects that have similar results.

One is that the negative feedback (seen from an AC perspective, not from the HT rail perspective) causes Ca-g and any other capacitance that it placed between anode and grid to appear multiplied by the gain of the stage.

A separate effect is that input impedances are reduced at the frequencies affected by the feedback, because it is shunt-applied feedback. Enough feedback and it would become a virtual earth.

It helps me to keep these distinct to consider that series-applied negative feedback would increase the input impedance, while still reducing the gain.

Only, now that I think of it, I'm not sure that series-applied negative feedback would make Ca-g appear any larger... oh dear, right when I thought I had it...

 

[Miles Prower]

Only, now that I think of it, I'm not sure that series-applied negative feedback would make Ca-g appear any larger... oh dear, right when I thought I had it...

You can't touch C_ag except in very limited ways since it's inside the device itself and inaccessible. The only thing you can do is parallel it with an inductor to make a parallel tuned trap. It's done to stabilize RF amps that operate at one frequency or over a narrow band of frequencies.

Stopping Miller Effect requires either common grid where C_ag becomes a parallel capacitance across the load, or common plate where it parallels the input while the normal C_gk is bootstrapped to a much smaller effective value. Why both CG and CP have better high frequency performance at the cost of current gain and input impedance (CG) or voltage gain (CP).

Any application of feedback is external to the device, and doesn't affect the internal operation at all.