Treble shunt - before or after grid-stoppers?

[Gaz]

I'd like to add a treble shunt cap from grid to ground, but was wondering if it should go before or after the grid stopper. I cannot hear a difference switching positions, so my assumption is that it doesn't matter. Thanks for any advice.

[txstrat]

From a gut feeling (I know this is not how you should plan a tube amp ) I'd put the treble shunt before the grid stopper.
That way all of the leftover signal has to go through the resistor to the grid.

[R.G.]

A grid stopper works because there is some parasitic capacitance from grid to cathode and grid to plate. If the source is not grounded directly or through another cap, the grid-cathode capacitance is paralleled by an even smaller capacitance to ground from the grid wiring. The sum of grid-source, grid-ground and amplified grid-plate capacitance looks like an equivalent capacitor to ground. The amplified grid/plate capacitor is usually the biggest cap in this three-way unless the tube is being used as a follower. The grid-plate cap looks exactly like a grid-ground cap that is the voltage gain times the actual grid-plate capacitance. This is known as the "Miller capacitance" in textbooks.

In any case, there is an equivalent capacitor from grid to ground. The grid stopper puts a resistor in front of this, and causes a single-pole RC lowpass filter at the grid. This is generally enough to both roll off the highs coming into the stage at some frequency, and also to drive the grid from a lower impedance at higher frequencies, both of which contribute to RF stability of this single stage in isolation, although it can make oscillation worse if there is overall feedback around grid-stoppered stages.

You may or may not hear a difference. That's because the RC lowpass filter may be above audio or above the frequencies you're putting into it. But I can assure you that you *can* make it make a difference. I once used 100K grid stoppers in a two-12AX7 preamp box I built inside a very small box. The grid stoppers were the only way to keep it from squealing, at about 47K. At 100K, they gave me a butter-smooth high end. Note that this was hand tuning on that one circuit/setup. YMMV.

[Gaz]

RG, thanks for the explanation about grid-stoppers. So to answer my question, the cap would need to be after the grid-stop, and effectively in parallel with the miller capacitance?

And you're right about not being able to hear because the roll-off is above audio frequencies, but the different I couldn't hear was not with or without the grid-stop altogether, but moving the cap before or after it. I'm not sure if you misunderstood that part of my post, or I'm just missing your point.

Specifically, I'm wanting to put a small capacitance (220pf) from grid to ground at the power tubes to protect against oscillation in the power stage. I'm using four output tubes (EL34) in push-pull with 5.5k grid-stoppers. If the cap can come before the grid-stoppers, then I only have to use one cap per side, and layout becomes simpler as well. Otherwise, if the caps have to be in parallel with the Miller capacitance, and after the grid-stoppers, I will have to use four caps, one for each tube. Hope that all makes sense, and clarifies my question.

Thanks for the advice so far.

[Dave H]

...but the different I couldn't hear was not with or without the grid-stop altogether, but moving the cap before or after it.

What is the source impedance of the driving stage? If you can hear a difference between having a grid to ground cap and no cap but don't hear a difference when the cap is moved to the other end of the stopper resistor it could be because the output impedance of the driving stage is much greater than the 5.6k grid stopper.

Dave H.

[R.G.]

RG, thanks for the explanation about grid-stoppers. So to answer my question, the cap would need to be after the grid-stop, and effectively in parallel with the miller capacitance?

And you're right about not being able to hear because the roll-off is above audio frequencies, but the different I couldn't hear was not with or without the grid-stop altogether, but moving the cap before or after it. I'm not sure if you misunderstood that part of my post, or I'm just missing your point.

I did misunderstand that part. However, it's a bit more complex than that.

To get a low frequency rolloff from a cap to ground, you need the cap to be a shunt to ground after some impedance in series with the signal. A grid stopper resistor fills the bill if you put additional caapcitance after the grid stopper, adding to the grid-cathode and Miller capacitance that is already there.

If you put it in front of the grid stopper, it's still a low frequency rolloff with whatever impedance is before that, as Dave H points out. It's just that with the cap before the grid stopper, the rolloff point is determined by the capacitance you add and whatever the source impedance is from the driving stage, which is usually poorly known. It's no longer strictly controlled by the grid stopper resistance.

It actually gets even worse than that. Using a grid stopper on power tubes gives you one low frequency rolloff. Putting a cap ahead of that to ground causes a second rolloff, with the source impedance of the driving stage, and both of these rolloffs cause phase shifts. If you use feedback around the output tubes to the input of the stage driving the capacitor/grid stopper/capacitor, you have at least two 90 degree phase shifts and probably three, the third being caused by the feedback network itself. Unless these are carefully done and the gain limited, it *will* oscillate. Three lowpass (and also highpass) networks makes enough phase shift to form a phase shift oscillator like the one in the tremolo of many amps. All it takes is enough gain to make it sing.

From the stability standpoint, you want to do all your frequency rolloff in one dominant place to get the gain down at high frequencies before you encounter another rolloff. So it's more forgiving to only use one rolloff, that being the cap after the grid stopper. Deliberately adding one in front of it is adding another chance to oscillate.

I'm simplifying very heavily, because a complete explanation runs for a long time. This gets off into gain-phase feedback network, pole-zero pairs and all the rest of the feedback compensation stuff.

Specifically, I'm wanting to put a small capacitance (220pf) from grid to ground at the power tubes to protect against oscillation in the power stage. I'm using four output tubes (EL34) in push-pull with 5.5k grid-stoppers. If the cap can come before the grid-stoppers, then I only have to use one cap per side, and layout becomes simpler as well. Otherwise, if the caps have to be in parallel with the Miller capacitance, and after the grid-stoppers, I will have to use four caps, one for each tube. Hope that all makes sense, and clarifies my question.

I understand.

There's a simpler way. First of all, do you know you need more capacitance? Generally something up to a few K in an output tube grid will kill off the RF gain while not interfering in audio, and that lets you set the frequency response and stability with the feedback network; that's a lot simpler and more straightforward. Second, if you do need a lower rolloff point on each power tube, you can get the same effect by raising the grid stopper resistance, instead of adding more capacitance.

The rolloff frequency of a series R and cap to ground is F = 1/(2*pi*R*C), where pi = 3.14159... and it calculates like F = 1/(2*3.14*5.1K*220pF) = 141.8kHz; this is what you'd get if you used a 5.1K and a 220pF cap. That ignores the tube capacitances. The grid-cathode and Miller capacitances will increase the effective value of the "220pF" and lower the frequency of rolloff. Doubling the resistor halves the rolloff frequency and does it without introducing more phase shift.

It is in general a bad idea to use extra grid capacitance for stopping oscillation in the audio band. It is better to make the output tubes go well above audio, control the gain/phase response with a single compensation cap either on the phase inverter, in the feedback network, or in the transformer implicitly. Grid stoppers are useful for killing oscillations at RF, where the "black magic" stuff takes hold. Adding more low frequency rolloffs down in the audio band can make oscillations worse.

... unless you've tried everything else and can't get it to work right any other way. That does happen.

[Gaz]

Doug, driving the power amp is a typical long-tail pair, and I believe the output impedance is the plate resistors in parallel, so about 50k. The low pass filter frequency in that case is about 16khz. with the 220pf caps after the 5.6k grid stoppers the frequency is increased to 142k! It still makes sense I can hear a big difference between the too - even 16khz is up there for guitar. Someone let me know if my calculations are off for the corner frequencies, or there's something I'm not taking into account.

R.G. thanks for the further explanation, it was especially enlightening to hear how phase shifts play a factor in the equation - I was completely unaware of that. In my design I would like to include a variable feedback control around a typical 5F6A Bassman/Plexi 100-watt power amp, and just want to make sure the amp stays stable when the most amount of feedback is applied. I'd like to vary the NFB from about 56k off the 8ohm tap to 250k. As you can see I don't understand the mathematics involved, but I do understand that too much NFB can cause instability, which is why I'd like to introduce a pretty good low pass filter somewhere in the feedback loop. Thanks again for the help!

[R.G.]

R.G. thanks for the further explanation, it was especially enlightening to hear how phase shifts play a factor in the equation - I was completely unaware of that. In my design I would like to include a variable feedback control around a typical 5F6A Bassman/Plexi 100-watt power amp, and just want to make sure the amp stays stable when the most amount of feedback is applied. I'd like to vary the NFB from about 56k off the 8ohm tap to 250k. As you can see I don't understand the mathematics involved, but I do understand that too much NFB can cause instability, which is why I'd like to introduce a pretty good low pass filter somewhere in the feedback loop. Thanks again for the help!

You're very welcome. The detailed math is a PITA, but the results are easy to remember.

Let's say you have an amplifier with no feedback with a gain of A times. So put in 1V of signal, get out A volts of signal. If you then feed back the whole output to the input, you get 1V => A volts => A*A volts => A*A*A, and the whole thing runs a way with itself because it provides more input for itself than it needs to run, so the input becomes inconsequential. It can provide its own input and ignores whatever you want it to do.

If you put a divider in the feedback network so you only feed back a fraction of the input, as long as the divider ratio cuts the fraction fed back to the input by at least the gain of the amplifier, then it can't make enough fed back signal to keep itself going, so it can't oscillate. The "loop gain" is the gain from input, through the amp, through the feedback network, and back to the input. If the loop gain is less than one, no self sustaining oscillation is possible.

That happy circumstance happens two ways: (1) when the feedback network attenuates the feedback signal from the output by more than the forward gain, so that if the feedback network is a 1/A divider, no oscillation happens; or (2) if the forward gain is inverting, so that the fed-back signal always opposes the input.

And this is where the phase shift comes in. It's impossible to build amplifiers where there are no capacitive or inductive components to change phase and gain; even the wires themselves have self-capacitance and inductance, so eventually, all negative gain amplifiers accumulate enough phase shifts to get turned around from a phase of 180 degrees to a gain of 360 degrees. It's impossible for this to NOT happen in the real world. Phase shift makes inverting negative feedback amplifiers into positive feedback amplifiers.

The saving grace is that phase shift components can also make gain change with frequency. So if you pick ONE phase shift part, and make that start reducing gain at a much lower frequency than the other phase shifts, it is possible to make that one phase shift part duck the loop gain under unity before the other phase shift stuff gets into operation and shifts you into oscillation. Picking one of the phase shifts and deliberately making it drop the gain under unity before phase shifts build up is called "dominant pole compensation".

One low frequency rolloff from an R-C or R-L network can contribute a maximum of 90 degrees of phase shift. Each other R-C or R-L adds more shift. An L-C can put in 180 degrees all by itself. Realistically, it takes an infinite amount of frequency to get to a full 90 degrees per R-C, so in practice, any amplifier with only one or two lowpass rolloffs is unconditionally stable, so long as the ones you know about overwhelm the ultra-RF funny stuff that comes in eventually. But with as few as three RC lowpass sections in any amplifier with feedback around it, there is enough phase shift to have oscillation. All that is needed to make it sing is enough forward gain.

That's why I was saying not to introduce yet another R-C rolloff ahead of the grid stoppers. Never add more phase shift than you have to.

[Gaz]

Brilliant! you've answered my initial question at a fundamental level, and in plain english to boot. Thread closed!! Thanks for your time, R.G.

[Dave H]

Doug, driving the power amp is a typical long-tail pair, and I believe the output impedance is the plate resistors in parallel, so about 50k. The low pass filter frequency in that case is about 16khz. with the 220pf caps after the 5.6k grid stoppers the frequency is increased to 142k! It still makes sense I can hear a big difference between the too - even 16khz is up there for guitar. Someone let me know if my calculations are off for the corner frequencies, or there's something I'm not taking into account.

It won't be 142k with the cap after the 5.6k grid stopper. The impedance you need to use in the calculation in this case is not 5.6k but the output impedance of the driving stage (50k) + 5.6k or 56.6k which will make the frequency about 14.5k not 142k.

Dave H.

[R.G.]

It won't be 142k with the cap after the 5.6k grid stopper. The impedance you need to use in the calculation in this case is not 5.6k but the output impedance of the driving stage (50k) + 5.6k or 56.6k which will make the frequency about 14.5k not 142k.

If you're driving it from a 12AX7 plate before the grid stopper, you're absolutely correct. The plate impedance adds directly to the grid stopper for the bulk effect. I was (misleadingly, I see now!) doing the modelling thing where a voltage source drives the network being considered. Good catch!

Grid stoppers really shine where there is local capacitive feedback from the anode or cathode of the power tube into the cables between the plate of the power tube or another tube nearby, or when the wires pick up radiated RF. This is why you don't want any distance between the grid lug and the grid stopper resistor. But for overall high frequency rolloff, yes, the plate impedance of the driving plate is added to the resistance of the grid stopper.

[Gaz]

Thanks for the clarification, Doug, it makes perfect sense to me now that the resistances would just be added in series.

[txstrat]

A grid stopper works because there is some... ...one circuit/setup. YMMV.

R.G. Thanks for not reproving or even slapping me for my benighted post.

[martin manning]

To get a low frequency rolloff from a cap to ground, you need the cap to be a shunt to ground after some impedance in series with the signal.

Interesting discussion. For clarification, a series R followed by a shunt C is usually called a low-pass filter, so I would think that the response would be called a high frequency roll-off, no? Just want to make sure I'm not missing something here...

[R.G.]

R.G. Thanks for not reproving or even slapping me for my benighted post.

I'd say you're welcome, but that would imply that it was an act of kindness. There was no need to be kind.

Your posts were well meaning and correct to the extent of your knowledge. Through my own mania... er, interests! ... I had happened to run onto an extra bit of knowledge. You're still learning, and so am I. I raise my metaphoric glass to you.