Low Noise high gain amp based on 18W kit: How and why

[nickb]

LTSpice? If so, which tube model are you using?

One, that I now know, doesn't include any noise sources, which explained the unexpected results

[Mike Sulzer]

The cathodyne phase splitter and hum from the power supply.

In a quiet moment (not so many here in the Puerto Rico countryside with all the insects, birds, and what not) I noticed that this amp had more hum than expected with gain 3 (master volume) turned down. It is not really a problem in any but the quietest locations with the listener paying close attention, but it should not have been there. I found that it was due to the cathodyne phase splitter. It turns out that the upper output has essentially no power supply rejection, while the lower has a lot. Thus, there is no cancelation, as there is with the cathode coupled splitter since the hum on the two plates is nearly the same and cancels in the push pull power amp.

Here is why the cathodyne has this asymmetry: Think of it as a cathode follower with an extra resistor inserted in the series with the plate. The cathode follower is a feedback circuit in which the gm of the tube is used to make changes in the grid voltage nearly reproduced on the cathode. The plate current that flows is whatever is necessary to make that happen, and so the plate is a controlled current source. Voltage variations on the plate are rejected: the plate takes the current it needs. Now put a resistor in the plate circuit to get the negative of the cathode voltage variations. Because the plate is a very high impedance, variations in the power supply voltage at the top of this resistor appear on its lower side, attenuated only a bit by the load. Thus, in effect, the power supply ripple on the supply for the cathodyne splitter appears on the grid of one of the power tubes.

In this amp, I got rid of the hum by moving the cathodyne supply from supply A to supply B, much lower in ripple. I also moved the second stage from B to C in order keep these stages on separate supplies.

I will eventually post a new schematic with new voltage measurements.

[Mike Sulzer]

An Additional benefit of the long tail biasing: simplification of grounding issues

In the normal "short tail biased" common cathode stage of gain, the signal and power supply "grounds" are intertwined, and it takes great care to avoid getting unwanted signals in series with the inputs to high gain, low level early stages. Suppose the dc current out of the cathode is handled instead by a perfect current source. Ac gain is enabled by a cathode bypass capacitor, and the low side of this capacitor can be thought of as the positive input of a differential amplifier, albeit not the normal kind we are used to since the negative input, the grid, is high impedance and dc coupled, while the positive input is low impedance and ac coupled. However, the latter is used in this context only as a signal "ground", and low impedance is just fine. This suggests that the wiring can make use of a signal ground bus (that is, separate from the power supply ground bus), where these "positive inputs" are connected together in a chain. This must be connected to the power supply "center" at one point, and it makes sense to connect the negative terminal of the filter capacitor used to supply positive voltage to the lowest level stage to the signal ground bus at this tube.

Can this work in practice? Instead of perfect current sources we have cathode resistors that are very large compared to the input impedance at the cathode. For example in the second stage we have a 330K resistor, and an input impedance of roughly 1/1600e-6 = 625 ohms. The power supply rejection then is (330 + .625)/.625 = 529 or about 54.5 db. This is high enough so that if we do a good job with the filtering on the negative supply, any remaining ripple on it is out of the picture, and we have the condition necessary to have a signal ground bus.

Of course, I did not think of this when wiring the amp, but I have modified the wiring. The signal ground bus starts at the right hand grid of the input stage, runs forward to the first gain control, and then along the pots behind the front panel. The positive inputs of the various stages attach at appropriate points, although after the first two stages, it does not make a lot of difference exactly you do.

On a related matter, I will try what I think is a better way to connect the guitar input jack to the first stage. Rather than use 2 conductor shielded cable as it now is, I think it would be better to use two pieces of single conductor. The one for the low side signal, externally the shield of the guitar cable, has some level of rf on it. Buy running it through a piece of shielded cable with the shield connected to the chassis, the rf is confined, and the capacitance of the cable acts to reflect/attenuate the rf. The rf is kept away from the high side of the signal; that is, coupling to it is much reduced. I will try this soon. Currently, the amp is apparently completely insensitive to digital cell phone transmissions, but it cannot hurt to improve the rejection further.

[Malcolm Irving]

In the normal "short tail biased" common cathode stage of gain, the signal and power supply "grounds" are intertwined, and it takes great care to avoid getting unwanted signals in series with the inputs to high gain, low level early stages. Suppose the dc current out of the cathode is handled instead by a perfect current source. Ac gain is enabled by a cathode bypass capacitor, and the low side of this capacitor can be thought of as the positive input of a differential amplifier, albeit not the normal kind we are used to since the negative input, the grid, is high impedance and dc coupled, while the positive input is low impedance and ac coupled. However, the latter is used in this context only as a signal "ground", and low impedance is just fine. This suggests that the wiring can make use of a signal ground bus (that is, separate from the power supply ground bus), where these "positive inputs" are connected together in a chain. This must be connected to the power supply "center" at one point, and it makes sense to connect the negative terminal of the filter capacitor used to supply positive voltage to the lowest level stage to the signal ground bus at this tube. ...

Would it be correct to say that for the 'long-tail stages' in your design, the DC current from the power supply goes in at B+ and comes out at B-. Only a reference ground is needed for this power supply, in order to hold B+ and B- at the correct voltage levels relative to the signal ground. No significant current is needed in the power supply reference ground.

[Mike Sulzer]

Would it be correct to say that for the 'long-tail stages' in your design, the DC current from the power supply goes in at B+ and comes out at B-. Only a reference ground is needed for this power supply, in order to hold B+ and B- at the correct voltage levels relative to the signal ground. No significant current is needed in the power supply reference ground.

Yes, I think that it is a good way to express it. I would just add that it is the fluctuations in the power supply current that matter since a pure dc current appearing in series with an input would be inaudible. And further, lacking a perfect current source that would have infinite impedance with the terminal away from the tube attached to any point whatsoever, we must provide a supply that a "real world" approximate current source can operate from. Thus the high voltage negative supply and the large value resistors. I have used solid state current sources in some designs, a better approximation in principle, but for a guitar amp where the signal from one stage can "beat to death" the input to the next stage it seems better to stick with a simple approximation, showing that it is good enough to do its job and so not adding to the problem.

[Merlinb]

You don't need to get rid of the input grid stopper to get low noise, just make it smaller than the Fender classic 34k. Use a 1k stopper and it will be as quiet as it's ever going to get. You can add a cap after the stopper to kill some of the highs and RF that have been 'rediscovered'.

[Mike Sulzer]

You don't need to get rid of the input grid stopper to get low noise, just make it smaller than the Fender classic 34k. Use a 1k stopper and it will be as quiet as it's ever going to get. You can add a cap after the stopper to kill some of the highs and RF that have been 'rediscovered'.

If you use a 1K resistor, it adds almost as much noise as a 12AX7 does, which is about 1.5K. (The two tube circuit only adds another 0.5K compared to using a 1K resistor.) 1K might be enough to stop oscillation in some cases, but it is not in general.

In any case, ac grounding the plate is a more effective way of preventing oscillation than Rs and Cs at the input. The circuit used here is one of the two topologies that have been called cascode at one time or another, although is not the one people normally use, or think of by that name. But it does allow actually ac grounding the plate.

This is a very high gain amp, but adjustable to be a clean amp as well. Designs such as this work out best when they begin with circuit topologies that favor success.

[Malcolm Irving]

... when you eliminate the Miller effect by ac grounding the plate, you also do a better job of shielding the grid from other effects....
... I think things are more critical with the tube, perhaps because it is more open to external fields.

Are you just thinking of the shielding effect of the grid being physically inside the plate, here? If so, would a grounded can over the tube in a conventional stage do a similar job? Or is there more to it than that?

[Mike Sulzer]

Are you just thinking of the shielding effect of the grid being physically inside the plate, here? If so, would a grounded can over the tube in a conventional stage do a similar job? Or is there more to it than that?

A tube shield does not do the same thing; I do not know why. Of course, that shield does not protect you from the other tube in the glass, and the two tube circuit both makes that other tube part of the solution and also means it cannot be otherwise part of the problem. So you get less information than you would like when you build this circuit and say, aha, it works pretty well.

[Merlinb]

If you use a 1K resistor, it adds almost as much noise as a 12AX7 does,

That's sort of the point -as long as the resistor generates less noise than the tube, it will hardly affect the total. A 12AX7 generates about 1uV EIN, whereas a 1k generates 0.6uV. The total is which is about 1.17uV, only 1.4dB worse than the tube alone (the situation is actually even better n a guitar amp, owing to its limited high frequency bandwidth).That's quieter than adding another tube in the form of a cathode coupled amp. Not that you'd notice the difference between these minute noise levels in this situation, since residual heater hum from the unregulated heater supply will dominate.

1K might be enough to stop oscillation in some cases, but it is not in general.

1k is safe with a 12AX7, especially if you add an extra cap after the resistor.

In any case, ac grounding the plate is a more effective way of preventing oscillation than Rs and Cs at the input.

Actually the opposite is true. A cathode follower is more likely to oscillate when driving stray capacitive loads. There are only capacitive loads at this point in the circuit. Not that it's likely to be a problem in either case in this situation, so it's purely academic.

[Mike Sulzer]

That's sort of the point -as long as the resistor generates less noise than the tube, it will hardly affect the total. A 12AX7 generates about 1uV EIN, whereas a 1k generates 0.6uV. The total is which is about 1.17uV, only 1.4dB worse than the tube alone (the situation is actually even better n a guitar amp, owing to its limited high frequency bandwidth).That's quieter than adding another tube in the form of a cathode coupled amp. Not that you'd notice the difference between these minute noise levels in this situation, since residual heater hum from the unregulated heater supply will dominate.


1k is safe with a 12AX7, especially if you add an extra cap after the resistor.


Actually the opposite is true. A cathode follower is more likely to oscillate when driving stray capacitive loads. There are only capacitive loads at this point in the circuit. Not that it's likely to be a problem in either case in this situation, so it's purely academic.

I do not agree with your numbers; so let's consider the simple case first. The noise resistance of a triode is 2.5/gm. For a 12AX7 this is 1.562K. If you have two cathodes in series as this differential input does, you have twice the noise resistance, that is, twice the power, or 3db, on an increased noise voltage of the square root of two. If you put a 1K resistor in series with 1.562K, the power increase is somewhat less than a factor of two. I get 2.15 db. Yes, the 1K resistor is less, but not as much as you say..

I do not see what the bandwidth of the amp has to do with this. The ratios still work out the same.

This amplifier has about 25mv of ripple on the filament supply for the preamp tubes. I can hear some hum from the input stage; I doubt that it is all from the filament supply, but in any case the tube noise is just as audible as the hum.

Stability: this cathode follower, or the left tube of a differential pair, operates into another cathode. This is a stable situation.

I do not see how you can claim that a 1K stop resistor is sufficient in a high gain tube amp. It just is not in my experience. We are not dealing with just one stage, but the whole amp. The impedance of the guitar and cable are factors as well. The circuit I am using cuts down on the influence of these things.

[Merlinb]

The noise resistance of a triode is 2.5/gm.

Ah, I see the problem. This is common mistake. Tube noise only approximates to 2.5/gm at radio frequencies. At audio frequencies, flicker noise dominates. This raises the equivalent input noise of good triode (20-20kHz BW) to around 1uV, though this is quite variable between individual samples of course. Many 12AX7s will actually be a good bit noisier than this. A few will be quieter.

I do not see what the bandwidth of the amp has to do with this.

Because the grid stopper only generates white noise, whereas the tube generates a lot of flicker (pink) noise. Reducing the high frequency bandwidth reduces the white noise, but has little effect on the (low frequency) pink noise. i.e. the BW has a greater reducing effect on the resistor noise than on the tube noise.

I do not see how you can claim that a 1K stop resistor is sufficient in a high gain tube amp. It just is not in my experience.

It is in mine...

[nickb]

I do not agree with your numbers; so let's consider the simple case first. The noise resistance of a triode is 2.5/gm. For a 12AX7 this is 1.562K. If you have two cathodes in series as this differential input does, you have twice the noise resistance, that is, twice the power, or 3db, on an increased noise voltage of the square root of two. If you put a 1K resistor in series with 1.562K, the power increase is somewhat less than a factor of two. I get 2.15 db.

Setting the flicker noise aside for the moment, since these are uncorrelated noise sources, shouldn't we be calculating this as the square root of the sum of the squares of the noise voltages?

i.e. increase due to 1k resistor is 20*log10(sqrt(1 + 1.5625^2 + 1.5625^2)/sqrt(1.5625^2 + 1.5625^2)) ~=0.8dB


PS: Merlin must to too humble to mention but he did present a very practically useful paper "Noise in Triodes with Particular Reference toPhono Preamplifiers"

[Mike Sulzer]

Setting the flicker noise aside for the moment, since these are uncorrelated noise sources, shouldn't we be calculating this as the square root of the sum of the squares of the noise voltages?

i.e. increase due to 1k resistor is 20*log10(sqrt(1 + 1.5625^2 + 1/5625^2)/sqrt(1.5625^2 + 1/5625^2)) ~=1.5dB


PS: Merlin must to too humble to mention but he did present a very practically useful paper "Noise in Triodes with Particular Reference toPhono Preamplifiers"

You are putting noise resistances, which are proportional to power, into an equation which needs voltages.

[Mike Sulzer]

Ah, I see the problem. This is common mistake. Tube noise only approximates to 2.5/gm at radio frequencies. At audio frequencies, flicker noise dominates. This raises the equivalent input noise of good triode (20-20kHz BW) to around 1uV, though this is quite variable between individual samples of course. Many 12AX7s will actually be a good bit noisier than this. A few will be quieter.


Because the grid stopper only generates white noise, whereas the tube generates a lot of flicker (pink) noise. Reducing the high frequency bandwidth reduces the white noise, but has little effect on the (low frequency) pink noise. i.e. the BW has a greater reducing effect on the resistor noise than on the tube noise.


It is in mine...

Well, I am not so sure it is a mistake, common or otherwise. The 1 micro volt level is too high for a high gain guitar amp because the low frequency cutoff is a lot higher than 20 Hz, maybe 200 Hz or even higher. Since this is 1/f noise, knocking off the bottom decade makes a difference. Other effects tend to reduce the relative importance of low frequency noise. For example, the speaker lacks bass, and human hearing peaks up in the low KHz range. When I measure noise at about 2KHz, I get about 3.4K while the no flicker noise equation predicts about 3.1K. This is not a huge difference, and my conclusion is that 1/f noise is not the dominant audible effect in a high gain tube guitar amp. So I tend to ignore it, and forget I am doing so. Maybe that is a mistake, but I do not think so.