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

[nickb]

I am confused. Do you agree that in the normal guitar input stage, the dominant source of noise is the 68K, or 33K, grid stop resistor, and that the noise equivalent resistance of the 12AX7 is more like 1.5K? That is source of the improvement. Yes, it only matters if the source is quiet enough, but that is certainly true with a guitar with a 10 K pot turned down most of the way.

In an rf filter, the shunt element would not be the 1M resistor but rather some capacitance.

Complete agreement. I'm just saying that if you do loose the input resistor on the original design so we are comparing apples to apples, then the noise is performance is better (in the original). The original has twice the gain and the two parallel triodes' noise sources are not correlated so it ends up with approx 2 x 2^0.5 ~= 2.8 times lower noise.

[Mike Sulzer]

Complete agreement. I'm just saying that if you do loose the input resistor on the original design so we are comparing apples to apples, then the noise is performance is better (in the original). The original has twice the gain and the two parallel triodes' noise sources are not correlated so it ends up with approx 2 x 2^0.5 ~= 2.8 times lower noise.

Yes, for sure two tubes are quieter than one if they are put in parallel*, but two tubes are noisier than one if the cathodes are connected together. The question becomes how do you get rid of the grid stop resistor? To be sure of doing this you must eliminate the Miller effect. The apple2apple comparison is the working circuit, not the tubes alone.

To eliminate the Miller effect, eliminate the feedback from plate to grid by grounding the plate and passing the signal to another tube cathode2cathode. Or use what people normally call the "cascode". I have explained why I like the circuit I used in he original post.



*But why would you want to put two tubes in parallel just so you can use half the value of grid stop resistor? You do not get the advantage of the quieter compound tube, just the reduced resistor. Why not eliminate the grid stop resistor, getting 3.4K rather than 33K + .75K, or 68K + 1.5K if you use just one tube?

[nickb]

Yes, for sure two tubes are quieter than one if they are put in parallel*, but two tubes are noisier than one if the cathodes are connected together. The question becomes how do you get rid of the grid stop resistor? To be sure of doing this you must eliminate the Miller effect. The apple2apple comparison is the working circuit, not the tubes alone.

To eliminate the Miller effect, eliminate the feedback from plate to grid by grounding the plate and passing the signal to another tube cathode2cathode. Or use what people normally call the "cascode". I have explained why I like the circuit I used in he original post.



*But why would you want to put two tubes in parallel just so you can use half the value of grid stop resistor? You do not get the advantage of the quieter compound tube, just the reduced resistor. Why not eliminate the grid stop resistor, getting 3.4K rather than 33K + .75K, or 68K + 1.5K if you use just one tube?

OK. I think I see where you are coming from. As I see it, the primary purpose of the grid resistor in this position is not as a grid stop in the sense of self HF oscillation prevention but as a 20KHz+ suppressor. By eliminating the Miller effect you solved the stability problem but lost the suppression. That's why the grid resistor should not be eliminated. In other words, the Miller effect works to our advantage by allowing us to use a lower series resistance and so lowers noise.

If we were to use an RC LPF in the input it would be a bit of a compromise between noise, suppression and loading on the pickups, although the latter may not be a problem with lower Z pickup designs.

The two tubes in parallel was the original circuit.

[Mike Sulzer]

OK. That's why the grid resistor should not be eliminated. In other words, the Miller effect works to our advantage by allowing us to use a lower series resistance and so lowers noise.


The two tubes in parallel was the original circuit.

But the grid resistor must be eliminated if you want a quiet front end. The stability problem is easily solved as this design shows. True, I have not used an RF suppression filter, but one with a 33K or 68K noise resistance is neither necessary or in any way an acceptable engineering solution. You can do a lot better than that! It only sort of works for guitar pickups because they have such high output. But if you want to use the guitar volume control over a wide dynamic range, you must first make the guitar circuitry better, and then make the amp better.

Yes, two tubes in parallel was the original circuit. I just cannot see anything good about that design with such a large grid resistor: a 16.5 db degradation in potential SNR while attempting to reduce noise by putting tubes in parallel!

[Malcolm Irving]

It’s great to see such an original design, and the discussion in the thread is very interesting.

But I don’t understand why eliminating the Miller effect would mean that no grid stopper is required. Doesn’t it work the other way around? If you eliminate the Miller effect then the input capacitance is very small and you need a bigger grid stopper resistance to get the same HF cut-off?
(Maybe I’m missing something here – it wouldn’t be the first time! )

Even with a conventional input stage you could get away with no grid stopper, if you are able to very carefully shield the input signal.

[Richard]

Thanks for the design analysis.

[Mike Sulzer]

It’s great to see such an original design, and the discussion in the thread is very interesting.

But I don’t understand why eliminating the Miller effect would mean that no grid stopper is required. Doesn’t it work the other way around? If you eliminate the Miller effect then the input capacitance is very small and you need a bigger grid stopper resistance to get the same HF cut-off?
(Maybe I’m missing something here – it wouldn’t be the first time! )

Even with a conventional input stage you could get away with no grid stopper, if you are able to very carefully shield the input signal.

Yes, what I have been saying is not right. Perhaps it would be better to say: 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 do not think as a practical matter that you could shield the input signal and necessarily be able to eliminate the grid stopper on the input stage. At least I cannot do it reliably. I do not think that shielding the input signal assures that you can completely eliminate signals getting onto the grid of the tube, and so I think you need to do something else as well. AC grounding the plate does it, just as the analogous technique using source coupled pairs and ac grounding the left drain works with JFETS. I think things are more critical with the tube, perhaps because it is more open to external fields.

[Mike Sulzer]

"By eliminating the Miller effect you solved the stability problem but lost the suppression."

Eliminating RF from audio circuits does to necessarily work that way. This: http://audiosystemsgroup.com/SAC0305Ferrites.pdf discusses the issue and solutions.

[Mike Sulzer]

Very interesting design, could you please provide the power supply voltages for the various stages?

Voltages in red.



Are people having trouble seeing this? In my browser, I just double click on it to put it into a separate large window (actually a tab) and click on it to zoom to even larger.

[jazbo8]

Voltages in red.
Are people having trouble seeing this? In my browser, I just double click on it to put it into a separate large window (actually a tab) and click on it to zoom to even larger.

Thank you for updating the schematic, I have no trouble seeing it in Google Chrome.

The original stage gives the ENIV 2.67uV 100Hz-10KHz, gain 68.5 and -3dB at 7KHz

Is the 7kHz a typo? Seems awefully low...

[nickb]

Thank you for updating the schematic, I have no trouble seeing it in Google Chrome.


Is the 7kHz a typo? Seems awefully low...

Yes, indeed, it's closer to to 26KHz - thanks A bit of funger trible on the kalcalatr!

[nickb]

"By eliminating the Miller effect you solved the stability problem but lost the suppression."


Eliminating RF from audio circuits does to necessarily work that way. This: http://audiosystemsgroup.com/SAC0305Ferrites.pdf discusses the issue and solutions.

If we take it that this method is suitable then lets's apply it to both topologies.


The problem I have is we are not comparing like with like. We have eliminated the large series grid resistor and then claimed low noise performance due to the topology. I don't think that is fair.


Hopefully we can agree that it would be reasonable to compare what is above with the original design but with any ultrasonic oscillation suppressed. Low gm triodes are not particularly susceptible to this and so just a few hundred ohms (or perhaps even a ferrite) will be sufficient. Let's be excessive and say 1K ohm.


The ENIR will be about 1.6k/2*1.4 + 1k = 2.12k whereas this topology is around 3.2k. If we can use a ferrite for parasitic suppression, the performance would be even better. The point is that this design is noisier than the original as far as I can see ( which generally is not very far ).

On a side note, I was using a simulator to derive the ENIV's earlier but I've found a case where I'm getting silly answers so I need to understand what is going on before presenting those as reliable.

On the side, of the side-note, I've suddenly stopped getting email notifications for this thread even though I'm still subscribed.

[Mike Sulzer]

If we take it that this method is suitable then lets's apply it to both topologies.


The problem I have is we are not comparing like with like. We have eliminated the large series grid resistor and then claimed low noise performance due to the topology. I don't think that is fair.


Hopefully we can agree that it would be reasonable to compare what is above with the original design but with any ultrasonic oscillation suppressed. Low gm triodes are not particularly susceptible to this and so just a few hundred ohms (or perhaps even a ferrite) will be sufficient. Let's be excessive and say 1K ohm.


The ENIR will be about 1.6k/2*1.4 + 1k = 2.12k whereas this topology is around 3.2k. If we can use a ferrite for parasitic suppression, the performance would be even better. The point is that this design is noisier than the original as far as I can see ( which generally is not very far ).

On a side note, I was using a simulator to derive the ENIV's earlier but I've found a case where I'm getting silly answers so I need to understand what is going on before presenting those as reliable.

On the side, of the side-note, I've suddenly stopped getting email notifications for this thread even though I'm still subscribed.

In the first post, I said that this design had more tube noise than the original, but got rid of the noise from the resistor, and so had a lot less noise over all. Perhaps you can get rid of the oscillation with a ferrite; Carvin apparently does, but what I know is that I can get rid of it with this design, and that the noise is so much lower than the original, it might not be worth worrying about. Everyone has been accepting really bad noise performance for decades; i am not worried about something that is much better but not quite as good as you can do.

[jazbo8]

On a side note, I was using a simulator to derive the ENIV's earlier...

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

[mhuss]

All this theory is very interesting , but how does it sound?