Why some amps use two triode in parallel?

[Alan0354]

Parallel does reduce noise to a certain degree and have other virtues. I never argue about that. Some people just insist on getting 3dB noise reduction, that is absolutely false. That's the whole point I am driving at.

In noise reduction design, anything happen in the plate region is much less critical, reducing the plate resistance only give minor improvement. The major problem is the input stop resistor, the grid current. Stop resistor does not change, grid current likely double. This alone increases the noise regardless of the gain increase of the parallel triodes.

People need to realize NOT EVERY NOISE SOURCE IS REDUCED WHEN THE GAIN GOES UP, THAT IS JUST WRONG. The above two major noise source is gain independent already!!!

Only flicker noise in the plate circuit get reduced if the gain goes up, any flicker noise source in the grid circuit is.......AGAIN, GAIN INDEPENDENT.

People really need to read noise calculation. Again, read post #13, #18 and #19!!!! Don't want to take my word, Google noise!!! Hundreds of articles showing you how to model and calculating noise. None easy, but they pretty much doing the same thing in the article and in post #19. I did not invent this, this is common knowledge, anyone design low noise circuit HAS to know what I am talking about.


For guitar amp,

1) Eliminate the 68K grid stop resistor,

2) Find a preamp tube that has the lowest grid current and low input flicker noise.

3) Put a cap to limit the frequency either at the input or the first stage to limit to 5KHz. You'll be way way ahead of the game.

This, you can take it to the bank. then you can talk about parallel tubes.

[Alan0354]

Just to show my point. Let me just calculate the noise of the grid stop resistor and grid current alone. Both are not gain dependent. I use typical 68K grid stop resistor. I cannot find the input grid current spec of 12AX7, from grid leak bias, let's just ASSUME Ig=1uA for this calculation. Also, I limit the bandwidth BW to 5000Hz using a cap.

The first part is calculation for single triode. As you can see, the 68K grid stop resistor alone generates 2.36uV already!!! The noise current from 1uA grid current is 40pA. All these are RMS value.


The total input noise due to 68K and grid current for a single triode generates 3.6uV RMS, that is 2X3.6X1.414=10.1uV peak to peak.

The total input noise due to 68K and two parallel triodes is 4.5uV RMS=12.8uV peak to peak.

As you can see, the input noise alone for these two contribution alone, the parallel triodes is higher than single triode.

The gain depend part of the noise of the parallel triodes has to make up the difference in order to be lower.

10uV at the input for guitar amp is quite high.

Both noise is related to the 68K grid stop resistor. The first is thermal noise of 2.36uV. Then the next part is the current of the shot noise due to grid current. This times the grid stop 68K gives 2.72uV from each tube. Eliminating this grid stop resistor get rid of all these. The only one left is about 10Kohm of the pickup resistance from the guitar.

Edited: I made a Bobo in the math, it is updated now. The result still stands.

[tubeswell]

To stick with your original question about why some amps use two triodes in parallel, its possible to get a noticeable increase in gain or improvement in bandwidth from paralleling triodes, but you can get more gain from cascading the triodes or lower output impedance by using one of them as a follower. What you do with the spare triode is up to you, and its a pointless discussion trying to 'disprove' the worth of some people choosing to have parallel triodes, because there may be situations where people prefer to use triodes this way and no amount of academic argument is going to change that.

[Alan0354]

To stick with your original question about why some amps use two triodes in parallel, its possible to get a noticeable increase in gain or improvement in bandwidth from paralleling triodes, but you can get more gain from cascading the triodes or lower output impedance by using one of them as a follower. What you do with the spare triode is up to you, and its a pointless discussion trying to 'disprove' the worth of some people choosing to have parallel triodes, because there may be situations where people prefer to use triodes this way and no amount of academic argument is going to change that.

No, all my talk has nothing to do with other strong point of parallel triode. It is just to disprove the narrow assertion of gain 3dB of noise advantage. That's it. In fact, parallel triode even have some advantage in the noise performance too. Just not the 3dB. 3dB is a Big Big deal in noise reduction. We are usually trying to shave off a little dB here and there only!!!

I think the sound of using parallel triode is slightly better, I even put a switch to connect and disconnect the second triode and definitely sounds a little better. I never question the improvement. It dose seems to sound a little different from cathode follower. CF seems to be a little more aggressive and is not necessary better.

[soundmasterg]

Keep in mind that any sound difference between a triode or two triodes in parallel that you are hearing is partially a result of the greater volume you get from the parallel mode. To really hear the difference in sound you need to equalize the volume between the two modes and then listen.

Greg

[markr14850]

With parallel triodes, wouldn't the input capacitance double, implying that the grid stopper should be cut in half to achieve the same roll-off point?

[J M Fahey]

people here seems to believe you have a big noise advantage using two triodes in parallel, that you actually lower the noise power by half.

You seem to find contradiction between those 2 statements, where there is none.
Lowering noise power by half is a BIG noise advantage.
We are talking the *triode* noise here.

If there are other, external noise generators is a different problem, which must be dealt with on its own.

Do you know how Formula 1 races are won?
You don´t make a car which is 100MPH faster than others, because that is very difficult, and others are not fools, they are also up to date with "state of the art".

Any groundbreaking "secret" is soon leaked or rediscovered by others.

But if you change pneumatic grip a little and get 1 MPH extra, then slightly change roof curve to be a little more aerodynamic and gain another MPH, and so on, all apparently despicable gains, in the end your car will go , say, 5MPH faster than others ...... and win the race

Same here: amp elements are many, improving any of them will advance the whole design.

Plus there is something called "synergy", where the whole is more/better than the sum of the parts

[Alan0354]

You seem to find contradiction between those 2 statements, where there is none.
Lowering noise power by half is a BIG noise advantage.
We are talking the *triode* noise here.

If there are other, external noise generators is a different problem, which must be dealt with on its own.

But remember triode noise mostly only generated from interaction with the rest of the circuit. Here is more the complete picture of the triode noise and I separate the ones generated due to external and the ones totally from the triode?

1) The shot noise from grid current do not by itself generates voltage noise. It is only when it goes through the input resistance of the grid stop resistor and the resistance of the pickup in the guitar, then V=IR gives you the noise. From the article you sighted, that is if you assume Rsg=0, then there is no noise generated due to shot noise of grid current.

2) If you assume grid stop resistor resistor and pickup resistance equal to zero. Then there is no noise from it.

3) Flicker noise are the one really generated by the tube and the tube alone. But you still need to divided the flicker noise at the grid and the flicker noise from the plate side. Flicker noise from the grid WILL NOT be reduce if you increase the gain of the stage. ONLY the flicker noise on the plate side will be reduced by the gain of the triode. You can never count on reducing flicker noise by increasing gain.

4) The shot noise from the plate current is from the plate current you set up for the tube. It is not generated by the tube. It's only the noise generated when you bias the tube. So it is not due to the tube.

5) The plate resistor generates thermal noise. This is not from the tube.

6) I don't know how to assume the noise from the internal rp. Let's just say it is from the tube.

You see, out of all the noise that hamper the amp, ONLY #3 and #6 is TERULY from the tube!!! If you insist on only looking at the tube, #3 and #6 can be used!!! and only the flicker noise from the plate will see the advantage of parallel tubes. In fact, the flicker noise at the grid will increase by 3dB when you parallel tubes and it will not be reduced by increase gain due to paralleling.

So in conclusion, yes, you can see more advantage if you isolate and look at the tube alone as you ONLY consider #3 and #6. but still it is not 3dB because the flicker noise from the grid is not divided by the gain. AND also, 3dB is idea situation where you double the gain when paralleling. Because of the internal plate resistance, you never double the gain. So you never get 3dB period.

Also, you cannot NEVER isolate the tube and look at the merit of the tube alone as I have shown the large noise contribution. I will post the calculation of noise contribution due to plate resistance and plate current to give a more complete picture. There is no data on the flicker noise of the tube, this is another big unknown.

[Alan0354]

Keep in mind that any sound difference between a triode or two triodes in parallel that you are hearing is partially a result of the greater volume you get from the parallel mode. To really hear the difference in sound you need to equalize the volume between the two modes and then listen.

Greg

That's the reason of my original post. I want to learn more. But we get stuck on the noise part which is not even of significant from what I see here.

[Alan0354]

With parallel triodes, wouldn't the input capacitance double, implying that the grid stopper should be cut in half to achieve the same roll-off point?

that's the whole point as I conclude in post #46. get rid of the grid stop resistor, you'll are way ahead of the game. I never see the reason for that. I put a 20K on my design just to protect and current limit to the input tube. I sure do not find the noise excessive in high gain mode.

[Alan0354]

Now I calculate two of the gain related noise source namely 1) the plate resistance and 2) the plate current. plate resistance used is 100K and plate current is 1mA. This is the most common value used. Internal plate resistance of tube rp used is typical 60K.




Noise due to output resistance:
For single tube, Output resistance Rout of a single tube=RL//rp=37.5K and thermal noise voltage is 1.75uV RMS. You refer back to the input which is divided by 54=0.032uV.
For parallel tube, Rout =RL//rp//rp=23.1K, noise is 1.37uV. refer to input is 1.37uV divided by 78=0.0177uV.


Noise from 1mA plate current. Noise current is 1.27nA.
For single tube, noise voltage is 1.27nAX37.5K for single tube =47.6uV
Refer to input is 47.6uV divided by 54=0.88uV.
For parallel, refer to input is 0.37uV.


You take this result and compare to the updated input noise calculation for the input part, these are very small and literally disappeared after route mean square.



Conclusion

From these result, you can clearly see the gain dependent part of the noise is ABSOLUTELY insignificant, the noise is DOMINATES by both the thermal noise of the grid stop resistor and the noise of the grid current with the grid stop resistor.

Biggest unknown factor is the flicker noise, where part of it is at the input and is gain INDEPENDENT and part is at the plate side that is gain dependent. From common experience, it is just like all the calculation above, that the gain dependent part is insignificant compare to the input part,.

You see now, the assertion of getting 3dB noise advantage is absolutely WRONG. You really need to go through the detail calculation to see the picture. So far in the picture without Flicker noise, parallel tube is loosing big time. Unless the Flicker noise from the output side dominate the picture, DON'T USE PARALLEL TRIODES for noise reduction, use it because you get better sound only.

[loudthud]

Some quick comments:

1) You can't calculate the peak to peak noise from the RMS noise by multiplying RMS * 2 * 1.414. The conversion factor is about 6.

2) The lowest noise commonly available JFET is the LSK170. Of the two very low noise RIAA preamps I've seen, both used four of them in parallel. One design also used an opamp but the power supplys had to have noise below 100uV. When a PCB was layed out, 2 inches of etch was too long between the regulators and the opamp. That ruined the noise floor. The layout had to be revised. The engineer claims it's the lowest noise and lowest distortion RIAA preamp available.

3) The typical Fender input has two 68K resistors in parallel. I used to use 33K CF but now I'm using a 10K metal film. It makes a difference.

[Alan0354]

Some quick comments:

1) You can't calculate the peak to peak noise from the RMS noise by multiplying RMS * 2 * 1.414. The conversion factor is about 6.

2) The lowest noise commonly available JFET is the LSK170. Of the two very low noise RIAA preamps I've seen, both used four of them in parallel. One design also used an opamp but the power supplys had to have noise below 100uV. When a PCB was layed out, 2 inches of etch was too long between the regulators and the opamp. That ruined the noise floor. The layout had to be revised. The engineer claims it's the lowest noise and lowest distortion RIAA preamp available.

3) The typical Fender input has two 68K resistors in parallel. I used to use 33K CF but now I'm using a 10K metal film. It makes a difference.

Regarding to the FET, you really have to look at the situation case by case. That is really what I am trying to convey. You mention one FET, there are hundreds of low noise FET and BJT. Why? BECAUSE they are all a little different. The application depends on the kind of input impedance you are dealing with.

1) If you have input impedance of hundreds of Kohm or Mohms. You cannot use BJT as they have base current and the base current is going to kill you. You need to use JFET. If the input impedance is way in the Mohm, you need to use MOSFET that has high flicker noise. But because they have low gate current, you win.

2) then in other applications where the input impedance is below say 100ohm. Input current is not important. Then BJT shines.

3) In audio application like here. Flicker noise (1/f) is very dominant, BJT has very low flicker noise compare to JFET or MOSFET. You don't use JFET or MOSFET for the first input stage unless you can justify it.


Point is you can never tell until you spend time to calculate all the noise contribution before you can tell which one is better. Blanket saying parallel two tubes lower noise by 3 dB not only is misleading, it is irresponsible and WRONG. Anyone saying that NEVER work a day in low noise design. AND you can take this to the bank.

As for any particular transistor, you might find it the best in your situation, but it's not for others. Everyone of the low noise transistors are a little different. You want to know why they have hundreds of low noise FETs? Because people find an application that they need a certain characteristics and they want enough of them that they make the semi conductor company make one for them!!!! I can never repeat enough, Low noise design is SO SO SPECIFIC, more often then not you don't know what is the gating factor until you go through all the calculations., that's the reason I do spread sheets for that.

As for peak to peak vs RMS, as long as you compare apple to apple OR orange to orange, It works.

[tubeswell]

Some quick comments:

1) You can't calculate the peak to peak noise from the RMS noise by multiplying RMS * 2 * 1.414. The conversion factor is about 6.

2) The lowest noise commonly available JFET is the LSK170. Of the two very low noise RIAA preamps I've seen, both used four of them in parallel. One design also used an opamp but the power supplys had to have noise below 100uV. When a PCB was layed out, 2 inches of etch was too long between the regulators and the opamp. That ruined the noise floor. The layout had to be revised. The engineer claims it's the lowest noise and lowest distortion RIAA preamp available.

3) The typical Fender input has two 68K resistors in parallel. I used to use 33K CF but now I'm using a 10K metal film. It makes a difference.

Also the input capacitance on 2 parallel 12AX7 triodes is about 240pF, which will affect the roll off.

[Alan0354]

Sorry I don't know what is a RIAA amplifier. I have no idea what is the requirement for it's usage. PCB trace dose not increase or decrease noise. I think you are mixing EM radiation noise with the noise we are talking about. Now you are getting into electromagnetics and antenna. This is a totally different subject from the discussion. I am happy to talk about it if you want as that is even more of my field.