Pentode Noise Tests

[DaveP]

After the thread on the Lorenz mic pre recently, I decided to try the circuit out which got me thinking about the whole subject of pentode noise.

I did some testing but none of it seemed to correlate with the standard formulae on the subject.  After a lot of research and reading, I began to realise that the pentode and triode noise issue was all about radio reception back in the day, then after that it was television.  They needed valves that could be used for VHF and UHF frequencies without too much noise, hence the frame grid types.

HiFi sales were very low compared to radio and TV sets in the 50's and 60's so hardly anyone was working on low noise audio tubes back then, you had the EF86 and EF804 and the 5879 in the US and that's it really, by the time the HIFi market started in earnest, transistors were all the rage.

On this basis, I decided to disregard the standard literature and try to find out for myself.  I prepared a test with the minimum variables, so only the noise changes would be picked up.  I set the HT B+ as 250V exactly.  I chose the Philips 220k/1Meg/2.2k
as the  Ra/Rg2/Rk set up as a ratio standard, this makes Rg2 4.545 times Ra.  The grid resistor was 100k.

I used 100k, 150k, 180k, 220k, 270k, 330k, 390k, and 470k  as the plate/anode resistors and made all the g2 resistors exactly 4.545 times these.  The Rk resistors were all 1/100 th of the Ra in the test,i.e. 1k 1.5k etc, etc.  This would mean that the vk voltage would be fairly constant and would not give a boost as the current changed.

I found the gain of each circuit and used it to calculate the gm for each value of Ra.  The same tube was used throughout.  I calculated RL as the parallel of Ra//Rgf=1M//ra=2.5M.  gm is then RL /gain.

I sited the power supply over a metre away and just used an incandescent bulb for light to minimise noise radiation.  It was only necessary for these test to be relative measurements, not absolute figures.

Here are my results, they are completely contrary to the literature, but in agreement with the commercial circuits of the time.

I simply measured the noise with a good DMM  (a few mV) then divided this figure by the gain to get the uV of noise on the grid.

First of all the Gain chart.

This is how the noise decreases as Ra is increased (up to a point)

This is how the noise decreases with Ig2

This is how the noise decreases with gm.  These are "Hockey Stick" graphs similar to tests made by Merlin Blencoe on triodes.

Finally, for reference here is Ra versus gm for an EF86.

These are all real world figures that anyone can reproduce and probably exceed if they enclose everything in a screened box.

Conclusions:  Noise decreases as you increase Ra and reduce the operating current, then it starts to increase again at very high values of Ra (that are hardly ever used) I don't yet know what that mechanism is.  I believe that current through the heater and cathode causes the thermal noise from the number of electrons in transit, so reducing the current reduces the noise. 

The standard Req = 2.5/gm does not apply to pentodes as I have shown that noise reduces along with the gm and the current.

I would be interested to hear your comments, I  will certainly no longer be worrying about partition noise, I have always liked pentodes, maybe this is why!

best
DaveP

 

[ruffrecords]

Hi Dave,

Good to see someone else doing fundamental noise work. It is a pity you did not short the grid to 0V for the test because the 100K resistor contributes over 5uV of noise at the input.

Cheers

Ian

 

[DaveP]

It is a pity you did not short the grid to 0V for the test because the 100K resistor contributes over 5uV of noise at the input.

Of course you are right Ian, but I wanted to know the noise in a usable circuit.

From what you have said, I've just realised that the very high Ra resistances of 390k and 470k used in the test where the noise starts to increase again, may be due to resistance noise in these components.

Best
DaveP

 

[ruffrecords]

Of course you are right Ian, but I wanted to know the noise in a usable circuit.

Tubes (of all types) are relatively noisy. The only way to obtain reasonable noise performance from them for mic pres is to precede them by a 1:10 transformer. The noise factor of the transformer will then dominate and this can be as low as 1dB. Thetypical 150 ohm microphone source reflected to the secondary will 'look like'  about 15K. The secondary winding resistance is often a couple of K so a more realistic value for your resistor for a usable circuit would be just under 20K. The noise due to 20K in a 20K bandwidth is about  2.5uV. A good triode tube stage will achieve an input self noise in the same ball park or less.

To remove the 100K from your figures to get to the actual tube noise you need to undo the summing of the tube and resistor noise. So if the measured noise is 8uV and the resistor noise is 5uV we have:

5 x 5 +tube noise squared = 8 x 8.

So tube noise squared is 64 - 25 = 39 so tube noise is just over 6uV

This is about double what a good triode will achieve. I have some measurements I made a long time ago of triodes. I will see if I can dig them out.

From what you have said, I've just realised that the very high Ra resistances of 390k and 470k used in the test where the noise starts to increase again, may be due to resistance noise in these components.

Best
DaveP

Usually by the time you reach the plate circuit, the noise contribution of passive components is 2nd order.. The noise in a 470K resistor is about 12uV. There may also be some excess noise due to the HT voltage across the plate resistor (this tends to be worst in old fashioned carbon resistors) but it is unlikely to be greater than the Johnson noise. Lets say the noise in the 470K is 24uV. Even with a stage gain of only 100, this is just 0.12 uV when referred to the input so definitely second order.

By the way, my handy rough calculator for resistor Johnson noise is square root of (resistor value divided by 3096) micro volts.

Cheers

Ian

 

[abbey road d enfer]

To remove the 100K from your figures to get to the actual tube noise you need to undo the summing of the tube and resistor noise. So if the measured noise is 8uV and the resistor noise is 5uV we have:

5 x 5 +tube noise squared = 8 x 8.

So tube noise squared is 64 - 25 = 39 so tube noise is just over 6uV

This is about double what a good triode will achieve.

You've just neglected one factor: noise measured by Dave is not limited to 0-20kHz. Measuring noise with a DMM is not correct, unless an appropriate filter is applied.

 

[abbey road d enfer]

I used 100k, 150k, 180k, 220k, 270k, 330k, 390k, and 470k  as the plate/anode resistors and made all the g2 resistors exactly 4.545 times these. 

Have you plotted the Vg2 variations (or Ig2)?

Here are my results, they are completely contrary to the literature, but in agreement with the commercial circuits of the time.

They don't really agree with the "German" trend of sub-mA current, though.

I simply measured the noise with a good DMM  (a few mV) then divided this figure by the gain to get the uV of noise on the grid.

As I mentioned earlier, measuring with an instrument of unspecified bandwidth allows comparison, but does not lend itself to confronting to academic formulae.

This is how the noise decreases with Ig2

I would think the X-axis is in mA, not uV...? It woud be interesting to graph noise vs. Ig2 with constant Ia. I'm not sure that partition noise is a direct function of Ig2...

Conclusions:  Noise decreases as you increase Ra and reduce the operating current, then it starts to increase again at very high values of Ra (that are hardly ever used) I don't yet know what that mechanism is.

The mechanism is pretty well described in chapter 5 of the AES article.  At very low current, noise increases because Gm is low. Then as current increases noise decreases as long as Gm increases faster than flicker noise, but there is a point where flicker noise takes over.

I  will certainly no longer be worrying about partition noise, I have always liked pentodes, maybe this is why!

I may agree when you produce the same tests with triodes and sompe kind of psophometric filter! 

 

[ruffrecords]

You've just neglected one factor: noise measured by Dave is not limited to 0-20kHz. Measuring noise with a DMM is not correct, unless an appropriate filter is applied.

You are correct, I assumed a 20KHz bandwidth. However, a little research shows that most 'good DMMs' (e.g Fluke) have a 20KHz bandwidth and the not so good ones have a much lesser bandwidth. Of equal concern is whether the meter reads true rms or whether it is an average reading mechanism calibrated for sine waves. This can make a significant difference in the readings of high crest factor signals like noise.

Cheers

Ian

 

[abbey road d enfer]

You are correct, I assumed a 20KHz bandwidth. However, a little research shows that most 'good DMMs' (e.g Fluke) have a 20KHz bandwidth and the not so good ones have a much lesser bandwidth.

Frequency response of DMM's is one of the most unpublished data (cf e.g. Fluke spec sheets). In fact, there is no consistency; it looks like Tektronix are good up to 100kHz (-0.5dB) and Fluke only up to a few kHz.
Googling "digital multimeter frequency response" shows it's all over the place.
100kHz BW for measuring noise is definitely inappropriate; 7dB pessimistic.

BTW, 20 kHz is for a brickwall response; equivalent to a 15.7kHz 1st-order low-pass, assuming pink-ish noise.

 

[DaveP]

Have you plotted the Vg2 variations (or Ig2)?

These are the charts you are asking about:-
Current drawn by  anode and g2 versus Ra.
The Rg2 resistance was made to be 4.545 times Ra so as to be a constant.

The ratio between the Ia and the Ig2

Ra versus Va and Vg2 The meter readings were checked 3 times (Meter 10M/V)

My Meter is not Fluke but it was not cheap.  It does measure RMS up to about 10kHz then it slowly loses sensitivity.
But this is irrelevant because my intention was to take relative measurements, I don't have the equipment, time or inclination to produce a scientific paper on the subject.

They don't really agree with the "German" trend of sub-mA current, though.

I don't understand this statement, nearly all the measurements are sub mA.

The mechanism is pretty well described in chapter 5 of the AES article.  At very low current, noise increases because Gm is low. Then as current increases noise decreases as long as Gm increases faster than flicker noise, but there is a point where flicker noise takes over

That is what the literature says, but for this EF86, the opposite is happening, the noise is decreasing as gm and current decreases.

DaveP

 

[abbey road d enfer]

I don't understand this statement, nearly all the measurements are sub mA.

You're right, I completely misread the 4th graph.

That is what the literature says, but for this EF86, the opposite is happening, the noise is decreasing as gm and current decreases.

One can see that noise increases when Ra>330k. Probably because the non-variable noise sources take over, among which the Johnson noise of the 100k grid res.

 

[merlin]

Of course you are right Ian, but I wanted to know the noise in a usable circuit.

In a 'usable circuit' the source impedance (presumed low) will shunt the grid resistor. You really do need to short the grid if you want useful noise measurements of tubes.

 

[DaveP]

In a 'usable circuit' the source impedance (presumed low) will shunt the grid resistor. You really do need to short the grid if you want useful noise measurements of tubes.

Yes, I take your point, but this is where I'm coming from.  The Mullard /Philips data sheet gives the well known typical noise figure of 2uV  with the tube set up with 100K Ra and 390k g2,  My test has shown me that this is not the best arrangement for low noise,  much higher values are better.  Now it may be that other factors like frequency response are much worse with higher resistance values, but I now know that noise is definitely not.  Now I know why the V72/V76 has 200k Ra and the V-241 has 390K Ra.

Thank you for your triode noise work in the A.E.S paper, it was partly that and a post on DIY audio by Bondini in Australia that gave me the incentive to do this test.  http://www.diyaudio.com/forums/tubes-valves/235070-pentode-noise.html

I am doing the research to make a top class mic-pre, so this knowledge is absolutely vital for me.

Best
DaveP

 

[abbey road d enfer]

The Mullard /Philips data sheet gives the well known typical noise figure of 2uV  with the tube set up with 100K Ra and 390k g2,  My test has shown me that this is not the best arrangement for low noise,  much higher values are better.

I think there's a misunderstanding there, or misformulation; Merlin is talking about the impedance of the source that drives the input grid. For the input tube in a mic pre, that impedance is about 20kohms, or 2.5uV of Johnson noise @20kHz BW. 100kohms generates 5.6uV. That is not negligible.
The figure in the datasheet is to be taken with a lot of caution; it says it's for a BW of 25 to 10kHz, which implies more than that for a 20kHz BW; how much more is difficult to tell, since this figure includes all sorts of noises, in particular heater noise, which is invariable, and Johnson noise, which is BW-dependant.
The 2uV figure cannot have been achieved with an unloaded 100k grid resistor, since alone it develops about 4uV of Johnson noise in a 25-10k BW.

 

[DaveP]

I don't seem to be having much luck explaining things, no matter how carefully I use  the language.

1. I will not ever be using a 100k resistor from grid to earth, there will only ever be a Jensen transformer secondary in the design.

2.  I used a 100k resistor as it was a simple constant that cancels out during the tests with the other load resistances.

3.  Using a transformer for a test just makes life difficult (unless it is very well screened) because it amplifies the noise from the environment by about 10.  A simple 100k resistor has, as you say a known thermal voltage of ~5uV which for relative measurements can be disregarded.  I.E. all the measurements are probably 5uV less than I reported.

4. The equivalent noise resistance inherent in an average EF86 is around 25k depending on the bandwidth, this has nothing to do with a grid to earth resistor, it's just a convenient way of expressing the 2uV of noise generated in the tube.

5.  I am not trying to re-write the textbooks, I just believe that they were written for RF work not audio.  If the 2.5/gm formula was true for pentodes, then my graphs would have gone the other way.  They would have shown lower noise with a 100k Ra than with a 220k Ra for example. 

I just want to set up an EF86 to run with the lowest noise and highest gain, if anyone out there has a better idea, then I'm all ears.  No, I will not be triode wiring it

Best
DaveP

 

[ruffrecords]

I don't seem to be having much luck explaining things, no matter how carefully I use  the language.
Best
DaveP

I think we understand you perfectly well and everything you said is true. The only point Merlin and I made was the tests are more useful if made with a zero ohm grid resistor rather than a 100K one. That way what you measure is the noise of the tube and nothing else. That's all.

You are right there is very little work done in this area for audio applications and you Merlin and I are probably the only people to have really tackled this. You are also right that the 2.5/gm formula is derived solely from RF considerations and though much touted by some audio designers, it is largely irrelevant for audio frequencies. 1/f (flicker) noise is a significant factor in audio but rarely treated for tubes. I know of only one book that covers it in any detail and only Merlin has made comprehensive measurements. I see it regularly now I use REW for measuring mic pre noise spectra but I have not studied it. There are a whole range of other factors to do with the way a tube mic pre is constructed that affect the ultimate noise level so I want to remove the majority of those before looking closely at 1/f noise.

Then there is the thorny question of weighting and detection. The range from A weighting with an average detector to C weighting with a quasi peak detector - the measured value of noise using these two methods on one circuit can vary by up to 12dB - Merlin and I have debated this at length on another forum.

Bottom line is we welcome anyone prepared to join us in better understanding what really goes on noise wise in tubes at audio frequencies.

Cheers

Ian

 

[abbey road d enfer]

1. I will not ever be using a 100k resistor from grid to earth, there will only ever be a Jensen transformer secondary in the design.

OK, that's good practice; then, why not make the measurements with a resistor that is consistent with the intended use? 20-22k would be more correct.

2.  I used a 100k resistor as it was a simple constant that cancels out during the tests with the other load resistances.

Does it? IMO it tends to swamp the differences. What I mean is, when the tube's shot-noise is modelled with a 25k resistor, and there's already 100k in the grid (actually in series with the stage's source), how can you acurately measure the meagre variations of this 25k resistor?

3.  Using a transformer for a test just makes life difficult (unless it is very well screened) because it amplifies the noise from the environment by about 10. 

That's why I didn't suggest a xfmr, but a suitable resistor; 100k is NOT a suitable resistor. Merlin agrees with me (or do I agree with Merlin?).

A simple 100k resistor has, as you say a known thermal voltage of ~5uV which for relative measurements can be disregarded.  I.E. all the measurements are probably 5uV less than I reported.

NO. Noise contributions do not add linearly, they add quadratically. You could deduct 100k from the calculated total noise resistance, though, but it would be tortuous IMO, since all the other noise resistances are fictitious (the result of calculations, no physical reality). I like to deal with physical factors, convert everything into noise  voltages and currents and sum them quadratically.

4. The equivalent noise resistance inherent in an average EF86 is around 25k depending on the bandwidth, this has nothing to do with a grid to earth resistor, it's just a convenient way of expressing the 2uV of noise generated in the tube.

That would be correct if noise was constituted only of Johnson noise, that could be scaled to the actual BW of measurement, but since it contains other components that are not BW dependant, it is not convenient.

5.  I am not trying to re-write the textbooks, I just believe that they were written for RF work not audio.  If the 2.5/gm formula was true for pentodes, then my graphs would have gone the other way.

Is it even true for triodes? Although it seems to be cast in stone, it needs to be taken with a pinch of corrective salt, i.e. the corrective factor Sigma (Eq. 2 in Blencowe's article). Actually, this formula is related only to one component of noise, shot-noise. If a triode noise was constituted only of shot noise, the spectrum would be exactly that of a resistor, but the actual noise of a triode stage is the sum of many components: shot-noise (Johnson) of the conductance (Gm), flicker noise (1/f), grid-current noise (that would be negligible in most audio applications, but would contribute in view of the excessive source impedance you've chosen for your tests), heater induced-noise, B+ noise, Johnson and excess noise in associated resistors.
When it comes to pentodes, you have to add partition noise.
BTW, the French Mazda/Belvu datasheet for EF86 has an interesting comment:
"In common cases, the resistance in series with teh screen will be 4 or 5 times the value of the anode load resistance. In cases where a particularly high signal-to-hiss ratio is sought, to the detriment of the gain of the stage, the value of this resistance can be increased up to 10 times the anode resistance."

They would have shown lower noise with a 100k Ra than with a 220k Ra for example. 

I already mentioned the noise sources that are not current-dependant AND those that are also directly current-dependant, while shot noise is inversely current-dependant (hence the "hockey-stick" curve).

I just want to set up an EF86 to run with the lowest noise and highest gain 

It's your privilege.  But what are you trying to prove? If you want to prove that a pentode front-end is better than a triode front-end, then make comparisons between those two topologies.
I certainly do not want to denigrate you, for the remarkable work you have undertaken, the achievements you have made, and the task you endeavour but it is clear that your enthusiasm does not compensate for your shortcomings in theoretical knowledge and lack of formal training in metrology. I'm convinced that, by establishing a more "academic" framework, you will achieve a better understanding and control of the numerous factors at play.

 

[moamps]

................. but it is clear that your enthusiasm does not compensate for your shortcomings in theoretical knowledge and lack of formal training in metrology......

Are you a scientist?

@dave
Thanks for very interesting thread.

 

[DaveP]

But what are you trying to prove? If you want to prove that a pentode front-end is better than a triode front-end, then make comparisons between those two topologies.

I think this is where the confusion lies, I am not trying to prove anything.  If I had wanted to compare pentodes and triodes I would have done so, I think you have made this assumption because of your training and background.  If I was preparing a paper for publishing then I would offer it for peer review and you would make an excellent job of that.

I make  recording gear.  To do that I need information and when that is not available I try to find out for myself. (Groupdiy)
After reading nearly all the literature on the subject it has come as a complete shock to me that the published formulae do not apply at audio frequencies (I think I am not alone in this).  There were clues in the fact that the best German  mic-pres disregarded the high gm route, but there was nothing confirming this in the literature.  My simple test has convinced me that I will get better results at  sub mA currents rather than mA currents.

I would like to make very accurate measurements and maybe publish something, but as I am married and want to stay that way,  I have chosen to make recording gear in what spare time I'm allowed rather than prove a point on technology that is at best a historical curiosity to the modern generation.  This forum is called Groupdiy and I try to push that ethos as far as I'm able.

I have a decent DMM a basic 2 channel scope and a sig gen but I will never own a psophometer with CCIR weighting.  I cannot make absolute noise measurements with this gear only relative measurements.  The tests were designed to minimise variables so that a trend could be uncovered which luckily it did.
I do know about quadratic summing but I am trying to make the information accessible  to the very wide range of ability found on this forum.  To be honest with you, after all the hard work this involved, I had hoped for some insight into the results, rather than  pedantic tangents on grid resistors and DMM's. 

I would like some insight into why we have been mislead all these years over RF equations.  I am also aware that no-one knows the cause of 1/f noise, although if we are dividing the speed of light by frequency , then we are talking phenomena from enormous distances.  I would also like to benefit from any tips that experienced members have to reduce noise in practical designs.  I will be doing some tests myself later with some mumetal foil and copper foil.  I am trying to design a mic pre that can compare with the spec of a 1950's V series mic pre, no small undertaking.  The  recording engineer where it ends up will not care if the noise is Shot noise, Flicker noise or Johnson noise, he just wants the best S:N ratio for his mic.

I appreciate your comments Abbey and the articles you have already kindly passed on to me, I also realise why you would want to remind us of academic rigour, but I am here to apply this knowledge to practical effect (groupdiy remember) so my approach is different.
Best
DaveP

 

[abbey road d enfer]

Are you a scientist?

What do you think?

 

[abbey road d enfer]

After reading nearly all the literature on the subject it has come as a complete shock to me that the published formulae do not apply at audio frequencies (I think I am not alone in this).

They do apply. Only in RF applications, only shot noise is taken into account, because 1/f and hum are filterd out. 

My simple test has convinced me that I will get better results at  sub mA currents rather than mA currents.

Taking into account all the noise sources relevant to audio frequencies would probably lead to the same conclusion.

To be honest with you, after all the hard work this involved, I had hoped for some insight into the results, rather than  pedantic tangents on grid resistors and DMM's. 

I was afraid you would take it badly, but you can't blame me for pointing out procedural errors.

I would like some insight into why we have been mislead all these years over RF equations. 

I think I and some others have given you enough clues about the "why".

The  recording engineer where it ends up will not care if the noise is Shot noise, Flicker noise or Johnson noise

You should, because the different spectra of these noises explain why some identical noise levels don't sound identical. Not everything is coloured pink...