[DaveP]

Not wanting to re-invent the wheel,

Does anyone have a method for measuring tube noise?

Steve Bench has published a table of readings around 0.67uV and my scope or DMM won't go below 100uV.

Would using an amp work?

best
DaveP

[abbey road d enfer]

A DMM won't do. You need an AC voltmeter with 20-20kHz BW. Noise is BW dependant. Type of detector is important too; should be RMS, but average is okay, gives results a tad pessimistic.

[DaveP]

Ok Abbey,

I'll look into it.

DaveP

[ruffrecords]

You definitely need some for of specialised instrument to do this whether it is home built or commercial. You can make basic measurement (down to about -80dBV) with something like the Ferrograph RTS2 test set (see link below) which can still be obtained for a reasonable sum on eBay. (the Ferrograph will also measure distortion at 1KHz).

http://www.morphet.org.uk/images/ferro/ferro-rts2man/cover.gif

To make accurate, repeatable noise measurements you need 3 things:

1.  Low noise gain
2. An appropriate filter
3. Appropriate meter ballistics

Opinions differ a  lot on 2 and 3 with the marketing guys preferring 'A' weighting filters and rms ballistics as these give the most favourable readings. Manufacturers of professional audio measuring equipment use different filters and quasi peak ballistics because they are of the view it gives readings which more closely represent what we actually hear. I use a Lindos Minisonic MS10 which can accurately measure noise down to -90dBu and conforms to  IEC268 (CCIR 468) weighting and Quasi-Peak dynamics and has balanced inputs. The basic unit cost 400GBP but it also does frequency response and distortion measurements too.

http://www.lindos.co.uk/ms10

Whatever route you take, making noise measurements is not easy (especially of prototype equipment). It is amazing where noise gets in (from soldering irons, flourescent tubes, laptop PSUs etc) so you need to take special care to ensure what you are reading really is noise.

Cheers

Ian

Edit: Here is a picture of my tube noise test jig. It fully encloses the tube under test in a screened box. Heaters wires go straight out the back away from HT and signal wires. The jack plug goes straight into my Lindos.

[DaveP]

Thanks Ruffrecords,

I've set up something similar, what values of resistors and caps are you using?  Are you changing them to suit various tubes or keeping them the same?

best
DaveP

[ruffrecords]

Thanks Ruffrecords,

I've set up something similar, what values of resistors and caps are you using?  Are you changing them to suit various tubes or keeping them the same?

best
DaveP

I have found the operating point of the tube makes a significant difference to the amount of noise it produces so I vary the anode and cathode resistors to achieve the desired operating point.

I now have two separate tube noise measuring rigs. The one I showed in my last post is a mu follower topology because I use that a lot in my mixer designs. It is basically the circuit used in the poor man's gain make up stage but I alter resistor values for different tube types and operating points. The bottom cathode is decoupled by a 220uF electrolytic.

At one point I suspected that the mu follower topology itself was the reason my measured noise levels did not agree with those predicted by accepted theory so I built the second rig. This wires up the two triodes as separate common cathode amplifiers. Again I change anode and cathode resistors to set suitable operating points and again the cathodes are decoupled by 220uF electrolytics.  I used this rig to check many of the tubes I had already checked in the mu follower rig. I found that the noise was no better after accounting for the lower gain of the CC compared to the mu follower.

In both rigs the input grid is shorted directly to ground. In both cases the opulent from the anode is via a 1uF capacitor into a 100K resistor.

Attached is a picture of the second rig.

Cheers

Ian

[DaveP]

Thanks Ian,

That's very useful info.  Steve Bench had a similar procedure.

I shall build a dedicated VVM for the job along the lines of the Heathkit.

With about 80dB gain and maximum feedback I should be able to get enough low noise sensitivity to differentiate between tubes.

Did you find that the lowest noise was when the load resistor was ~twice the rp/ra?

best
DaveP

[DaveP]

One other thing,

How have you cleaned up the HT supply for the rigs?  At these very low noise levels there may be some residual ripple coming through despite the PSRR.  I suppose I could use filters to separate hum from noise measurements.

best
DaveP

[ruffrecords]

One other thing,

How have you cleaned up the HT supply for the rigs?  At these very low noise levels there may be some residual ripple coming through despite the PSRR.  I suppose I could use filters to separate hum from noise measurements.

best
DaveP

I use a four stage RC filter with each stage consisting of 1K and 470uF. Theoretical hum is around one microvolt.

Cheers

Ian

[ruffrecords]

Thanks Ian,

That's very useful info.  Steve Bench had a similar procedure.

I shall build a dedicated VVM for the job along the lines of the Heathkit.

With about 80dB gain and maximum feedback I should be able to get enough low noise sensitivity to differentiate between tubes.

Did you find that the lowest noise was when the load resistor was ~twice the rp/ra?

best
DaveP

Basically your VVM needs to be able to measure a few tens of microvolts.

I did not find that noise was lowest when the load resistor was twice the tube ra although with some tubes like the 12AX7 that does lead to a relatively low noise operating point but the tests I have done indicate that this is not universal. What does generally lead to lower noise is lower Va and lower Ia. In my experience, running hi gm tubes at high currents so the operating point gm is high does not lead to low noise.

Cheers

Ian

[DaveP]

Hi Ian,

Thanks for all your help, I'll let you know how it turns out.

Interesting about the low current noise, even though the triode formula is Req.= gm/2.5, the designers of the V72~V76 series went for very low currents too, even though the pentode formula is more biased for high current.  You would have thought that more current equals more noise from more electron collisions etc.

Another factor is that low currents imply high load resistors so the gain is higher and signal to noise ratio higher.

best
DaveP

[ruffrecords]

Hi Ian,

Thanks for all your help, I'll let you know how it turns out.

Interesting about the low current noise, even though the triode formula is Req.= gm/2.5, the designers of the V72~V76 series went for very low currents too, even though the pentode formula is more biased for high current.  You would have thought that more current equals more noise from more electron collisions etc.

Another factor is that low currents imply high load resistors so the gain is higher and signal to noise ratio higher.

best
DaveP

I have nearly completed my investigation of the causes of tube noise and, without wishing to go into detail, I can say with a fair degree of certainty that Req=2.5/gm is the least of your worries when designing low noise tube audio circuits. In practice flicker noise is much more important and there is a formula for predicting it and basically it is proportional to Ia/gm where Ia is the quiescent current. As you can see, unless gm rises faster than Ia (and I have not found one tube yet where it does), higher currents lead to higher gm  but higher noise. In practice Ia/gm gets smaller as Ia gets smaller. It also gets smaller as the plate to cathode voltage gets smaller. So in a nutshell I would aim for low Ia and low Vak to minimise noise. Practical tests back this up and as you pointed out, most if not all of the early mic pre designs adopt this approach.

Cheers

Ian

[DaveP]

Thanks Ian,

I've just found all the posts in "Tube Noise Myths" you had back in August so I can see where you are coming from with the 1/f issue.

I've just read some very heavy maths on the subject online and one important fact,  that the metal used for the cathodes varied a great deal and the interface between the emissive material and the metal is the source.  The metal used varied between manufacturers and caused some european tubes to be less noisy in this respect.

Ref: Smullin L.D. & Haus H.A., Noise in Electron Devices, MIT Press, 1959

I see that temperature issue was brought up to correct the 2.5 to 2.3 but as the cathode heater is operating at~800C or 1073K it might even be far worse!

Chemistry is my primary field and in my youth I made about 100kg of the triple carbonate used by British tube manufacturers in the early 70's. (Calcium Strontium Barium Carbonate) so I guess all the noise is my fault!

best
DaveP

[PRR]

> unless gm rises faster than Ia (and I have not found one tube yet where it does)

For plain tubes at useful currents, Gm can't rise faster than Ia.

For plain tubes at useful currents, Gm rises about as square-root of Ia.

This can be warped, Super-Control (remote-cutoff) types.

But still Gm rising faster than Ia overall leads to a paradox, and rising fast over only a small range would give annoying (for most uses) nonlinearity.

[ruffrecords]

Thanks Ian,

I've just found all the posts in "Tube Noise Myths" you had back in August so I can see where you are coming from with the 1/f issue.

Yes, I had been struggling fro some time to understand why I could never measure noise levels anywhere near as low as the 2.5/gm formula would lead one to expect and the tube myth post was at least partly in frustration at that. Fortunately Merlin gave and excellent reference:

Cherry & Hooper (1968) Amplifying devices and low pass amplifier design. p76

I got this book from my local library (via the British Library because it is a rare book) and it provides chapter and verse on tube noise and goes into detail about flicker noise.

The bottom line is it gives predictions much closer to what I measure but more importantly it gives the clue to achieving the minimum noise by reducing Ia and Vgk and I am pleased to say my tests yesterday on some newly arrived brand new production EH  6922 tubes and Ei PCC88 tubes and some NOS Brimar ECC88 has confirmed that significant improvements are obtained by this means. Because mu is reduced a low Ia and increased by low Vgk I modified my test rig so that I could also measure stage gain at the same time and hence calculate the true EIN. I need to make one further modification to test a 12AX7 at below 1mA because from the data sheets this should give the lowest value of the ratio Ia/gm and hence lowest flicker noise.

Cheers

Ian

Reducing SE amp noise by injecting noise

Started by adamasd Drawing Board

A practical method for measuring signal to noise ratio

Started by truzz The Lab

measuring mic pre noise

Started by briomusic « 1 2 » Drawing Board

best method for measuring noise level of a 6AU6?

Started by outoftune The Lab