what is the relationship between input transformer ratio and tube noise?

[C12VR]

I have read a lot of stuff suggesting 1:10 as acceptable for noise, but also PRR suggests the optimum is the highest practical ratio. How does one determine the noise penalty of using a lower turns ratio, and is it really worth the expense of pursuing old, often expensive high ratio units? What causes a transformer to improve the noise figure in a tube pre?

 

[rock soderstrom]

I have read a lot of stuff suggesting 1:10 as acceptable for noise, but also PRR suggests the optimum is the highest practical ratio. How does one determine the noise penalty of using a lower turns ratio, and is it really worth the expense of pursuing old, often expensive high ratio units? What causes a transformer to improve the noise figure in a tube pre?

I consider input transformer in this application as (more or less) free amplification as far as the noise floor is concerned. The SNR gets better because the input tube, which essentially defines the amount of noise of the amp, does not have to provide this gain. The improvement corresponds in the theoretical optimum case to the gain in dB of the transformer.

1:10 transformers have become the sweet spot for technical and resulting economic reasons. They are a good compromise between technical performance and effort, producibility, costs on the one hand and the gain in "free" amplification/ SNR on the other hand.

Higher step-up ratios go hand in hand with significant increase in production costs and the required know-how, otherwise the technical performance suffers, especially the high-frequency reproduction. The possible gain in SNR is then bought more and more expensively, also in view of other limiting factors with regard to SNR.

With 1:10 you can build excellent tube microphone amps, for me a good compromise.

 

[merlin]

Ideally you want the mic's internal resistance to be the dominant noise source, swamping the tube effective noise resistance (the transformer wire resistance is usually negligible). The effective noise resistance of a decent tube is usually around 3k, so if the mic resistance is 150 ohms you'll want to magnify it by more than 20 times to swamp the tube. That's a turns ratio of 1:4.5 minimum. When the apparent mic and tube resistances are equal you get a 3dB noise penalty, which rapidly shrinks as the mic resistance becomes more dominant. A 1:10 ratio would magnify the mic resistance by 100 times leaving the tube in the dust, which is what you want.

 

[ruffrecords]

The 20KHz bandwidth rms noise at a tube's input is typically little better than -110dBu. The noise due to a 150 ohm mic is close to -131dBu. Hence a 10:1ratio transformer (20dB gain) can get the tube's EIN down to the the same region as the mic's which can achieve a 3dB theoretical noise figure. In practice it is hard to achieve an EIN much better than -125dBU but in the vast majority of situations this is fine because the electronic noise is swamped by acoustic background noise.

Cheers

Ian

 

[merlin]

The 20KHz bandwidth rms noise at a tube's input is typically little better than -110dBu.

I'd say that is a touch conservative; most audio tubes can muster -114dBu no problem, and a really fine one maybe -119dBu, but yeah nature usually gives about -125dBu as a practical limit.

 

[emrr]

In the modern age consider higher input levels allow ratios on the lower end due to the swamping effects, though that approach will penalize low level sources. Occasionally you see ratio switching approaches used, but pretty much never in commercial devices and there are very few choices available for DIY.

 

[ruffrecords]

I'd say that is a touch conservative; most audio tubes can muster -114dBu no problem, and a really fine one maybe -119dBu, but yeah nature usually gives about -125dBu as a practical limit.

A lot depends on how you measure noise. When I first started serious designing with tubes I used a Ferrograph test set with an rms calibrated average reading analogue meter. The crest factor of white noise is much higher than that of a sine wave so heaven only knows what the reading I took on that actually represented. What it did teach me though was that there is a wide variation in measured values between tubes of the exact same type with the variation being greatest in NOS tubes. Later I purchased a Lindos Audio Test Set which measured C weighted quasi peak values of noise which is perhaps the most pessimistic noise reading you can make and makes exceeding -120dBu EIN in a tube preamp very difficult to achieve. These days with FFT real time analysers like REW you can pick almost any analysis you like. True rms C weighted is my preferred choice and with this -125dBu is not too hard.

Cheers

Ian

 

[merlin]

A lot depends on how you measure noise.

Correct, it is easy to mislead if it is not measured right. The only measurements I trust are true RMS unweighted, with a correctly defined noise bandwidth. At a push I will accept A-weighting for the marketing department who want an extra 2-3dB to put in the brochure. I'm not sure when C-weighting would ever be applicable since it is meant to represent higher SPL, which isn't really where you find noise?

 

[ruffrecords]

Correct, it is easy to mislead if it is not measured right. The only measurements I trust are true RMS unweighted, with a correctly defined noise bandwidth. At a push I will accept A-weighting for the marketing department who want an extra 2-3dB to put in the brochure. I'm not sure when C-weighting would ever be applicable since it is meant to represent higher SPL, which isn't really where you find noise?

Of course, you and I have discussed this topic at length. I was thinking of the weighting curve used for ITU-R 486 which I thought was very similar if not identical to the C weighting curve, but I may be wrong.

Cheers

Ian

 

[dmp]

A lot depends on how you measure noise...

These days with FFT real time analysers like REW you can pick almost any analysis you like. True rms C weighted is my preferred choice and with this -125dBu is not too hard.

Cheers

Ian

I'd like to understand better how to measure a noise figure. I've been measuring noise with RMAA which I think is similar to REW. It displays the noise vs frequency and a single dBA number.
For instance, here I measured the noise with RMAA. I had my sound card plugged in to the input and output (and the green line shows the soundcard at -140 dB). I was bringing the soundcard send level down 42 dB and the preamp was adding 42dB gain.
How is a single noise figure calculated from the noise vs frequency data? For unweighted, it it a RMS average, in this case probably about -105dB? Subtracting the gain from this would make it -147dB, which doesn't seem realistic.
And to find an EIN does the input need to be terminated with a 150 ohm resistor?

 

[ruffrecords]

I'd like to understand better how to measure a noise figure. I've been measuring noise with RMAA which I think is similar to REW. It displays the noise vs frequency and a single dBA number.
For instance, here I measured the noise with RMAA. I had my sound card plugged in to the input and output (and the green line shows the soundcard at -140 dB). I was bringing the soundcard send level down 42 dB and the preamp was adding 42dB gain.
How is a single noise figure calculated from the noise vs frequency data? For unweighted, it it a RMS average, in this case probably about -105dB? Subtracting the gain from this would make it -147dB, which doesn't seem realistic.
And to find an EIN does the input need to be terminated with a 150 ohm resistor?

The key concept is that an FFT display usually shows the signal level per root Hertz. Roughly speaking it shows the noise voltage at every single frequency in the audio spectrum. To work out the rms noise such as would be shown on a true rms meter, you need to square the noise level at each frequency, add these all up, then divide by the number of frequencies and take the square root.m Usually the software will do this automatically. Clearly the answer you get depends on the bandwidth over which you do the sums but most good software will allow you to set this. Sometimes you can A weight the answer in which case the A weighting curve is applied to the noise spectrum before taking the rms

If your software does not do this then you can estimate it as follows. If the noise spectrum looks essentially flat (with perhaps a little rise at low frequencies) then you can say it approximates to white noise. Because white noise has equal amplitudes at all frequencies you can simplify the rms calculation because all the values are the same. Bottom line is if you look at the level of the noise spectrum on the FFT, the rms in 20Khz bandwidth will be approximately 43dB higher. Why 43dB? Remember the display shows noise per root Hz. If we take the square root of the bandwidth we have effective done the calculation. The square root of (20,000 -20) is 141.35 which expressed in dB is 43dB.

Cheers

ian

 

[dmp]

If your software does not do this then you can estimate it as follows. If the noise spectrum looks essentially flat (with perhaps a little rise at low frequencies) then you can say it approximates to white noise. Because white noise has equal amplitudes at all frequencies you can simplify the rms calculation because all the values are the same. Bottom line is if you look at the level of the noise spectrum on the FFT, the rms in 20Khz bandwidth will be approximately 43dB higher. Why 43dB? Remember the display shows noise per root Hz. If we take the square root of the bandwidth we have effective done the calculation. The square root of (20,000 -20) is 141.35 which expressed in dB is 43dB.

Thanks for this great answer! So for the plot I linked to, the plot is pretty flat, so using the white noise approximation, the unweighted RMS will be approximately -105+43dB = -62dB
Is the gain then subtracted? So -62dB - 42dB = -104 dB?
And is this the same as a EIN?

 

[ruffrecords]

Thanks for this great answer! So for the plot I linked to, the plot is pretty flat, so using the white noise approximation, the unweighted RMS will be approximately -105+43dB = -62dB
Is the gain then subtracted? So -62dB - 42dB = -104 dB?
And is this the same as a EIN?

OK, once you have measured the noise at the output the equivalent input noise is simply the output noise minus the gain. So in your example if the output noise was -62dBu and the gain was 42dB them the EIN is -124dBu. Remember noise and EIN are absolute levels so they must be referred to an actual voltage level. This also means you need to calibrate your FFT analyzer. The simplest way to do this is to feed it a 0dBU signal at 1KHz and note the level in dB shown on the FFT analyser. Some analysers (REW for instance) allow you to set this level as 0dBu and it will then show its readings in dBu.

Cheers

Ian

 

[Newmarket]

OK, once you have measured the noise at the output the equivalent input noise is simply the output noise minus the gain. So in your example if the output noise was -62dBu and the gain was 42dB them the EIN is -124dBu. Remember noise and EIN are absolute levels so they must be referred to an actual voltage level. This also means you need to calibrate your FFT analyzer. The simplest way to do this is to feed it a 0dBU signal at 1KHz and note the level in dB shown on the FFT analyser. Some analysers (REW for instance) allow you to set this level as 0dBu and it will then show its readings in dBu.

Cheers

Ian

"if the output noise was -62dBu and the gain was 42dB them the EIN is -124dBu."

20 out ? Or am I missing something from the thread ?

 

[dmp]

So in your example if the output noise was -62dBu and the gain was 42dB them the EIN is -124dBu.

Was -124 a typo? -62dBu - 42dBu = -104 dBu
I'll check the dB units in RMAA. I might try REW also, since RMAA does not let the user make any selections about this.

 

[ruffrecords]

Was -124 a typo? -62dBu - 42dBu = -104 dBu
I'll check the dB units in RMAA. I might try REW also, since RMAA does not let the user make any selections about this.

Ooops, brain fart. Yes, it should be -104dBu.

Cheers

ian

 

[john12ax7]

Optimum transfer occurs when source impedance matches load impedance, this holds true for electronics in general. Look up the device voltage noise divide by current noise and you get desired source impedance. The issue for a tube is the current part is very low so the optimum impedance is very high, you want a very large turns ratio for a transformer. In practice 1:10 is about the limit for still having a good transformer, so you use 1:10, which is as high as practical.

Note : For a bjt you would generally want a lower turns ratio depending on its parameters.

 

[ruffrecords]

Optimum transfer occurs when source impedance matches load impedance, this holds true for electronics in general.

Maximum POWER transfer takes place when source and load resistances are identical. However, in a mic pre we want maximum voltage transfer which means the load impedance should be about 10 times the source impedance

Look up the device voltage noise divide by current noise and you get desired source impedance. The issue for a tube is the current part is very low so the optimum impedance is very high, you want a very large turns ratio for a transformer.

Tube data does not include input noise current or voltage. The current factor in a tube is so small it can safely be ignored so all you need to know is the noise voltage. You can convert this into an equivalent noise resistance using the Johnson noise equation.

Cheers

Ian

 

[john12ax7]

Maximum POWER transfer takes place when source and load resistances are identical. However, in a mic pre we want maximum voltage transfer which means the load impedance should be about 10 times the source impedance

Optimal noise figure is achieved at an optimal source impedance. As a practical matter using maximum voltage transfer is ok for a tube since current noise is very low. It is more correct, however, to use the optimum source impedance concept as it applies to all electronics. Without that an inexperienced person might go around changing the 1:2 in an optinzed bjt circuit with a 1:10 thinking they are improving things, while actually making noise performance worse.

 

[ruffrecords]

Optimal noise figure is achieved at an optimal source impedance.

Indeed it is but this is nothing to do with matching source and load impedances.

As a practical matter using maximum voltage transfer is ok for a tube since current noise is very low. It is more correct, however, to use the optimum source impedance concept as it applies to all electronics.

You will need to define correct in this context. Matching the optimum noise impedance is only one aspect. Loading the microphone with the 'optimum' load (whatever that might be) is another. Most mic pres present a load of about 1500 ohms or so to a microphone whose source impedance might be anything from 50 ohms to 300 ohms. Optimum noise matching is not going to occur.

Without that an inexperienced person might go around changing the 1:2 in an optinzed bjt circuit with a 1:10 thinking they are improving things, while actually making noise performance worse.

Yes they might do just that but once again you will need to define 'optimised' in this context.

Cheers

Ian