Comparison of JFETs for mic applications

[moamps]

It doesn't work to measure uncorrelated noise. Which is why I insist that mickeysoft cannot measure -20.3dB (A) random noise 20Hz-20kHz using two microphones with 5.5dB(A) self noise.r

As someone else mentioned, averaging works only for separating non-correlated noise from correlated signal.
When both are random, it just doesn't work.

I think some terms are confused here, I will try to clarify them a little, the way I see it with my limited knowledge and sometimes lack of accurate information.

The MS anechoic room measurement was made by experts from the B&K company, the result was recognized as a Guinness record. I see no point in denying the accuracy and relevance of this measurement.

When it says -20dB(A) it means exactly that, 20dB lower signal than 20uPa, measured using the A type curve (the A type curve is determined in the frequency range from 20Hz to 20kHz, and cannot be applied to a non-wideband signal, e.g. a 1/3 octave filtered signal).

For example, the noise of the BK 4179 cartridge is declared -5.5dB(A), with the amplifier it is -2.5dB(A), while for 1/3 octave at 1kHz this noise drops to -16dB, and with additional band reduction it can be reduced to -34dB(re 20uPa).

When measuring with two microphones, the uncorrelated noise is the Brownian noise generated on each microphone membrane. The second noise that we want to measure is actually the signal that has penetrated our anechoic room. It is correlated in both microphones (that is why the microphone membranes are placed next to each other). So, we have two different types of noise.

Finally, the coherent power measurement technique allows us to measure the values of noise which can be lower than the value of the total noise produced by an individual microphone. So, it is possible to use this technique using microphones with 5.5dB(A) noise to measure the presence of a signal (in this case, it is the unwanted noise that has entered the room) which is lower than that value.

 

[thor.zmt]

When measuring with two microphones, the uncorrelated noise is the Brownian noise generated on each microphone membrane. The second noise that we want to measure is actually the signal that has penetrated our anechoic room. It is correlated in both microphones (that is why the microphone membranes are placed next to each other). So, we have two different types of noise.

What I said, no?

Finally, the coherent power measurement technique allows us to measure the values of noise which can be lower than the value of the total noise produced by an individual microphone. So, it is possible to use this technique using microphones with 5.5dB(A) noise to measure the presence of a signal (in this case, it is the unwanted noise that has entered the room) which is lower than that value.

Yes, but by doing so we reject ALL random noise, including Brownian motion noise, so we cannot measure -20.3dB Brownian motion noise this way.

Theory suggests that for 20Hz to 20kHz bandwidth the Brownian motion noise on the diaphragm is a smidgen above "0dB".

Of course, we can measure -20.3dB(A) Brownian motion noise, we just need to pick a bandwidth of 20Hz -200Hz, apply A-weighting.

To figure the theoretical A weighted Brownian motion noise under these conditions is left as exercise to the reader. It's lower than -20.3dB (A) incidentally.

Thor

 

[moamps]

Yes, but by doing so we reject ALL random noise, including Brownian motion noise, so we cannot measure -20.3dB Brownian motion noise this way.

What was measured here was not Brownian noise, but a signal (you can call it noise if you want) that penetrated the anechoic room. And its value is the famous -20dBA.

 

[thor.zmt]

What was measured here was not Brownian noise, but a signal (you can call it noise if you want) that penetrated the anechoic room. And its value is the famous -20dBA.

Yes, my point from the beginning. The -20.3dB figure is not Brownian motion noise. I think the debate has reached QED for the third or fourth time now.

Thor

 

[ricardo]

Eliminating non-film capacitors so only very high grade coupling capacitors are used .... balanced supply currents, that is generally expected for microphones.
... loadsa stuff some good but some dodgy ...
, for transformer coupled inputs the "simple P48" will magnetise the core permanently, degrading performance and causing the transformer in effect to bd destroyed, if using a DC coupled inputs, the DC difference will stop the circuit from working.

Some dodgy stuff here (along with some good stuff of course as expected from Thor )

I'll just answer the stuff about transformer input preamps. Various MicBuilders have tested SimpleP48 with these and these are known to be OK long term with Lundahls, Sowters & Jensens. Lesser brands might have problems. And several $$$ mikes also have this sin with even MORE EVIL unbalanced currents. It's all in SimpleP48.pdf

I'm not sure Thor is of the vintage that unnerstans transformers in preamp design so he is forgiven for doubting Great Guru Baxandall et al.

For his other criticisms, you can build a rather good omni with SimpleP48 for less than $10, probably less, with stuff in your junk box in less than an hour, and see (hear) for yourself.

Having been responsible for probably the most complex mike of the 20th century, these days, I look with a jaundiced eye at any complexity that doesn't give me more practical advantages compared to SimpleP48

The "high grade capacitors" and other stuff comes under "my stuff is hand carved from Unobtainium & Solid BS by Virgins"; really important for any commercial maker. But you can claim this without putting any s**t into your product as long as the user is ... Shut up! Just SHUT UP!!!

 

[ricardo]

The second noise that we want to measure is actually the signal that has penetrated our anechoic room. It is correlated in both microphones (that is why the microphone membranes are placed next to each other).

This is what I find dodgy. If the noise is correlated in both microphones, would it be also correlated between different 'samples' that you are averaging.

Unless of course Microsoft, in their infinite wisdom built their anechoic on top of a generator which was the main 'external' noise source. Such a noise can, with suitable digital juju, be suitably aligned such that subsequent samples to be averaged, build up.

They could have used a zillion 4955s (?) and done a 'single' sample ... but then this begs the question of how a zillion 'closely spaced' mikes will pick up a 'correlated(??)' random signal A page or two of obfuscating maths re spaced mikes in a 'diffuse' soundfield deleted.

Gotta watch the video again to see if any old Naerum B&K people appear that I know from da previous Millenium

Duu.uh! Only 2 x 4955s. No Danish B&K engineers. Indian engineers must be as s**t as Thor's Chinese engineers.

And I can hear an annoying Microsoft bell sound even at their -20.6dBA

 

[thor.zmt]

I'll just answer the stuff about transformer input preamps. Various MicBuilders have tested SimpleP48 with these and these are known to be OK long term with Lundahls, Sowters & Jensens.

Unless specified for DC current, the permanent magnetisation of the core will happen, including reduced primary inductance and increased distortion.

Lesser brands might have problems.

No, all of them, unless explicitly designed to handle unbalanced DC.

And several $$$ mikes also have this sin with even MORE EVIL unbalanced currents.

I am sure. The circuit floated around for decades.

I'm not sure Thor is of the vintage that unnerstans transformers in preamp design

I understand audio transformer design itself quite well, thank you.

For his other criticisms, you can build a rather good omni with SimpleP48 for less than $10,

I don't really need this. What if I want to build a good LDC with externally biased capsule? Which is what my designs are for.

The "high grade capacitors" and other stuff comes under "my stuff is hand carved from Unobtainium & Solid BS by Virgins"; really important for any commercial maker.

The objective performance difference is quite stark.

Thor

 

[thor.zmt]

This is what I find dodgy. If the noise is correlated in both microphones, would it be also correlated between different 'samples' that you are averaging.

That is the point.

External noise (say the subway passing under Kingsway Hall or the distant FLAK on the first hifi stereo tape recordings of music in Berlin in 1943) is both from a distant source and not random.

Both microphones react to it equally.

Random electronic noise and Brownian motion noise from the microphones is not correlated but random.

On the rule that equal levels of random noise adds 3dB while two correlated signals add 6dB, we gain 3dB SNR by using two microphones or two samples.

Every time we double the number of microphones or of distinct individual samples we gain another 3dB.

With 2.5dB(A) effective self noise we only need to average 256 Samples with (say) 64k sampls size to get below -20.3dB(A).

Or 512 individual microphones will do the same.

Thor

 

[Paul W]

This is what I find dodgy. If the noise is correlated in both microphones, would it be also correlated between different 'samples' that you are averaging.

I would think not. Recording simultaneously with two microphones will record any ambient sound in phase. Recording at different times will record different sounds, or if they're steady sounds, record them at randomly different phases.

So, for example, if you have an ambient 60 cycle hum, recording 512 versions of it randomly staggered in time will tend to make it mostly cancel out.

And any uncorrelated, one-shot noises would be averaged way, way down with recordings where that one shot doesn't happen at the same time.

 

[abbey road d enfer]

Unless specified for DC current, the permanent magnetisation of the core will happen, including reduced primary inductance and increased distortion.



No, all of them, unless explicitly designed to handle unbalanced DC.

Actually, ALL transformers are victims of magnetization, not only permanent but also that generated by the actual circulation of current through the windings (H field).
A xfmr "specified for DC current" is just one the designer defined an acceptable deterioration of performance. A transformer "specified for DC current" by manufacturer X maybe unacceptable by manufacturer Y or client Z. It's all a matter of expectations.
No common xfmr is actually 100% ungapped. And no xfmr "specified for DC current" is actually immune to the effects of DC.

 

[thor.zmt]

Actually, ALL transformers are victims of magnetization, not only permanent but also that generated by the actual circulation of current through the windings (H field).

Yes, but to apply significant DC current to a Transformer with Mu-Metal cores not explicitly designed to operate thusly causes permanent, practically irrecoverable deterioration of performance.

A xfmr "specified for DC current" is just one the designer defined an acceptable deterioration of performance.

Not quite. It is one that is DESIGNED EXPLICITLY to withstand the specified DC current while the deterioration of performance is within the specified paramebters.

A transformer "specified for DC current" by manufacturer X maybe unacceptable by manufacturer Y or client Z. It's all a matter of expectations.

A transformer specified for x mA DC should provide rated performance at this level of DC, so it is not a question of expectation, but of reading the datasheet, presuming there is one that is adequate.

No common xfmr is actually 100% ungapped.

In 2024 many are made on gapless amorphous/nanocrystalline toroid cores. Nagra uses these, for example.

And no xfmr "specified for DC current" is actually immune to the effects of DC.

Any transformer designed explicitly to handle a specified level DC will perform as designed with this level of DC and performance will deteriorate at greater levels of DC.

If it is explicitly designed to handle 0.00 mA than applying ANY DC will degrade performance.

Perhaps the degradation and the permanent degradation of performance is acceptable, but it needs to be warned against.

Finally, as I am critiquing the SimpleP48 circuit for being too simple, it does behoves me to show an alternative that addresses somebunall of the grosser shortcomings.



Both this and the original circuit share the same shortcomings. Absolutely awful HD approaching 10% at 124dB SPL (assuming a -40dB sensitivity Capsule), poor cable capacitance handling, really, there is nothing whatsoever to recommend either the original 2 component circuit or this one when we have better options.

Yes, either version will work and some sound will come out. Which may be enough for some.

For the rest...



The 10nF and smaller Capacitors should be C0G (in 0603 to fit the BM-800 PCB which uses lower grades) or film. For the 100n C0G are preferred as well, but X7R will probably not be the end of the world. 1uF need to be X7R.

The BSS84 need to have matched threshold voltages. The ones I had so far from a reel all matched so well, no extra action was needed.

The bias chain (given 1% resistors) will have > 35dB suppression of Audio across all frequencies, the lowpass filter effect adds ~ -60dB @ 20Hz, so any audio will be attenuated by at least 95dB @ 20Hz, with a 60dB/decade slope.

As shown here, we have 40V 35V bias, not 60V, which will take 3.5dB 4.7dB of the sensitivity, so a -34dB/pa @ 60V sensitive LDC capsule will produce -37.5dB/pa -38.7dB/pa or 11.7mV/94dB to convert into modern money. Using P68 instead of P48 and adjusting R16 (switchable?) would recover this sensitivity, but it is unlikely to improve SNR materially.

4Hz-100kHz -3dB with an option for a HF roll-off (increase C2) to limit how bright mic's are.

Predicted is an unweighted 14dB 20dB self noise 20Hz-20kHz, so perhaps 10dB(A) 16dB(A) without accounting for Brownian Motion Noise and acoustic resistances etc. I would say "low enough for music recording" but not phantastically low. About 30dB 36dB worse than -20.3dB(A).

With a 2.2k Mic Pre input load we can handle +35dB/pa or 129dB at ~ 0.2% THD +40dB/pa or 134dB at < 0.5% THD (H3 dominates) with a -34dB/pa @ 60V bias capsule.

Higher loads allow more SPL.

As mentioned before, this will easily be able to be realised on the BM-800 PCB with a few cut's & jumps.





Add the favoured AliExpress K47 or whatever capsule plus saddle, a true studio grade recording LDC is done.

Anway, enough from me.

Thor

 

[abbey road d enfer]

Yes, but to apply significant DC current to a Transformer with Mu-Metal cores not explicitly designed to operate thusly causes permanent, practically irrecoverable deterioration of performance.

I am not aware of Mu-Metyal that can't be demagnetized.

Not quite. It is one that is DESIGNED EXPLICITLY to withstand the specified DC current while the deterioration of performance is within the specified paramebters.

That's exactly what I said. Specified parameters are arbitrary. Deterioration accepeted by designer may be inacceptable for another individual.

In 2024 many are made on gapless amorphous/nanocrystalline toroid cores. Nagra uses these, for example.

Many for you may be very few for others. They are still exceptions in products available to DIYers.

Perhaps the degradation and the permanent degradation of performance is acceptable, but it needs to be warned against.

Acceptable is the word.
There is no absolute zero current like there is no absolute evil or absolute good. Everything is variable, with more or less abruptness.

 

[thor.zmt]

I am not aware of Mu-Metyal that can't be demagnetized.

By the average DIY enthusiast with the tools he has. I work with transformer makers who demag their transformers before shipping. So I do know how it's done.

Maybe you describe the process for those who are ignorant?

That's exactly what I said. Specified parameters are arbitrary. Deterioration accepeted by designer may be inacceptable for another individual.

Then the initial stated parameters would also be unacceptable, as they would/should be quoted at rated DC. So we are back to kindly read the datasheet.

Many for you may be very few for others. They are still exceptions in products available to DIYers.

So you agree your previous statement was not correct?

This statement:

No common xfmr is actually 100% ungapped.

Clearly now there are many that are. Look for the watchwords "amorphous" and "nanocrystalline".

Acceptable is the word.

Again, I will state that this is a design issue.

If I define a transformer to have a given frequency response, loss and distortion behaviour in a given circuit but I specify "0mA" DC in one case and "1mA DC" in the other, I will get different transformers.

There is no absolute zero current like there is no absolute evil or absolute good. Everything is variable, with more or less abruptness.

We design transformers for a set of performance requirements. The Amount of DC to be handled is part of the specification and alters the design of the transformer.

Do you really want to argue that this is not the case?

Thor