Comparison of JFETs for mic applications

[abbey road d enfer]

You are assuming something.

Of course. Who doesn't?

Brownian noise was covered in my reference from 1974 and was suggested as -100dB/PA (now I got my unit conversion fixed) in a physical sense (-5dB re 0.0002ubar).

I would like to have a look at this "reference". I've spoken with several acoustics guru, had all different answers.

Yes, but if a microphone has (say) 6dB(A) self noise, like the Microtech Gefell M 940, how can we possibly measure noise levels 26dB lower?

Their method is not described, but I doubt they even used microphones as we know them.

 

[moamps]

.....

Their method is not described, but I doubt they even used microphones as we know them.

"...used a two-microphone coherent power measurement technique with two Type 4955 low-noise microphones. The acousticians measured the same overall dB(A) multiple times."

 

[thor.zmt]

I would like to have a look at this "reference".

I already posted it. You didn't notice?

The Journal of the Acoustical Society of America - February 1974 - Noise Due to Brownian Motion of the Diaphragm, of Electret Microphones

I've spoken with several acoustics guru, had all different answers.

Maybe cut out the Gurus and talk to the scientists.

Thor

 

[thor.zmt]

"...used a two-microphone coherent power measurement technique with two Type 4955 low-noise microphones.

5.5dB(A) self noise. Use two, 3dB SNR improvement.

There is a lot that can be done with averaging to remove random noise (e.g. Brownian noise).

Each time we double the number of samples we average, we gain 3dB SNR for correlated signals. So 256 averages with the correct window should do.

I routinely use this method to measure below the noisefloor of AP2, but this method rejects any random noise, including random noise from the source. It is good to look for correlated signals well below the noise floor.

Not sure if that is what mickeysoft did.

The acousticians measured the same overall dB(A) multiple times."

So it's -20.3 dB(A) not -20.3 dB, plus the way it is measured is questionable SMD tells us nothing about Brownian noise, so we can safely ignore all of it in context.

Thor

 

[moamps]

I guess we can kill Brownian noise by using our microphones in a Vacuum and the resistor noise by doing it at very low temperatures close to absolute zero... 555

And just cooling to absolute zero would be enough, no vacuum is needed, no Brownian motion is expected at that temperature. But at that temperature, the microphone capsule membrane probably wouldn't be able to move, and there wouldn't be any sound to set it off, because the air molecules would be completely devoid of energy and motion.. Hmm..

 

[moamps]

Here is a very interesting graph showing the contributions of different types of noise.
Source: B&K, Microphone handbook, Volume 1, Theory

 

[ricardo]

5.5dB(A) self noise. Use two, 3dB SNR improvement.

There is a lot that can be done with averaging to remove random noise (e.g. Brownian noise).

Each time we double the number of samples we average, we gain 3dB SNR for correlated signals. So 256 averages with the correct window should do.

Er.rrh! In this case, our 'signal' is also "supposedly random" noise of a quiet room. So I'm not sure this would give you something to measure da 'quiet' room.

10 pages of obfuscating theory about correlated or non-correlated room noise between 2 closely spaced B&K Supa Lo Noise mikes deleted. I'm suspicious of da 'acousticians' in that youTube video but I'm only a pre10 acoustician and mike designer.

 

[ricardo]

Should also point out that well before we hit da noise limit of Brownian motion against da diaphragm, da EVIL acoustic resistance used by us horny handed mike designers to get flat response will dominate.

There are exceptions like the B&K Supa LN omni, some Sennheiser mikes and even an experimental one I did circa 1980 to go with Great Guru Baxandall's RF circuit which I considered for the Mk4 Soundfield.

But Zephyr.doc in MicBuilders shows how your Mike Preamp and even the EVIL small resistors (not just da 1Gs) in a good Schoeps variant can be audible. The curve there is similar to the B&K stuff that moamp posted but in more detail.

BTW, GG Baxandall's 1968 tome on Condensor Microphone noise is also in MicBuilders. It is the best accurate and accessible explanation I know off. Wannabe LN gurus should read this daily on their knees facing Malvern, UK. Even old gurus like Thor might find pearls of wisdom when re-visiting it in their old age.

 

[Voyager10]

I expect the point of using two mics is that noise 'in the room' arrives at both mics together, and noise 'in the mic' (electronic, Brownian motion) is independent for each mic. So the correlation between the mic signals gives you the in-the-room noise.

So, I'm theorizing here, they multiply the signals from the two mics together, then average over time. If the signal is s, and the mics' noise is e1 and e2, you're multiplying (s+e1)(s+e2) = s^2 + (e1+e2).s + e1.e2 . If e1 and e2 are random and uncorrelated then both (e1+e2) and e1.e2 average out to zero and we're left with s^2, our mean-square signal value.

 

[abbey road d enfer]

Brownian noise was covered in my reference from 1974 and was suggested as -100dB/PA (now I got my unit conversion fixed) in a physical sense (-5dB re 0.0002ubar).

This doesn't add up with the Brownian noise figure of -20.3dB, even if we consider the weighting effect.
If noise in the range 1-5kHz is -5, how could the 20-20kHz be -20?

 

[thor.zmt]

This doesn't add up with the Brownian noise figure of -20.3dB, even if we consider the weighting effect.

Probably because "-20.3dB" is not the figure fof Brownian noise.

If noise in the range 1-5kHz is -5, how could the 20-20kHz be -20?

It cannot be -20. And it is not.

The theory of Brownian motion was well understood in 1974. The article I referenced was testing empirically how well the theoretical noise from air molecule matche actual noise measured with a specific microphone.

The conclusion is in the article. There is no point debating physical constants and basic laws of nature, so I'm done.

Thor

 

[thor.zmt]

BTW, GG Baxandall's 1968 tome on Condensor Microphone noise is also in MicBuilders. It is the best accurate and accessible explanation I know off. Wannabe LN gurus should read this daily on their knees facing Malvern, UK. Even old gurus like Thor might find pearls of wisdom when re-visiting it in their old age.

I am not a wannabe, or a guru. I just like to keep things real where solid science exists.

I am not of the opinion that extremely low noise microphones are a valid goal FOR RECORDING MUSIC.

They are interesting from a purely technology and technical standpoint, the way stealth fighters and nuclear warheads pique my intellectual curiosity, but I have little practical use for the knowledge.

For a microphone, I simply will restate this:

Use a Schoeps circuit, with a 2SK660 as J-Fet. Place the FET directly at the capsule, apply bias to the backplate.

Cascode said J-Fet with a J113 or any other J-Fet that gives enough operating voltage.

Use a BSS84 Mosfet to make a Sziklai circuit with our cascoded 2SK660. Use this as the frontend. It will be low enough noise and very low distortion. We expect around 2mA and 24V available for our frontend. So 2 x 2.7k load resistors.

Replace the PNP transistors in the output stage with BSS84 Mosfets, arranged suitably and use 0.1uF C0G coupling capacitors or 0.1uF film capacitors. Each transistor has around 1.2mA Source current and around 10V across it.

This circuit can actually be build up on the original Chinese PCB that came with our BM-800 we bought to use as shell.

The flip side of the PCB can be used to tap off the 40V that remain from our P48 phantom power and use a simple RC chain to make 40V bias. This looses us around 3dB sensitivity and SNR, but otherwise is fine.

Add your favorite Chinese AliExpress capsule and a 3D printed saddle, bobs your uncle.

Result?

Super simple DIY microphone with an objective performance materially better than most Chinese microphones a available. It will be adequate to get a clean and neutral (depending on capsule) sound, will not overload easily, will be low enough noise and generally just a reasonably open window on the performance.

All for less than the price of a steak dinner for two at a cheap chained steakhouse for two mic's.

If you want a Vintage U87 this is not it, neither will be any replica, only the real thing will be real. So why deceive oneself with poor knockoffs, DIY or Commercial?

I made a few like this in China for vocalists who liked them (and we're cute and willing to return favours horizontally) and tried to convince a big factory to switch to my design in slightly modified BM-800 shells. They were interested, liked what they heard, but we're unwilling to pay me anything. So this went nowhere.

Thor

 

[abbey road d enfer]

Probably because "-20.3dB" is not the figure fof Brownian noise.

Then, what is it? Noise inside an enclosed volume cannot be smaller than that due to Brownian movement. Just like noise in a complete circuit cannot be smaller than that in the real part of the source impedance. They measured -20, so Brownian noise should be smaller than that.

It cannot be -20. And it is not.

If you say so...

The conclusion is in the article.

The link you posted is only a short abstract that doesn't say it all. I'd like to see a real demonstration.

 

[abbey road d enfer]

Unless we have more than two braincells to rub together and use a diode or a special J-Fet designed for microphones.

You should re-read. Ricardo spoke of "acoustic resistance", not the gate resistor.

 

[thor.zmt]

Then, what is it? Noise inside an enclosed volume cannot be smaller than that due to Brownian movement.

BUT THE MEASURED NOISE CAN BE, by using averaging many individual measurements. Anything random disappears "magically". Brownian noise is random.

Just like noise in a complete circuit cannot be smaller than that in the real part of the source impedance. They measured -20, so Brownian noise should be smaller than that.

No, this doesn't follow at all.

The link you posted is only a short abstract that doesn't say it all.

??? Here is what it says:

Measurements on a circular electret microphone with a radius of 1.8 cm and a sensitivity of 5 mV/µbar yield Brownian‐motion noise voltages of about 9 nV/Hz1/2 in the frequency range between 1 and 5 kHz, in good agreement with calculated results. The total noise voltage over this frequency range corresponds to a sound pressure level of −5 dB relative to 0.0002 µbar, or an average diaphragm displacement of 5X10−12 cm.

So it says EXACTLY what it says. And what I claimed it said.

So please take up your beef with the universe for being the way it is, or with the marketing / publicity people at mickeysoft for not giving a full scientific article with all details (which was not their aim anyways).

Thor

 

[abbey road d enfer]

BUT THE MEASURED NOISE CAN BE, by using averaging many individual measurements. Anything random disappears "magically". Brownian noise is random.

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

No, this doesn't follow at all.

Please say why.

So it says EXACTLY what it says. And what I claimed it said.

Then the people at Microsoft are off by at least 20dB. I guess it would have been noticed by the scientific community. Do you think they are not under scrutiny?

So please take up your beef with the universe for being the way it is, or with the marketing / publicity people at mickeysoft for not giving a full scientific article with all details (which was not their aim anyways).

This shows that you completely ignore what I am. I have no beef, I try to put things in perspective, and when something seems weird to me, I look for proof. As to MS people, when the show at 3.01 in the video "brownian motion -20.3dBA", it seems enough detail to me. Unless they have a hidden agenda...

 

[thor.zmt]

I'd like to see a real demonstration.

Surely math will suffice?

L. J. Sivian and S. D. White, On minimum audible sound fields. Journal of the Acoustical Society of America, 1933, 4, 288-321



If we calculate 1k-6kHz, we get 5.3×10−11 bars = 5.3 µPa = −11.6 dB SPL. But this is across a very limited bandwidth.

Calculating total SPL, in the same way but over 20 Hz to 20 kHz, we get 21.8 μPa, very close to 0 dB SPL.

This equation also lets us calculate the spectral density, which seems to be violet noise, increasing by 6 dB every octave, matching other references shown here, not the flat white noise I erroneously assumed.

By "cherry picking" bandwidth (say 20Hz to 200Hz) and weighting (say "A") we can claim almost any SPL for Brownian motion noise we like to.

So unless mickeysoft tell us EXACTLY what they measured and how, their numbers are meaningless.

Thor

 

[thor.zmt]

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

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.

Then the people at Microsoft are off by at least 20dB.

Or they measure something very different to what we would like know for microphones.

I guess it would have been noticed by the scientific community. Do you think they are not under scrutiny?

Have they published in a peer reviewed journal?

As to MS people, when the show at 3.01 in the video "brownian motion -20.3dBA", it seems enough detail to me. Unless they have a hidden agenda...

What bandwidth, measured how?

Without full disclosure, it's just a random number.

What is the hidden agenda behind such videos?

Somehow translating a huge expense on a white elephant with no practical use into cash, by converting it into goodwill.

Obvious.



Thor

 

[ricardo]

Use a Schoeps circuit, with a 2SK660 as J-Fet. Place the FET directly at the capsule, apply bias to the backplate.
.... loadsa good stuff describing Supa Dupa Schoeps variant with everything changed ....
super simple DIY microphone with an objective performance materially better than most Chinese microphones a available.

What practical advantage does your Supa Dupa Schoeps variant give the user which da common or garden Schoeps variant lacks?

Or even just using your 2sk660 with a suitable electret capsule in SimpleP48?

I can think of only two. A 20m cable restriction and some 5dB less sensitivity .... provided your variant is at least as good as da common or garden Schoeps ... which I'm sure it must be.

SimpleP48 will also have about 1dB better S/N but this SHOULD be swamped by da EVIL acoustic resistance noise

 

[thor.zmt]

What practical advantage does your Supa Dupa Schoeps variant give the user which da common or garden Schoeps variant lacks?

Eliminating non-film capacitors so only very high grade coupling capacitors are used.

Reduced noise (Cds cancelled by cascode, Cgs bootstrapped by Sziklai circuit)

Improved linearity at all SPL's due to more linear follower input impedance and extremely linear frontend.

Much improved SPL handling compared to the original with a similar capsule.

SimpleP48 will also have about 1dB better S/N

That remains to be seen, I doubt it. In practice it will not be material, except with very low capacitance capsules.

Plus, no need to limit ourselves to electrets.

Plus balanced supply currents, that is generally expected for microphones.

I have seen this circuit recommend in letters in the 70's for use with cheap electret capsules for "studio recording microphone use". Didnt fly.

Also, 20m cable (1nF) with a source follower operating at a fraction of a mA and with a 47...100k load?



And as drumkit overheads where "Animal" plays drums and is especially fond of the crash cymbal?

I normally recommend < 10m for the ""Fat Schoeps".

There is a trick to make the output Push-pull on each signal line, with double the current available at HF. It adds two N-Channel Fets, one resistor and two more cap's.

Next, 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.

This circuit needs a list of health warnings longer than what's on a pack of ****.

So the simple P48 circuit is a clear example from "overmuntzing", or making something too simple, at the expense of universal usability.

Adding just two more resistors and another coupling capacitor could fix some drawbacks, without getting excessively unsimple.

Any intelligent fool can make things bigger and more complex. A. Onestone

Everything should be made as simple as possible, but no simpler. A. Onestone

Incidentally, all this veer started when I suggested that using very low voltage noise J-Fets in microphones was counterproductive, as the complete circuit with capsule, Brownian motion noise, be it direct, or amplified with acoustic resistance, swamp out all but the noisyiest J-Fet and that it was likely better fo focus on other figures of merit.

I didn't think of this statement as in any way controversial, as it is a position grounded solidly in science, validated theory, confirmed here partially by experiments here and further elsewhere decades ago and pages on pages later we still debate these simple facts?

Why this extreme need to endlessly debate simple statements of fact I make?

It doesn't change the facts. It's such a trumpian whatifism move. Why not stick to facts and implications from this?

If I got something wrong (happens, getting old here, brain ain't what it used to be) tell me and I post the correction. This is not an ego trip. This is about that is true and how to apply it.

Doing this will improve SNR a lot more than upgrading a 2SK170 to a 2SK660. And it will serve to help those in askance "what FET is better for my mic than 2SK170 that I can no longer buy".

Thor