Comparison of JFETs for mic applications

[ccaudle]

this method of biasing the 2SK660

I'm having trouble following, is this correct for the connections?


What is indicated with the connection (capacitor?) between source and case, and drain and case?

 

[k brown]

I'm having trouble following, is this correct for the connections?
View attachment 143471

What is indicated with the connection (capacitor?) between source and case, and drain and case?

Those are RF caps at the TB3M input jack of the pwr module (could have been drawn more clearly).

 

[thor.zmt]

I've always been curious about this method of biasing the 2SK660; it's from the Audio Technica AT8533 power module; so different from the way FETs are usually biased in microphones.

The PNP transistor acts in effect as a Zener Diode (or Vbe multiplier), which produces a fairly constant emitter voltage.

The emitter is supplied from a current source (CRD) from the centertap of the output transformer.

The 1k load for the Mic capsule FET comes from the PNP emitter. It seems to be a follower, with the Diode in the acting as "floating" powersupply of ~ 0.6V for the J-FET, that's DEEP in the triode region.

As the signal out of the J-Fet goes via a cap to the base of the PNP the PNP transistor in effect bootstrap's the 1k Resistor.

Overall capacitances are well cancelled, it's a fairly linear follower into a transformer.

It looks complex, but fundamentally it's simple.

I'm not surprised it sounds good, if we measure this circuit, I expect even order HD dominant with very low order of HD, plus likely a fair bit of pink noise.

Thor

 

[k brown]

The PNP transistor acts in effect as a Zener Diode (or Vbe multiplier), which produces a fairly constant emitter voltage.

The emitter is supplied from a current source (CRD) from the centertap of the output transformer.

The 1k load for the Mic capsule FET comes from the PNP emitter. It seems to be a follower, with the Diode in the acting as "floating" powersupply of ~ 0.6V for the J-FET, that's DEEP in the triode region.

As the signal out of the J-Fet goes via a cap to the base of the PNP the PNP transistor in effect bootstrap's the 1k Resistor.

Overall capacitances are well cancelled, it's a fairly linear follower into a transformer.

It looks complex, but fundamentally it's simple.

I'm not surprised it sounds good, if we measure this circuit, I expect even order HD dominant with very low order of HD, plus likely a fair bit of pink noise.

Thor

Thnaks

Did look like a bit of a bootstrap to me. Wonder why this type of bias isn't seen more often.

This is the Shure version (don't know for what FET):

 

[thor.zmt]

Did look like a bit of a bootstrap to me. Wonder why this type of bias isn't seen more often.

Most people get a headache looking at this.

The J-Fet operated at extremely low voltage.

There seem to be many thermally dependent voltages.

A lot goes against "received wisdom".

Peeps just "don't get it"!

Thor

"fast and bulbous

(Hey, do it again) I love that, I love those words

Fast and bulbous, that's right, the mascara snake

Fast and bulbous

Bulbous, also tapered (yeah but you gotta wait until I say it)

Also, a tin teardrop (oh, christ)

(Again, beginning)

Fast and bulbous, that's right, the mascara snake

Fast and bulbous, also a tin teardrop

Bulbous, also tapered, that's right

A squid eating dough in a Polyethylene bag

Is fast and bulbous, got me?"

 

[thor.zmt]

This is the Shure version (don't know for what FET):

Q115 Q103 (PNP) is the constant current source, there is implicit looped feedback from the follower (Sziklai circuit) emitter via the zener diode.

R115 (8.2k) with C110 acts as the "flying supply" for the J-Fet, with Q102 acting as the same follower as in the Audio Technica circuit.

The second follower is also switchable highpass and perhaps an overengineered part.

For a simple modern version with 21st century parts, use 2 X E102 CRD from pin 2/3 of XLR (2mA constant current), J-Fet and source resistor (say 10k for 500uA Idss) and BSS84 as follower and to bootstrap Cgd, coupling cap. Capsule to ground, J-Fet gate sits at ~5V.

As pin 2/3 with P48 will be at 41.2V, we can easily derive 36V bias Voltage for LDC Capsules. If using P68 (my "personal" standard) we get 56V Bias.

Marginally more components that the simplest possible circuit, but better performance all around, including noise.

Thor

 

[k brown]

Q115 (PNP) is the constant current source, there is implicit looped feedback from the follower (Sziklai circuit) emitter via the zener diode.

R115 (8.2k) with C110 acts as the "flying supply" for the J-Fet, with Q102 acting as the same follower as in the Audio Technica circuit.

The second follower is also switchable highpass and perhaps an overengineered part.

For a simple modern version with 21st century parts, use 2 X E102 CRD from pin 2/3 of XLR (2mA constant current), J-Fet and source resistor (say 10k for 500uA Idss) and BSS84 as follower and to bootstrap Cgd, coupling cap. Capsule to ground, J-Fet gate sits at ~5V.

As pin 2/3 with P48 will be at 41.2V, we can easily derive 36V bias Voltage for LDC Capsules. If using P68 (my "personal" standard) we get 56V Bias.

Marginally more components that the simplest possible circuit, but better performance all around, including noise.

Thor

Did you mean Q105 (CCS)?

 

[jp8]

The bootstrapped JFET principle was discussed on that other microphone forum and inspired me to build something similar, with such a circuit on both the inverting and non-inverting outputs of the microphone. The circuit would have zero electrolytics, lending itself very well to miniaturization. Gain close to 0dB, and according to simulations very low distortion, extremely high SPL handling and low noise. Everything except the low noise came out as predicted. It could handle 7V RMS on the input at 0.1% THD. Just noisy as hell and I still don't understand why. Maybe I'll publish it here later on GroupDIY.

Jan

 

[micolas]

Q115 (PNP) is the constant current source

Thor, please excuse me,
There is no BJT Q115 in the schematic

 

[thor.zmt]

Did you mean Q105 (CCS)?

Thor, please excuse me,
There is no BJT Q115 in the schematic

Q103... It was before my first coffee.



My brain and memory ain't what they used to be. Then again, they probably never were.

Thor

 

[thor.zmt]

The bootstrapped JFET principle was discussed on that other microphone forum and inspired me to build something similar, with such a circuit on both the inverting and non-inverting outputs of the microphone.

Theoretically good idea.

Everything except the low noise came out as predicted.

Excess noise from poor resistor choices? I had an SMD thick film accidentally in the tail of a folded cascode circuit where the input side was capable of < 0.5nV|/Hz and had 40mA/V transconductance.

Replacing with a generic (Yageo) thin-film resistor dropped noise a lot, going MELF got close to predicted results.

Thor

 

[micolas]

Q103... It was by my first coffee.

My brain and memory ain't what they used to be. Then again, they probably never were.

Thor

Ah, that's okay. Now it makes sense. After too few hours of sleep, before coffee and at my age, my eyes and mind still play tricks on me
Thanks for your objective analyzes of electronic circuits, it's a pleasure to follow your posts. And your humor is comforting Cheers!

 

[ricardo]

Correct, the datasheet shows the curves for 250uA Idss @ 10V. They obviously shift with Idss which is a bit variable, especially in older J-Fets.

The difference is ~ 50%.

My computer and office/lab is in boxes, until I unpack and set up, no model.

I pontificate slightly on this in FETbias.doc

I'll wait on your model. In the meantime, I'm trying out some easy stuff with the LTspice model. It's of course important that we are 'using' the same model.

It can plot against any axis format you elect to use. The stuff you dis.iss as "unnecessary bells and whistles".

Please use this "bell & whistle"; plotting noise spectral density against log frequency as it is fundamental to our comparison.

Can you leave Zeph out until.he starts using nbers that actually make sense and are within reasonable variations of reality?

I have zero confidence in what is in this doc.

Sigh! It may have escaped your notice that I did the measurements in Zephyr.doc ... and I used my own "bell & whistle" to show the noise spectrum. The exact details are in the document. As this is a constant relative bandwidth display, the details are important.

I'm not sure what numbers you are complaining about. I think you are querying the absolute level.

Alas, I don't have an absolute level calibration and I make this clear on pages 1, 7 & 10. The only number I have for the ECM 8000 is a cryptic "-60dB" number on the box. 2025 ECM 8000s have a different "70dB" cryptic number.

Jochen Schulze has 15mV/Pa while you claim 8mV/Pa

I'm not surprised that these 4 numbers are very different as I'm aware of at least 3 very different circuits for ECM8000. At least 2 of them are guilty of the heinous crime of taking power from only side of a P48 line .. but I forgive them cos there are a couple of big name Germans who are also perpetrators and SimpleP48 of course

I also explain my crude overload test and say "Can’t tell much from this VERY crude test except that overload probably very high" on page 2

But what have you got against the noise curves on page 10 ?

Absolute levels have no bearing on these curves which were all taken with the same setting on the preamp. So the info is all relative.

PLEASE post the circuits you want me to simulate as "SimpleP48" and "SimpleP48RCA", with all values according to the capsule and J-Fet as mentinoned.

I'm doing my best in between my beach bum activities. Dis LTspice sim stuff is new to me but I'm learning. I hope to have both circuits ready with the correct models by the time you have unpacked ... and hope to be also ready to tweak them with your model when available.

So far you have been endlessly prevaricating and sandbagging and moving goalposts, after first a lot of big gob trash talking.

There appears to be a lot of gob trash talking but I'm not sure I'm the generator

I already conceded that the theoretical electronic noise limit of the Schoeps derived circuit will be as much as 10dB greater, simply due to resistor noise.

I missed that Where did you post this?

BTW for the record, my claim is that Zephyr's Schoeps variant is 10dB quieter than yours with a capsule similar to his ... and SimpleP48 is perhaps a dB quieter than that.
When your TINA sims prove or disprove this claim, will you enlighten us as to why you think this the case?

The tradeoff is lower HD (dramatically so) and the ability to handle high SPL's, which is what I emphasised in my design.

If you want to use a modern electronic (not transformer) input mic pre (build into a USB soundcard) and you are not interested in low distortion at typical recording levels for music, but only in low noise, for very low sound levels.

But I AM interested in high SPLs & low distortion. That's the reason for SimpleP48RCA ... which BTW is on page 12 of SimpleP48.pdf ... as one of the 2 recommended variants. So I'm not sure it counts as "moving goalposts"

---

I think I've mostly avoided "wi**y wan*ing" in this post

With some luck we might all get something useful out of this if Thor & I stick to refining our models so the sims represent 'real life'

 

[thor.zmt]

I'll wait on your model.

You will wait forever. I'm not making a model. I told you, LT-Spice Library.

Please use this "bell & whistle"; plotting noise spectral density against log frequency as it is fundamental to our comparison.

Is it now? I suggest unweighted SNR 20-20kHz instead. It is a simple single unambiguous number, no debate. -127dB Ein beats -126dB Ein. No interpretation needed, no ambiguity and Curated Egging.

Sigh! It may have escaped your notice that I did the measurements in Zephyr.doc ...

Quel dommage!

No, it honestly never even remotely entered my mind they that....ahhhm...lost for words...ahhhm...thing with the professionalism, accuracy and genersl aptitude it evidences, might be associated with such august fellow as you, who designed soundfield microphnes at Calrec, is the only one in the world who knows how to calibrate ambisonic microphones, endorsed by all sorts of famous British engineer's working in the field of recorded sound.

I used my own "bell & whistle" to show the noise spectrum.

You mean you used generic software to do an FFT.

I'm not sure what numbers you are complaining about. I think you are querying the absolute level.

I am not complaining. I merely observe that ALL numbers are wrong. Observing, note. Not complaining.

Alas, I don't have an absolute level calibration

Then why even publish?

And why not pay a pittance and send your ECM8000 off for calibration? I did.

Or lacking this, BORROW a calbritated SPL Meter from your friendly neighborhood environmental officer, to see if your neighbors are too noisy. Might cost you a shout for a pint of Carlton.

The only number I have for the ECM 8000 is a cryptic "-60dB" number on the box.

Then why not make a comparison to a reference of known accuracy. The speaker you used as source may have full documented response and sensitivity, if not, it would probably have been possible to borrow one. Or as mentioned a well calibrated SPL meter might have been used.

Jochen Schulze has 15mV/Pa while you claim 8mV/Pa

That's the number in the calibration report I received, which I secondarily verified using a calibrated SPL Meter and AP2.

I'm not surprised that these 4 numbers are very different as I'm aware of at least 3 very different circuits for ECM8000.

I am aware of one circuit between the early 00's when I got my unit (with an RadioShack SPL Meter for SPL) and the mid teens, when I last purchased one for QC purposes and opened it up.

Many modifications abound on the net, these are not "original" ootb.


Transistor types varied widely, as did capsules. But these small electret capsules regardless of maker, tens to have similar sensitivity levels.

I found no real differences between the early 00's and the mid 10's unit

But what have you got against the noise curves on page 10 ?

You mean this:


I have nothing against it whatsoever. Mind you, I also see no possible use.

Absolute levels have no bearing on these curves which were all taken with the same setting on the preamp.

So you are comparing noise levels without correcting for sensitivity? I am unsure what you are trying to do. But whatever it is, its pointless.

So the info is all relative.

Yes, relative on multiple levels. So the lowest noise microphone in this chart, may actually also have low sensitivity, so in practice it would have the greatest actual noise.

Absolute reference levels and calibration are essential for measurements to have meaning.

I'm doing my best in between my beach bum activities. Dis LTspice sim stuff is new to me but I'm learning.

Why do you insist on not using TINA-TI but instead the Spice simulator with the least intuitive, least user friendly interface available?

There appears to be a lot of gob trash talking but I'm not sure I'm the generator

Well, I don't post how my circuits are so much better than everyone else's and how great a fellow I am.

I am only interested in what is right and to promote learning, not to big up my ego. I'm here to learn more than I teach hopefully.

I missed that Where did you post this?

A few posts back I did a rough noise analysis on paper.

In my Schoeps variation, my desire to handle high SPL's and to use the 2SK660 FET caused me to set the split load phase inverter/buffer resistors at 10kOhm for 0.5mA Idss FET's.

The "generic" version uses 2.2k. this will cause around 7dB more noise for my version with 2SK660.

However, by not using the 2SK170 I would probably claw back around 3dB of that noise. And by using diode biasing instead of 1G resistors we probably end up not sll that far apart in practice.

So practically, maybe a little advantage for the "generic" circuit, but unlikely 10dB.

I could complicate the circuit by adding FET's in parallel (5 pcs give 2.5mA and ~ 5V across each load resistor). Or I could make some kind of compound circuit that draws more current.

I found the noise of the original low enough for my uses and usefully lower than many generic china microphones sold sub 1k as recording microphones and left it at that.

As said, this was not a rigorous design for best performance.

It was simply a quick and dirty hack to a bunch of BM-X00 bodies, with a bunch of 34mm Capsules off Taobao and a hacked up BM-X00 PCB's to make of useful studio recording mic's with s few different "flavours", that did better all around than the kind of Mic's selling for ~ 500 USD equivalent on Taobao.



Not a huge love affair, but more like a quickie in the back at Coyote Ugly, fast, satisfying, few strings attached and generally guilt free.

BTW for the record, my claim is that Zephyr's Schoeps variant is 10dB quieter than yours with a capsule similar to his ... and SimpleP48 is perhaps a dB quieter than that.

Yes. And even basic back of envelope calculations call that into question.

When your TINA sims prove or disprove this claim, will you enlighten us as to why you think this the case?

I will not waste dny more time, except to input the SimpleP48 circuit you give to me, place it in the schematics sheet as my own send run MY analysis, plus perhaps yours...

But I AM interested in high SPLs & low distortion. That's the reason for SimpleP48RCA ... which BTW is on page 12 of SimpleP48.pdf ... as one of the 2 recommended variants. So I'm not sure it counts as "moving goalposts"

I do not see a complete circuit. What I see is the "Linkwitz Mod" (yes, another bloody Kraut, I know) being misattributed:


What I do not see is a circuit I can input into a simulator.

---

With some luck we might all get something useful out of this if Thor & I stick to refining our models so the sims represent 'real life'

As remarked before, my models are sufficiently refined and validated with real circuits and test gear.

And all my test gear is calibrated as well.

Thor

 

[jp8]

Theoretically good idea.



Excess noise from poor resistor choices? I had an SMD thick film accidentally in the tail of a folded cascode circuit where the input side was capable of < 0.5nV|/Hz and had 40mA/V transconductance.

Replacing with a generic (Yageo) thin-film resistor dropped noise a lot, going MELF got close to predicted results.

Thor

Thanks Thor for your comments. The few resistors in the circuit are all connected to low-impedance nodes and/or have only a low voltage across the resistor. So I expect little (noticeable) excess noise from the resistors. The noise also seems broadband, and not limited to 1/f noise. Furthermore, I have never had (noticeable) noise problems with the 0805 thick-film resistors from Yageo, Bourns and Vishay that I normally use. But I will keep an open eye for all possible causes and will also order thin-film resistors with my next Mouser order.

Jan

 

[thor.zmt]

Thanks Thor for your comments. The few resistors in the circuit are all connected to low-impedance nodes and/or have only a low voltage across the resistor. So I expect little (noticeable) excess noise from the resistors. The noise also seems broadband, and not limited to 1/f noise.

Noisy resistors is still my guess.

My case was 4 x 2SK170GR in parallel ~ 5mA per fet) switchable source resistor to ground to set transconductance.

Fixed extra low noise CCS to -6V to keep the source node at 0V (servo) with 6V across the 2SK170.

I had a BC860C as folded cascode, appx 100R to a super low noise + 9V PSU, servo to BC860 base to keep output at 0V.

Collector from BC860 to a -6V rail via a low noise 10mA CCS.

Output from a resistor to ground in parallel with a capacitor implementing a 75uS turnover.

So a 100R resistor at ~ 3V was adding 20dB or so excess noise and it was not 1/LF at that.

And the emitter node of a BC860 is a few Ohm at 10mA collector current.

Furthermore, I have never had (noticeable) noise problems with the 0805 thick-film resistors from Yageo, Bourns and Vishay that I normally use. But I will keep an open eye for all possible causes and will also order thin-film resistors with my next Mouser order.

Try identifying the hidden, high noise sensitive node in the circuit, to take out the guesswork.

There is one, or there would not be noise.

A other ready suspect are unfiltered or insufficiently filtered zenner diodes and fake J-Fets.

Thor

 

[jp8]

Try identifying the hidden, high noise sensitive node in the circuit, to take out the guesswork.

There is one, or there would not be noise.

Sure, that's what I tried all the time... But to no avail.

A other ready suspect are unfiltered or insufficiently filtered zenner diodes and fake J-Fets.

There is a zener in the non-driven leg, between collector of the cascode PNP transistor, powering the bias generator. No connection to the audio path. But noise is still there when shorted to ground and disabling the bias generator.

JFET is 2SK209 from Mouser. I don't trust any semiconductor if it's not from an official distributor. That said, I did buy some 2SK660 and other JFETs from Aliexpress, just out of curiousity, but I'm not expecting anything good from them. Haven't used them yet and it has low priority.

Jan

 

[thor.zmt]

Sure, that's what I tried all the time... But to no avail.

Do you want to post the circuit?

Thor