Comparison of JFETs for mic applications

[ricardo]

Drain voltage thus is 48V - ~17V = ~31V.

I mean this is an EE101 calculation. Nothing but Ohm's law.

Thanks for this Thor.

I make it 29v4. What does TINA say?

 

[thor.zmt]

Thanks for this Thor.

I make it 29v4. What does TINA say?

I just started to install the Ceiling light for my office. A huge, fairly heavy multiple ring, vaguely space styled (think mid century modern) thing.

I found the sheetrock quite rotten, so simple sheetrock anchors would not do, so I shot a bunch of construction adhesive into the ceiling. It will dry overnight.

I'll set all of the office tomorrow, after fixing the light.

From extremely hard stone cladding on bathroom walls that kept stripping my diamond drill bits and made me wish I had gotten a Bosch hydraulic Hammer drill instead of a Chinese "Bazooka" Hammer Drill, I had any annoying problem possible, renovating this place. So bear with me.

I can answer you after the office and lab is back up. But I suspect TINA will be on a similar voltage.

To be honest, with P48 actually being 44...52V via 6k8 x 2 at 10% tolerance, the precise Voltage will be subject to quite a range, so excessive precision here is probably pointless.

Vth on BSS84 I encountered across various vendors have been pretty consistently 1.5...1.9V, with device to device matches on a reel very good. Noise is also usually on spec, meaning ~ 1.5nV|/Hz.

BSS84 is very preferred part to use for me, together with 2N7002 and STN1NK60Z for MOSFET's, which in turn I like to use a lot as active parts in discrete designs.

A "standard" circuit of mine is MMBFJ113 (J113 in SMD) or LSK170 + BSS84 and STN1NK60Z as output, matching the old 3-transistor designs with BJT's but with vastly improved performance by using modern devices.

Note, this is not for microphones or even transformerless microphone preamps, but line level.

Thor

 

[ricardo]

I'm quite excited about our little SPICE comparison.

I've never used LTspice for noise before and it is showing and confirming stuff which Guru Wurcer and I played with at the beginning of this Millenium ... all in MicBuilders for those really interested and maybe his Linear Audio articles.

Let's tie down some important but non-controversial details.

---

THOR'S CAPSULES

You proposed 2 capsules. The first was a 34mm 'Brass' capsule of 68p capacitance and -34dBV/Pa sensitivity @ Vp = 60V

This is appropriate for the comparison between your circuit and Zephyr's. It is what Zephyr had on his capsule. I'll allow you to assume Vp = 60V though your circuit only has about 42V

The other was an electret 26mm, 42p and -35dBV/Pa. This is good to compare your circuit with SimpleP48 & SimpleP48RCA

Are you happy with these choices of capsules you recommended?

---

MIKE PREAMP

In #319, I assume a simplified version of Fig 9 in THAT Corporation 1510/1512 Datasheet rev9 5oct2017.

You assume a slightly different 'simplified' preamp in #245

Which one should we use?

I'm happy to use yours cos it sometimes gives SimpleP48 & SimpleP48RCS a slight advantage

 

[ricardo]

ZEPHYR'S CIRCUIT

I apologise for, not so much moving goalposts, as having difficulty remembering his circuit from 2019. I'm going diving this weekend so his marked up print might emerge from under da diving junk.

The aim of the exercise is to sim what I measured in feb2019 to check my claim that your circuit has >10dB more noise above 4kHz

I'm sure of ..

• C10 is grounded ... cos I refer to that in Zephyr.doc

But in Muntzing all the stuff to get #332, I forgot this changed the current draw which might adversely affect the yucky regulators ... which is why I asked if I could add a single resistor to replace the Muntz'd stuff.

If you are unhappy with that, please put all the stuff I removed back so the current consumption is as the original ... but GROUND C10 as that is how Zephyr had it for the Cardioid position (your single diaphragm requirement).

When you get TINA up & running and post a circuit we can check if this an accurate representation of what he built and I tested.

When I find Zephyr's marked up print, we can put his pot in the correct spot and assess whether it would have changed our sims.

 

[thor.zmt]

THOR'S CAPSULES

You proposed 2 capsules. The first was a 34mm 'Brass' capsule of 68p capacitance and -34dBV/Pa sensitivity @ Vp = 60V

This is appropriate for the comparison between your circuit and Zephyr's. It is what Zephyr had on his capsule. I'll allow you to assume Vp = 60V though your circuit only has about 42V

The other was an electret 26mm, 42p and -35dBV/Pa. This is good to compare your circuit with SimpleP48 & SimpleP48RCA

Are you happy with these choices of capsules you recommended?

Let's keep it simple and a level playing field, using only the 26mm Model. That way all circuits work under identical conditions.

---

MIKE PREAMP

In #319, I assume a simplified version of Fig 9 in THAT Corporation 1510/1512 Datasheet rev9 5oct2017.

Why don't you ever post schematics?


I will use this. It's from the datasheet, quickly reduced to the essential.

In addition I will place a VCVS (voltage controlled voltage source) at the output, to allow each circuits gain to be normalised losslessly, while capturing all effects of the presumed preamp input.

This will make sure we get comparable results on a single plot.

If I can find the file, I will also add an A-weighting filter to each circuit.

Thor

 

[MicUlli]

A-weighting is done by this (nearly noisless) filter (for example). Dont try to build it in reality with discrete components
Best regards
MicUlli

 

[thor.zmt]

A-weighting is done by this (nearly noisless) filter (for example). Dont try to build it in reality with discrete components

Thank you, I have a simpler circuit. Just need to find it.

Here is one that can be build and integrated into a test setup:

https://jensign.com/AWeight/index.html

Thor

 

[Dylan W]

Vth on BSS84 I encountered across various vendors have been pretty consistently 1.5...1.9V, with device to device matches on a reel very good. Noise is also usually on spec, meaning ~ 1.5nV|/Hz.

BSS84 is very preferred part to use for me, together with 2N7002 and STN1NK60Z for MOSFET's, which in turn I like to use a lot as active parts in discrete designs.

A "standard" circuit of mine is MMBFJ113 (J113 in SMD) or LSK170 + BSS84 and STN1NK60Z as output, matching the old 3-transistor designs with BJT's but with vastly improved performance by using modern devices.

Note, this is not for microphones or even transformerless microphone preamps, but line level.

Thor

This is interesting, what advantages do you find using MOSFETs over BJTs? Improved linearity, or maybe more favorable harmonic profile? Would be interesting to try a MOSFET ring of three in something like a Neve BA283. Would you expect it to sound significantly different?

 

[thor.zmt]

This is interesting, what advantages do you find using MOSFETs over BJTs?

MOSFETs have a close to infinite gate impedance, so for example my "MOS-SCHOEPS" uses 1M from gate to source (bootstrapped together with 73pF Cgs) and 4.7MOhm (in parallel with 5pF Cgd) between gate and drain.

So, the input impedance of my MOS buffer is 4.7MOhm // 5pf compared to (say) 135kOhm // 8.5pF for using 150k + 2N4403.

More critically perhaps, the input impedance is not signal/load dependent the way BJT's are. There is a reason that the latest TI audio Op-Amp's that have super low distortion, low noise.AND offer good subjective sound are all done on CMOS processes, with not a BJT in sight.

At 1mA collector/drain current output impedance's are similar.

As Class A follower MOSFET have a more tube like harmonic spectrum, for whatever that is worth.

Would be interesting to try a MOSFET ring of three in something like a Neve BA283.

As I'm more of a tube guy, I'd render it more as a 3-triode, 2 stage + follower Circuit.

My own take for that has a J113 (well, the SMT Version) as input, the BSS84 as direct coupled second stage, with STN1NK60Z as follower/active load for BSS84. CCS tail for STN1NK60Z. Loop feedback to J113 source.

Would you expect it to sound significantly different?

Yes, very.

The J113/BSS84/STN1NK60Z circuit sounds a lot like a classic 12AX7 2 stage Amp with 12AU7 follower and loop feedback, but with less obvious colorations. It is basically clean and uncoloured, with just a hint of "warmth" and colour.

Though by lowering loop gain and thus increasing HD we can get a more stereotypical "Tube" sound.

Simple BJT circuits by comparison are always slightly grainy and opaque sounding.

Of course, once we throw in transformers, they will contribute a lot of the sound.

Thor

 

[ricardo]

Let's keep it simple and a level playing field, using only the 26mm Model. That way all circuits work under identical conditions.

I'm happy with that. Just to confirm, it is

26mm, 42p and -35dBV/Pa.

View attachment 143882
I will use this. It's from the datasheet, quickly reduced to the essential.

I'm happy with that.

But can you explain what you mean by

I will place a VCVS (voltage controlled voltage source) at the output, to allow each circuits gain to be normalised losslessly, while capturing all effects of the presumed preamp input.

This will make sure we get comparable results on a single plot.

Why can't we just measure the output across R1 & 2?

LTspice allows you to plot the difference between 2 nodes. This 'difference' can be signal, noise or other stuff depending on your type of analysis.

If I can find the file, I will also add an A-weighting filter to each circuit.

That would be 'interesting' as long as we don't forget the main and original 'noise interest'. Comparing noise spectra above 4kHz

For those interested in the subject, there is a Jurassic MicBuilders discussion on noise weightings with Dipl. Ing. Wuttke of Schoeps, myself and various true and pseudo gurus. The concensus was (surprise, surprise) A wtg had little to do how quiet a mike sounded or the annoyance of the noise .. but yus true mike gurus will already know that.

 

[thor.zmt]

Why can't we just measure the output across R1 & 2?

Because each circuit has different gain. So direct measurements are not useful. A circuit with high gain will have more absolute output noise, but also mor signal.

LTspice allows you to plot the difference between 2 nodes. This 'difference' can be signal, noise or other stuff depending on your type of analysis.

So can TINA of course.

That would be 'interesting' as long as we don't forget the main and original 'noise interest'. Comparing noise spectra above 4kHz

Which is pointless unless we can model Brownian motion noise of the capsule and other related noise sources.

Otherwise HF noise will likely be dominated by this, so the electronic noise is irrelevant, if low enough.

My issue is how to add a good model of Brownian motion noise to my capsule model.

For those interested in the subject, there is a Jurassic MicBuilders discussion on noise weightings with Dipl. Ing. Wuttke of Schoeps, myself and various true and pseudo gurus. The concensus was (surprise, surprise) A wtg had little to do how quiet a mike sounded or the annoyance of the noise .. but yus true mike gurus will already know that.

This is a whole other debate.

Noise audibility vs SPL and Spectrum is WHAT SHOULD inform our tests, but it's not. Just like THD testing ignores relative audibility of distortion components.

So measured noise, either A-Weighted or unweighted tells nothing how audible or objectionable this noise will be, while measured THD tells nothing how audible or objectionable this distortion will be.

But as said, this is a separate debate.

Thor

 

[ricardo]

Why can't we just measure the output across R1 & 2?

Because each circuit has different gain. So direct measurements are not useful. A circuit with high gain will have more absolute output noise, but also mor signal.

Of course ! Hence the Adjusted for on-axis sensitivity curves on page 11 of Zephyr.doc

But I withdraw my question about your VCVS. I've had a good look at the beast in LTspice and it greatly simplifies my attempts to replicate your sims.

Even dis old beach bum can lern nu tricks

That would be 'interesting' as long as we don't forget the main and original 'noise interest'. Comparing noise spectra above 4kHz

Which is pointless unless we can model Brownian motion noise of the capsule and other related noise sources.

Please stick to YOUR original challenge to my statement that YOUR circuit has >10dB more noise than Zephyr's above 4kHz.

I'm sure your facility with modelling Brownian noise is better than that of a beach bum who last did such hard sums more than 45 yrs ago.

What we CAN do is compare the simulated electrical with the measured noise. If the measured noise is only slightly above the the simulated electrical noise, we can make a good estimate of the level of Brownian (and other) noise ... at least for our frequency bands of interest above 4kHz.

eg If the measured noise is 3dB above the sim'd electrical noise, the Brownian and other noise sources will have about the same power as the electrical noise in that frequency band. If less than 3dB above, the electrical noise is 'dominant'.

Noise audibility vs SPL and Spectrum is WHAT SHOULD inform our tests, but it's not. .... So measured noise, either A-Weighted or unweighted tells nothing how audible or objectionable this noise will be,

I'll refer those interested in the audibility and nuisance value of noise to the Jurassic discussion in MicBuilders where several gurus chime in who have conducted carefully controlled listening tests and other guru stuff on the topic.

For our little comparison, let's just plot the noise spectrum above 1kHz. This covers my claim nicely

while measured THD tells nothing how audible or objectionable this distortion will be.

Thanks for this Thor. I'll remember to quote you when your sims show your SupaDupa circuit has a zillion times less THD than SimpleP48, SimpleP48RCA or Zephyr's version of the Schoeps. As I'm sure they will ... cos you've told us so many times

 

[thor.zmt]

Please stick to YOUR original challenge to my statement that YOUR circuit has >10dB more noise than Zephyr's above 4kHz.

Let's see, but I remembery "back of envelope" noise calculations that you poh-pohed suggest it would have 13dB more noise in the flat band, contributed by resistors, but the 1G + capsule pink noise would be dramatically lower.

So I think I already demonstrated that my circuit should have greater flat band noise than "Simple P48 (zephyr we leave for later) but less LF noise.

I'm sure your facility with modelling Brownian noise is better than that of a beach bum who last did such hard sums more than 45 yrs ago.

To do it right is non-trivial. I'm kinda hoping to somehow use a simple RC circuit to simulate violet noise of appropriate levels.

What we CAN do is compare the simulated electrical with the measured noise. If the measured noise is only slightly above the the simulated electrical noise, we can make a good estimate of the level of Brownian (and other) noise ... at least for our frequency bands of interest above 4kHz.

Indeed.

eg If the measured noise is 3dB above the sim'd electrical noise, the Brownian and other noise sources will have about the same power as the electrical noise in that frequency band. If less than 3dB above, the electrical noise is 'dominant'.

This is not quite correct. If we have uncorrelated noise with identical spectrum (frequency response) and level, total noise will increase by 3dB.

Noise adds root sum square.

One noise source 3dB quieter still contributes to total noise.

1.4uV ^ 2 + 1uV ^ 2 = 2.96uV ^ 2

SQRT(2.96) = 1.72uV

QED

For our little comparison, let's just plot the noise spectrum above 1kHz. This covers my claim nicely

Nope, let's plot SNR and effective output noise density in V|/Hz from 20 Hz to 20kHz, as this is what most people are interested in.

Thanks for this Thor. I'll remember to quote you when your sims show your SupaDupa circuit has a zillion times less THD than SimpleP48, SimpleP48RCA or Zephyr's version of the Schoeps. As I'm sure they will ... cos you've told us so many times

On top of that, the spectrum of harmonics (say we use the weighting of Distortion proposed by D.E.L. Shorter of the BBC in the 1950's, nothing too newfangled and unproven) will induce less audibility of what Distortion is there.

Extra reading accessible directly: (ignore the malware warning, the site is legit):

http://www.nutshellhifi.com/library/tinyamps.html

Thor