Comparison of JFETs for mic applications

[AdamFrisen]

I found this from former technical director of Neumann in US:



from: https://repforums.prosoundweb.com/index.php?topic=36194.0

 

[abbey road d enfer]

I found this from former technical director of Neumann in US:

That's a very interesting post, however there is an element of confusion, which has been noted by Klaud Heyne.
The text refers to THD being lowered by decreasing the capacitive load, but improved by the presence of a NFB capacitor. This is all true, but slightly confusing, because the mechanisms must be considered as two totally different sources of distortion.
Distortion caused by parasitic capacitance is purely physical and does not pertain to the head amplifier, whether tube or FET. It acts on the global transductance of the system.
NFB capacitor reduces the THD of the amplifying stage, but, as it increases the apparent parasitic capacitance, it increases the capsule's transductance THD, so there are two antagonistic effects at work.
In most head amps, amplifying stage's THD is dominant, so using a NFB capacitor ("charge amp") improves THD at high spl, which is what most users are concerned with.

 

[Voyager10]

The ORS87 measurements to date have used a fixed 68pF input capacitor (for comparison with other circuits). We should have figures for 100pF (original U87) and 50pF (U87A in cardioid), as this will affect the gain and therefore THD.

If the U87 output transformer is 9.26:1, we'll need the FET gain to be about 5.8 to match the U87 numbers. With a 100pF inout cap that should be close. For the U87A the FET gain should be 9.26, which is some way off from my numbers.

Re. Abbey's point - yes, that's confusing. The capsule generated THD will go down, but that's not what is being measured when applying signal to the calibration input.

 

[kingkorg]

FWIW. By modern standards both u87, and km84 have mediocre THD performance. Just like most commonly replicated tube mics. However people seem to like that, as most modern low THD mics are frowned upon.

So do we want more of that "vintage THD", or less, or the same exact amount, or maybe focus on harmonic profile throughout the spectrum instead of TOTAL harmonic distortion, mostly expressed as a meaningless figure at 1Khz?

Or maybe just let it go, as most voices and commonly recorded instruments are never going to reach that SPL level, nor trigger any meaningful harmonic generation?

EDIT:
Here's u87ai 2nd and 3rd profile at 1% THD @1K. Clearly, simple 1%THD @1K figure is deceiving. Harmonic content generated by 1K content, will sound significantly different than 10K. Cymbals, sibilants? And C12 circuit, at slightly higher THD, all 2nd harmonic, except at low end where 3rd is generated by the transformer. Basically, total opposite of solid state. Ignore SPL axis, this is injection test.

 

[abbey road d enfer]

Re. Abbey's point - yes, that's confusing. The capsule generated THD will go down, but that's not what is being measured when applying signal to the calibration input.

So do we want more of that "vintage THD", or less, or the same exact amount, or maybe focus on harmonic profile throughout the spectrum instead of TOTAL harmonic distortion, mostly expressed as a meaningless figure at 1Khz?

It seems (correct me if I'm wrong) many prefer charge amp types over unloaded capsules.
A big difference is that capsule distortion is essentially 2nd harmonic, when head amp is mainly third.
Will it fuel the debate over "euphonic distortion"?
I fully agree with weak significance of THD at 1kHz; same for noise, for which spectrum is even more important.

 

[Voyager10]

OK, perhaps I've been misleading people here - the "THD" numbers here are largely a stand-in for headroom. For all these measurements, generally THD increases gently with signal level until it hits 0.5% or so, then all hell breaks loose:



(I didn't publish this chart earlier, it didn't seem to show anything exciting...)

All the THD numbers are telling you is how close to this threshold you are. Perhaps I should be reporting "max signal level for 0.5% THD" as with earlier posts - it's more time-consuming to measure though.

Certainly this one number won't tell you how a mic sounds when it's in its comfort zone, please don't read significance into the difference between 0.14% and 0.15% - that really is meaningless.

 

[Voyager10]

NFB Capacitor selection for ORS87Plus​

I've plotted the gain and maximum input/output levels for a variety of FETs in the ORS87+ circuit. A couple of changes were made to make it closer to a real U87 microphone circuit:

• Input capacitor now 47pF, close to the 50pF for a single-sided capsule
• Load resistor now 82K, representative of a 1K preamp input through a 9:1 output transformer

All measurements made at 1KHz, using 8pF, 5pF, 3pF (2 x 6pF in series) and no capacitor for C4.

Gain vs NFB cap​

Max input level vs NFB cap​

This is the maximum input level (V RMS at 1KHz) for 0.5% THD, as measured by REW.



I've also attached a plot of maximum output level (i.e. input level multiplied by gain) for reference.

Observations​

• The 2SK170, much used in "ADK" / MXL2001-style mics, has very low gain in this circuit, most likely because of a high input (Cgs) capacitance. Probably not suitable for charge-amplifier applications.
• With no C4 ("0pF") all other FETs have higher gain and lower max input level than is reported for the U87A. It's possible the actual U87A has more parasitic capacitance than the ORS87+, due to the latter's compact PCB layout.
• The J113 doesn't do particularly well at high gain / low C4 values.
• The J305 (InterFET, from Mouser) comes out best overall, it matches or exceeds the U87A figures for C4 in the 2.5-3pF range.
• OnSemi BF256B is a very close second, for a fraction of the cost of the J305 (11p each, 10 off, from CPC!) and also meets the U87A figures, for C4 ~ 2.5pF.

 

[Voyager10]

Charge amp noise measurements​

These measurements made with Cin = 68pF, and C4 = 10pF, which is comparable to the circuits used in earlier posts, especially the 'ADK/MXL2001' circuit. These aren't necessarily the right figures for a U87 application (Cin = 50pF).

The graph looks like this:



The A-weighted numbers for equivalent input noise are:

Device	Ein (dBV)
J113	-123.7
2N3819	-123.0
BF256B	-122.2
J305	-122.8
2SK117	-118.2
2SK30A	-121.4

These are looking noticeably better than the ADK circuit (post #33), which suggests removing the 2K2 source resistor and controlling gain with a feedback capacitor is a good move.

The 2SK117 has noticeably worse LF noise than the others; I've tried with two different devices, checked the bias, recalibrated REW, cleaned the board etc., but the same results. Not sure what's going on there.

(The 'effective noise floor' is not from some crazy liquid-nitrogen-cooled preamp, it's because I'm measuring the output noise (at the FET drain), and the circuit gain is between 4 and 5.).

 

[Voyager10]

Mo' FETs mo' data​

I've measuring some more FETs, from @GJB17's stash (see post #69). The devices not previously covered were (all Toshiba):

• 2SK246-BL
• 2SK369-V
• 2SK373-GR
• 2SK709-V

First up was the ORS87+ circuit, with the same setup as post #87: 47pF source capacitance and 82K output load. Bias was adjusted to allow the highest input signal level for 0.5% THD at 1KHz, then gain was measured. This was done for a variety of NFB capacitor values (C4 in the ORS87 circuit).

I've included the previously-measured 2N3819 values as a reference:




For reference, the DC bias points were as follows:

Device	Id (mA)	Vgs (V)	Rbias (calculated)
2SK246-BL	0.35	-3.15	9.0K
2SK369-V	0.21	-0.57	2.7K
2SK373-GR	0.35	-0.92	2.6K
2SK709-V	0.30	-6.0	20K

 

[Voyager10]

Noise measurements​

These are using MicUlli's circuit (post #38), with a 4.7uF bootstrap cap (post #39). I've also included BF256B and 2SK3557 measurements, which weren't previously included.

As before, measurements were made with 1nF and 68pF source capacitances.

Device	R4	Vds (V)	Gain 1nF (dB)	THD 1nF (%, 1KHz, 100mV)	Gain 68pF	THD 68pF (%, 1KHz, 100mV)	Ein, 1nF	Ein, 68pF
2SK246-BL	68K	1​	-0.21​	0.0026​	-0.29​	0.0029​	-124.0​	-120.8​
2SK369-V	33K	1.3​	-0.16​	0.0015​	-0.25​	0.0023​	-127.6​	-122.4​
2SK373-GR	33K	1.37​	-0.18​	0.0021​	-0.25​	0.0023​	-122.2​	-119.4​
2SK709-V	47K	1.9​	-0.2​	0.002​	-0.27​	0.0021​	-124.2​	-121.3​
2SK3557	33K	1.9​	-0.19​	0.0024​	-0.24​	0.0026​	-127.5​	-122.4​
BF256B	47K	1.2​	-0.17​	0.0014​	-0.22​	0.0016​	-123.2​	-121.4​

The plots below include the 2SK209 figures, which is amongst the best previously measured, for comparison.

 

[OneRoomStudio]

How variable do we think these devices are unit-to-unit? Are you testing just one of each, or if not, how many until you've established "average" behavior of each device? In my experience, most FETs vary widely.

 

[MicUlli]

Noise measurements​

These are using MicUlli's circuit (post #38), with a 4.7uF bootstrap cap (post #39). I've also included BF256B and 2SK3557 measurements, which weren't previously included.

As before, measurements were made with 1nF and 68pF source capacitances.

Device	R4	Vds (V)	Gain 1nF (dB)	THD 1nF (%, 1KHz, 100mV)	Gain 68pF	THD 68pF (%, 1KHz, 100mV)	Ein, 1nF	Ein, 68pF
2SK246-BL	68K	1​	-0.21​	0.0026​	-0.29​	0.0029​	-124.0​	-120.8​
2SK369-V	33K	1.3​	-0.16​	0.0015​	-0.25​	0.0023​	-127.6​	-122.4​
2SK373-GR	33K	1.37​	-0.18​	0.0021​	-0.25​	0.0023​	-122.2​	-119.4​
2SK709-V	47K	1.9​	-0.2​	0.002​	-0.27​	0.0021​	-124.2​	-121.3​
2SK3557	33K	1.9​	-0.19​	0.0024​	-0.24​	0.0026​	-127.5​	-122.4​
BF256B	47K	1.2​	-0.17​	0.0014​	-0.22​	0.0016​	-123.2​	-121.4​

The plots below include the 2SK209 figures, which is amongst the best previously measured, for comparison.
View attachment 142407View attachment 142408

Hi Voyager10
I want to congratulate to this time consuming and deeply interesting work you have done for the DIY community. Thanks to your investigations some myths are busted now.
Best regards MicUlli

 

[Voyager10]

How variable do we think these devices are unit-to-unit? Are you testing just one of each, or if not, how many until you've established "average" behavior of each device? In my experience, most FETs vary widely.

Good question. Generally, I'm testing with a single device, not batches. For some devices (J113, BF256B, 2SK209) I've used several in a number of builds and they all had measurements which agree with the test devices. So far it seems units within the same batch are pretty consistent.

The outliers I've had are 2SK170s, where a batch I had from eBay had 2-3dB worse noise than others (recovered from a dismantled mic PCB).

 

[abbey road d enfer]

The outliers I've had are 2SK170s, where a batch I had from eBay had 2-3dB worse noise than others (recovered from a dismantled mic PCB).

It seems that 2SK170 is one of the most counterfeited items.

 

[jp8]

The outliers I've had are 2SK170s, where a batch I had from eBay had 2-3dB worse noise than others (recovered from a dismantled mic PCB).

I only buy from trusted sources like Mouser and Digikey. I've bought several "original" (yeah, sure, haha) audio ICs on Aliexpress. All of them were counterfeits. Don't try to save a few pennies on JFETs when you're spending hundreds of dollars/euros on capsules...

 

[Khron]

I only buy from trusted sources like Mouser and Digikey.

Well, it doesn't help that 2SK170 has been discontinued for a while...

 

[soulsonic]

Interesting thread! I just wanted to chime in that I started using 2SK880 in my guitar pedals around 11 years ago when my old favorite 2SK117 became total unobtainium. I have been very satisfied with the results.

I hope this thread will help to show anyone new to the hobby that there are several fine choices available in current production devices, so that they don't feel compelled to go chasing after unobtainium JFETs with the mistaken idea that they are better.

 

[micolas]

Hi @soulsonic !
You have practically tested and you can
confirm that 2sk880 is the smd equivalent of 2sk117?

 

[thor.zmt]

Guys,

All this hard empirical work is commendable.

But we already have a good working understanding of what goes on and why we see these results.

Plus we can predict why all these Japanese J-Fets dedicated to microphone applications while eliminating gigaohm resistors are a really good option in microphones (I notice, non where tested, everyone pulls them out where they see them, meanwhile I refuse to use anything else except 2SK660 in mic's).

What are our noise sources, other than the J-Fet?

First, Brownian motion of the air molecules impacting on the diaphragm. This is the ultimate limit. And the larger the diaphragm, the more noise, this is white noise. SDC's are at advantage here.

Second, the noise of the bias resistors combined with the capsule capacitance. The capacitance itself is noiseless (other than Brownian motion noise), but resistors are not.

A 1GOhm resistor has ~0.6mV Johnson-Nyquist noise. So next to a -40dB sensitivity microphone this would be a 23dB SNR at 1PA (94dB).

But this noise is lowpassed by the capsule capacitance. With 1GOhm and 68pf we get an ~ 2Hz lowpass, so at 20Hz our SNR has improved to to 43dB and at 200Hz to 63dB (unweighted, spot).

An interesting factor is that while Johnson-Nyquist Noise rises by 10dB for a 1:10 increase in resistance the lowpass of resistor and capacitor shifts so that at any spot frequency above the corner noise is reduced by 20dB. So more resistance is good, as is more capacitance.

This resistor noise and RC lowpass are the core noise sources at low frequencies in all traditional capacitor microphone circuits at LF and cause the distinctive "pink" shape of LF noise in capacitor microphones.

Once frequencies become high enough the 20dB/Decade falling slope of the "pink" bias resistor noise intersects with the "flat white" Brownian motion noise.

So as long as resistors have no excess noise and gate noise current of the J-Fet is low enough, theory entirely predicts circuit noise.

The J-Fets voltage noise really needs to extremely sh!tty before it features in the equation.

So why measured differences, why different HD and why is the 2SK170 (and any derivates) such a garbage FET for microphones, despite Phanatically low noise and Phanatically high Transconductance?

It's the parasitic capacitance.

Let's take a J-Fet designed for microphones.

It's input resistance (shown on the datasheet and listed as actual value in the datasheet at a value that causes everyone to immediately reject it as garbage, as if the engineers at Sanyo, Toshiba etc. Ard completely clueless idiots) is actually the forward biased diode of the gate2source junction and an additional reverse polarity diode between Gate & Source.

If the voltage between gate and source is zero, the current flow is zero, so the actual input impedance tends towards infinity.

The greater the voltage between gate and source, the lower the impedance.

The FET has a low and fairly linear transconductance and usually Phanatically low capacitance (as these FET's are mainly for 3mm electret capsules).

The result, correctly applied the FET has very low noise, lower than circuits with expensive gigaohm resistors.

The J-Fet capacitance forms a voltage divider with the capsule, plus Mister Miller has its say. The lower this capacitance, the less signal is lost and the closer the circuit performs to the theoretica ideal.

Less transconductance is less gain, but also better linearity. The Idss of the microphone FET's is such that no degenetion is needed, they usually work "nekkid" so no added noise.

Good inherent linearity and low input capacitance (including Mr Miller) make for a linear Output and noise close to the theoretical ideal.

So why is the 2SK170 so awful?

It is not. The kind of J-Fets it represents are awesome on many applications. Want a Poweramp with 30dB gain and 115dB(A) SNR at 1W? You can, with 2SK170.

But the low noise and high transconductance come at the cost of high and non-linear capacitance.

And that is why a "designed for microphones" Fet like the 2SK660 is the bees knees in microphones, I might go as far as calling it the sum total of about half the erogenous zones of female pollinating insects on the western hemisphere, while the 2SK170 plainly sux a$$.

So, now that we reviewed theory, and being almost in 2568, can we use this knowledge to make better microphones?

Then again, why bother, people do not want actually better microphones, they want a recipy to build a perfect Neumann 87 Klown for 10 bux worth of Chinese garbage of AliExpress.

Thor

 

[thor.zmt]

Well, it doesn't help that 2SK170 has been discontinued for a while...

There are SMD FET's that use the same die. There are analogs from On-Semi and now as a suppa s3xy dual from TI/BB.

There credible second source parts from linear (LSK170) and another FET specialist.

Sure, they cost a dollar or so, not 5 cent like the AliExpress 2SK170. Ever wonder why?

Thor