Comparison of JFETs for mic applications

[Voyager10]

You're right. I was having doubts about the accuracy of using a 68M resistor, so I did an LTSpice noise simulation:



Green is 68M, blue is 1G, both in parallel with a 68pF capacitor.

The explanation is that noise voltage increases as the square root of resistance, but the attenuation due to the capacitor increases proportional to resistance. Total noise is the same, as per theory, but as you say it's shifted towards frequencies which will be filtered out.

It's worth re-measuring with R1 = 1G (and possibly 500M, as most versions of the circuit have 2 x 1G around this point).

 

[abbey road d enfer]

You're right. I was having doubts about the accuracy of using a 68M resistor, so I did an LTSpice noise simulation:

Don't think I've been dissing you. Your experiment is very valuable. Keep the good work.

 

[ccaudle]

(For reference, REW reports the source signal as 0.00073% THD when input to the audio interface directly).

And that is an iPhone into a Focusrite?! That is really decent performance for a setup that most people could put together from equipment they already have.

 

[Voyager10]

I was pleasantly surprised - this was straight from the headphone out to the XLR input. I guess it's driving a very easy load, which must help.

On the plot there is a noticeable noise floor, which vanishes when you mute the signal (and presumably the output stage is turned off), so I expect there's a bit of DSP magic involved.

 

[Voyager10]

Revised noise measurements​

So I swapped the 68M bias resistor R1 for a 1G one, and re-measured the noise, both with a 1nF and 68pF input capacitance. The A-weighted equivalent input noise, as reported by REW, is as follows:

Device	Cin = 1nF (dBV A)	Cin = 68pF (dBV A)
2N3819	-120.5	-118.2
2SK117-BL	-120.6	-118.1
2SK118-GR	-120.4	-118.0
2SK170-BL	-118.3	-116.2
2SK208-GR	-120.5	-117.8
2SK30A-GR	-120.2	-117.8
BF244B	-120.5	-118.1
J112	-121.1	-118.3

Plots​

and for 68pF:

 

[Voyager10]

To help make sense of it, here's a plot comparing a typical FET (2SK118) over the different input configurations:

Some thoughts​

• Yes, the 68M input resistor was misleading.
• However, there still isn't an appreciable difference between most of the FETs in this circuit, still.
  • The 2SK117 has more LF noise than the others I measured. However, with the 68pF input capacitor, and an A-weighted measurement, this isn't actually any worse that the others.
  • The 2SK170 does come out a couple of dB worse than the other FETs, which is still evident with 68pF input capacitor and A-weighting.
  • All measurements used a single example of each FET. This doesn't therefore justify ripping out 2SK170s everywhere. I might try others in the same batch, and others from different sources.
• Back-of-an-envelope calculations suggests that the noise from the 68pF / 1G input components will be less than the noise from R3 (2k2) at about 1.5KHz or so. I think this is borne out in the plots above (which flatten out at 1-2Khz). To get any better above 1.5KHz (where the A-weighting really counts) we need to do something about R3.
  • Suggestions welcome. The aim here is "simple mods to the ADK / MXL 2001 circuit to improve performance", rather than a total rebuild.
  • Potentially a charge-amplifier arrangement, where R3 is shorted and we control the gain with a feedback capacitor, might be worth trying.
• Those mains spikes at 50Hz / 150Hz are getting annoying...

 

[abbey road d enfer]

• Potentially a charge-amplifier arrangement, where R3 is shorted and we control the gain with a feedback capacitor, might be worth trying.

A quick sim shows that it's potentially possible to halve the noise density above 4-5kHz with such a configuration; however it may not be productive in practice since acoustic resistance noise becomes dominant.

• Those mains spikes at 50Hz / 150Hz are getting annoying...

Indeed, they alter the significance of the global noise figures.

 

[MicUlli]

I only want to point out again,
These measurements are far away from showing the noise capabilities of modern JFET circuits. Here some actual data from my own measurements.

2SK208: EIN A weighted -125,5 dBV @ 50pF capsule capacitance, noise density @ 20kHz 3,9nV/sqrt(Hz), the worst JFET
NSVJ3910: EIN A weighted -129,0 dBV @ 50pF capsule capacitance, noise density @ 20kHz 1,8nV/sqrt(Hz), best in class JFET
2SK209: EIN A weighted -128,5 dBV @ 50pF capsule capacitance, noise density @ 20kHz 2,0nV/sqrt(Hz), always a good JFET for all purposes
LSK170A: EIN A weighted -126,9 dBV @ 50pF capsule capacitance, noise density @ 20kHz 2,3nV/sqrt(Hz), not worth the 8 € i spent for

BR MicUlli

 

[abbey road d enfer]

I only want to point out again,
These measurements are far away from showing the noise capabilities of modern JFET circuits. Here some actual data from my own measurements.

2SK208: EIN A weighted -125,5 dBV @ 50pF capsule capacitance, noise density @ 20kHz 3,9nV/sqrt(Hz), the worst JFET
NSVJ3910: EIN A weighted -129,0 dBV @ 50pF capsule capacitance, noise density @ 20kHz 1,8nV/sqrt(Hz), best in class JFET
2SK209: EIN A weighted -128,5 dBV @ 50pF capsule capacitance, noise density @ 20kHz 2,0nV/sqrt(Hz), always a good JFET for all purposes

BR MicUlli

What do you mean by "noise density @ 20kHz"? With a capsule and 1G resistor, the voltage noise density is not constant with frequency. Is it the FET's noise voltage density?

 

[MicUlli]

It is mainly the JFET voltage noise density. Because of the frequency dependance I stated it at 20 kHz. The RC noise (capsule and pol resistor) at this frequency can be neglected. I do not use a pol resistor at all but a double diode (BAV199), therefore the thermal noise disappears..

 

[abbey road d enfer]

It is mainly the JFET voltage noise density. Because of the frequency dependance I stated it at 20 kHz. The RC noise (capsule and pol resistor) at this frequency can be neglected. I do not use a pol resistor at all but a double diode (BAV199), therefore the thermal noise disappears..

Could you actually evaluate the equivalent resistance of the double diode? I know it depends on the FET bias, but roughly, is it 1G, 10G, 100G...?

 

[MicUlli]

Here https://groupdiy.com/threads/replacing-gohm-resistors-with-diodes-the-theory-behind-it.83141/ you can find all details. 5pA gives about 5GOhm.
But currents of <1pA are not seldom, wonderful

 

[Voyager10]

Updated measurements​

I've made a few changes to the circuit, and made some more measurements. It now looks like this:


Here's what it looks like:


The main changes are:

• Now P48 phantom-powered, which gets rid of the 50Hz noise from the external PSU (and makes it more convenient to operate).
• VR1 is now a 10-turn, 10K preset, which makes it easier to set the bias more consistently.
• The output buffer is now closer to the ADK / MXL circuit, in case that makes a difference.
• Better mounting arrangements for SMT devices (on SOT-23 adapter boards)

Here are our test subjects:

Results​

Device	Origin	Idss (mA)	Rbias (K)	Gain (1nF)	Gain (68pF)	THD % (1nF)	THD % (68pF)	Ein dBV(A) (1nF)	Ein dBV(A) (68pF)
2N3819	Old	10.3​	5.4​	6.05​	5.00​	0.046​	0.017​	-120.4​	-118.2​
2SK117-BL	eBay	7.55​	0.83​	6.77​	4.86​	0.039​	0.059​	-120.9​	-118.0​
2SK118-GR	eBay	3.67​	2.8​	5.82​	4.52​	0.190​	0.037​	-120.2​	-118.1​
2SK170-BL ii	Old	9.1​	0.74​	7.01​	3.89​	0.037​	0.210​	-121.2​	-118.3​
2SK208-GR	Farnell	3.46​	2.52​	5.92​	4.48​	0.190​	0.015​	-120.2​	-118.0​
2SK209-GR	RS	5.98​	0.84​	7.06​	4.38​	0.080​	0.110​	-121.2​	-118.4​
2SK30A-GR	eBay	3.62​	2.94​	5.57​	4.69​	0.180​	0.077​	-119.9​	-118.0​
2SK880-GR	Farnell	5.64​	0.76​	7.09​	4.45​	0.074​	0.120​	-121.2​	-118.3​
2SK932-23	RS	14.9​	1.55​	7.14​	5.30​	0.890​	0.110​	-121.2​	-115.9​
BF244B	Old	9.43​	5.24​	5.94​	5.01​	0.098​	0.043​	-120.3​	-118.1​
J112	Farnell	35.1​	5.28​	6.67​	4.85​	0.093​	0.200​	-120.8​	-118.3​
J113	Farnell	23.8​	3.68​	6.84​	4.96​	0.018​	0.140​	-120.8​	-118.4​
MMBF J201	Farnell	0.24​	0​	5.65​	4.64​	0.550​	0.400​	-120.0​	-117.8​

​

Rbias is the value of VR1 adjusted to give 0.5mA drain current, which was used for all the measurements. The J201 at the end of the table had an Idss (i.e. drain current when Rbias=0) of 0.24mA, so it wasn't possible to achieve 0.5mA.

Gain gives the voltage gain with 1nF and 68pF input capacitors (measured using REW at 10mV RMS, 1KHz input signal).

THD was measured with a 100mV RMS, 1KHz input signal using REW.

Ein is the equivalent input noise (i.e. noise measured at the output accounting for the amplifier gain).

Footnotes​

• As discussed earlier, the ADK / MXL circuit is not a low-noise design. Although there is variance between FETs, it's largely irrelevant as we end up at -118dBV, A-weighted, for almost all cases. For a typical capsule (-40dBV at 94dB sensitivity) this corresponds to a noise level of (-118)-(-40-94) = 16dBA. This agrees with the published specs for many mics using this design. If you want a quieter mic, it needs a different circuit, not just a better FET.
• Given the above, 2SK209 and 2SK880 were the best for noise (I think the -880 is the same as the -209 in a smaller package?)
• 2SK170 - I used a different device to my earlier measurements, an old one (not ones in a batch bought on eBay). It was noticeably lower noise; the eBay ones are certainly sub-standard.
• 2SK932 - disappointing. Advertised as very low noise, and suitable for audio applications. Something very weird is going on at 68pF input capacitance: it's worse than all the others by some margin. I tried 2 different examples with identical results.
• J112 and J113 have good noise performance; high-ish THD. J113 was overall comparable to the (good) 2SK170, given a different bias resistor.
• J201 (at least in SMT) - disappointing. Very high THD and a steep rise in LF noise, maybe related to non-optimal bias arrangements.

 

[rogs]

With no supply regulation to the FET, surely you'll need to adjust the bias for every individual phantom power supply you connect it to?....

 

[MicUlli]

Quite weird. The main noise you see at frequencies about 20 kHz is produced by R3 in your circuit. So the results are not usable for qualified JFET selection. See attached a noise profile 1/3 RTA of the best JFET i have ever measured...

 

[Voyager10]

To be clear: this is not my own attempt at designing a low-noise amplifier/impedance converter using JFETs. It isn't trying to find the "best JFET", or the JFET with just the lowest noise. It's a survey of how different JFETs perform in this particular circuit.

Why? Because this circuit is used in literally dozens of common LDCs, and people ask whether a "better" JFET would improve the mic overall, or what a good replacement for the original 2SK170 would be.

I'm hoping to be able repeat the exercise with other popular mic circuits (e.g. Schoeps, KM84) to help answer the same questions.

 

[MicUlli]

To be clear: this is not my own attempt at designing a low-noise amplifier/impedance converter using JFETs. It isn't trying to find the "best JFET", or the JFET with just the lowest noise. It's a survey of how different JFETs perform in this particular circuit.

Understood. This circuit may be used in a lot of designs, but it has a very low performance in all aspects. "Better" circuits are available and should be tested.
For this special circuit any JFET can be used..

I'm hoping to be able repeat the exercise with other popular mic circuits (e.g. Schoeps, KM84) to help answer the same questions.

I would recommend to leave out the KM84 circuit because of similar bad behaviour regarding noise and distortion. Schoeps is far more better in low distortion, noise is a little bit lower.
My recommendations for an interesting comparison are the OPA Alice single ended output design and of course my SDC circuit:
https://groupdiy.com/threads/a-very-versatile-input-stage-for-sdc-mics.80172/post-1084357
BR MicUlli

 

[Voyager10]

New circuit, new data​

I've now built & tested @MicUlli's circuit with a range of FETs. This is the circuit used:

(This is the original version from Post #1, not the later one with two diodes).

Notes:

• R4 is selected to keep the Vds in the 1-2V range - i.e. comparable to that measured for the original's 2SK117 - for different FETs. The measured Vds is shown in the table.
• THD measurements were made with 1KHz 100mV RMS input, as earlier in this thread. Note the signal source is good but not perfect, so take these figures as an upper bound.
• Noise voltages (Ein, as dBV) are A-weighted, as reported by REW. As before this is the equivalent input noise, although because the circuit is so close to unity-gain, these are basically the same as the output noise.
• The same samples of each FET were used here as for previous measurements. I've also added a Interfet J305 (>$3 from Mouser).
• Measurements were made with a large (1nF), and "typical LDC" (68pF) input capacitance.

Results​

Device	R4	Vds	Gain 1nF	THD 1nF	Gain 68pF	THD 68pF	Ein 1nF	Ein 68pF
2N3819	68K	1.35​	-0.19​	0.0023​	-0.23​	0.0024​	-125.5​	-122.1​
2SK117-BL	33K	1.6​	-0.2​	0.0023​	-0.25​	0.0025​	-126.0​	-122.1​
2SK118-GR	47K	1.55​	-0.22​	0.0024​	-0.28​	0.0025​	-125.4​	-121.9​
2SK170-BL	33K	1.6​	-0.2​	0.0023​	-0.26​	0.0027​	-126.9​	-122.7​
2SK208-GR	47K	1.7​	-0.22​	0.0023​	-0.27​	0.0025​	-125.5​	-121.5​
2SK209-GR	33K	1.6​	-0.19​	0.0023​	-0.25​	0.0026​	-126.8​	-122.6​
2SK30A-GR	47K	1.55​	-0.23​	0.0024​	-0.29​	0.0025​	-124.8​	-121.5​
2SK880-GR	33K	1.6​	-0.19​	0.0023​	-0.25​	0.0026​	-126.8​	-122.2​
2SK932-23	33K	1.45​	-0.19​	0.0024​	-0.24​	0.0025​	-126.9​	-122.5​
BF244B	68K	1.35​	-0.17​	0.0023​	-0.22​	0.0024​	-125.4​	-122.2​
J112	68K	1.15​	-0.14​	0.0023​	-0.21​	0.0024​	-126.5​	-122.5​
J113	68K	1.9​	-0.14​	0.0023​	-0.2​	0.0024​	-126.6​	-122.5​
J201	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
J305	47K	1.25​	-0.22​	0.0024​	-0.28​	0.0026​	-125.4​	-121.8​

Observations​

• Gain and THD are very consistent across all the devices, likely due to the bootstrapping of both gate and drain terminals.
• Noise is much better than the ADK/MXL circuit, we can finally see some variation between FETs.
• I've attached plots of noise voltage, as measured. The nV/√Hz values given at the bottom are for 10KHz.
• The capsule voltage (V_CAP) varies according to the R4 value, 33K gives 39.0V, 47K gives 38.2V and 68K gives 37.0V.
• Vds took a long time to stabilise, the values given are approximate.
• My J201 FET has an Idss of 0.24mA, which is less that the 1mA operating current required by the circuit. It couldn't be used here (without changing a bunch of other resistors).
• The J305 seems nothing special here; the humble OnSemi J112/J113 are a tenth the price and measure better.

 

[Voyager10]

Just out of curiosity, I tried increasing the value of bootstrap capacitor C1 to 4.7uF.

With 1nF source capacitance the noise below about 100Hz is measurably improved. For the 2SK209, the green vs red traces below show the difference.



A-weighted, that's now -127.1 dBV (previously -126.8 dBV).

For a microphone application, with 68pF source capacitance, there's no significant difference (the top two traces), so don't get too excited.

More circuits coming, when the time allows...

 

[Voyager10]

OPA1641 Initial Results​

Here's the circuit - just a very simple voltage follower running off P48 power:


R6 can be chosen to set the supply voltage, in this case 24V. Supply current is approximately 1.7mA, and there's 41.6V available at XLR pins 2 and 3 for capsule polarisation.

Measurements (all at 1KHz) are as follows:

Value	Cin=1nF	Cin=68pF
Voltage gain (100mV in)	-0.3dB (x0.96)	-1.2dB (x0.87)
Input noise, A-weighted	-122.9 dBV	-119.4 dBV
THD, 100mV RMS in	0.00052%	0.00057%
THD, 500mV RMS in	0.00047%	0.00040%

Caveats​

• The construction might be responsible for some excess input capacitance (I used the IC soldered to a carrier board, plugged into an IC socket, soldered to a perfboard), which lowers the voltage gain (and raises the input noise) for Cin=68pF. I may try an 'air-wired' build.
• A practical microphone circuit would probably use a second high-value resistor at the input, which would increase the low-frequency noise.
• THD numbers really just mean "lower than I can measure" here. See the spectrum below - most of what you're seeing is the signal source (an iPhone!) rather than the amp itself.

The noise curves look like this - again, it's possible that lowering the input capacitance might bring the green one down by a fraction.