Comparison of JFETs for mic applications

[abbey road d enfer]

Has anyone tried to measure the input impedance (or input admittance) for some of these fet's? I assume the impedance would be some ultra high resistance in parallel with a few pf of capacitance.

The real part of input impedance is constituted of leakage, which is more or less constant, and reverse conductance of the gate junction, which varies with bias and Vds. Low Vds is desirable, as long as it doesn't compromise headroom.

Hopefully the input capacitance (reactance) of the fet is negligible compared to the 20 to 40 pf of the capsule in parallel.

Let('s take the ewample of a Neuman KM (53,54,..). The input is not bootstrapped, because of the source capacitor, so the input capacitance is directly loading teh capsule. With a typical 20pF capsule, a low Cgs capacitance is required; the typical 2N3819 is spec'd at 4pF, which accounts for about 2dB in sensitivity reduction.
That's why many designers use bootstrapping, either by not decoupling the source resistor, or using dedicated NFB.

 

[abbey road d enfer]

With bootstrapped JFETs Cin can be neglected, but in charge amps it plays a significant role..

Can you elaborate on that? I would have thought that, due to the low input impedance of a charge amp, Cgs would be less significant.

 

[OneRoomStudio]

I have tested the LSK170A regarding noise. Dont waste your money

LSK170X was supposedly designed to be lower noise than LSK170A/B/C/D.

Also, LSK189 is very interesting:
Ciss = 4pF
Crss = 3pF
En @ 1kHz = 1.8 nV/Hz

 

[ricardo]

LSK170X was supposedly designed to be lower noise than LSK170A/B/C/D.

It is in certain applications but NOT in HiZ stuff like condensor mikes.

Yes. Bootstrapping, and other fancier and less fancy circuit dodges help alleviate large Ciss & Crss. But don't eliminate their EVIL influence.

In every case, replacing 2sk170 with a more appropriate FET will give you less noise.

There are several noise sources in a condensor mike with different audible effects but I won't go into this huge topic. Just make sure you LISTEN to the noise to hear if it's OK for you.

Can you elaborate on that? I would have thought that, due to the low input impedance of a charge amp, Cgs would be less significant.

Ciss/Cgs in a charge amp still acts like a 'voltage' divider on the signal before it enters the charge amp.

 

[thor.zmt]

Has anyone tried to measure the input impedance (or input admittance) for some of these fet's?

Well, it's known.

A J-Fet is essentially a diode plus.

Commonly source and drain are interchangeable, so we have kind of two diode junctions.

As long as the J-Fet is operated with the Gate below ~ +0.3V with respect to the source, the gated is basically a diode.

It doesn't conduct. It is pretty much Gigaohm.

Capacitances are usually symmetric too.

The higher the J-Fet transconductance (thus gain) the higher the capacitance.

If the J-Fet operated with significant gain the intrinsic capacitance from gate to drain is amplified by what is called the "Miller Effect".

So if we have 5pF (a common capacitance) and a gain of 15dB (5X) the 5pF are boosted to 25pf and the gate to source capacitance remains 5pF, so 30pF in total.

With the J-Fet operated as follower, in turn the gate to drain capacitance is 5pF, the gate to source capacitance is bootstrapped by the follower action, so it pretty much disappears.

So as follower the same J-Fet that has 30pF input capacitance with a 5X gain, has 5pF input capacitance, which could be lowered dramatically using a cascode.

Also, the capacitances are not linear, nor is the transconductance.

So operating J-Fets as gain stage directly linked to a microphone capsule will likely be by far worse for objective performance than using it as follower.

Hopefully the input capacitance (reactance) of the fet is negligible compared to the 20 to 40 pf of the capsule in parallel.

Hope springs eternal. It also the main reason for sad disappointment.

Thor

 

[thor.zmt]

Can you elaborate on that? I would have thought that, due to the low input impedance of a charge amp, Cgs would be less significant.

Depends on open loop gain.

If we have (say) 60dB loop gain at 10kHz, then the voltage error on the gate will be 1/1000 of the input voltage , so even a high capacitance will be reduced to a small fraction, but if loop gain is low, the action of the feedback will be modest, large error voltages mean the fraction of the effective input capacitance that causes errors is larger.

And of course the input capacitance affects loop gain.

So all else being equal, THD lower capacitance and thus lower transconductance and higher noise J& FET tends to perform in a capacitor microphone.

Thor

 

[abbey road d enfer]

Ciss/Cgs in a charge amp still acts like a 'voltage' divider on the signal before it enters the charge amp.

I've done a simulation of a typical Neumann circuit, in 4 configurations;

1. Common-source
2. Charge amp
3. Common source with additional 80pF gate capacitor
4. Charge amp with additional 80pF gate capacitor

80pF seems a rather high value, but it's what I needed to produce about 6dB attenuation.
It would suggest that the input impedance of the 1st circuit is equivalent to a 80pF. Actually a complementary simulation confirmed that.
It doesn't add up with the 2N3719 characteristics, Cgs=2.4pF and Cgd=1.6pF, which, with a gain of 26dB would result in 34.4pF. I don't know why...
Here are the noise results of the simulation:

My understanding is that the difference between 1 and 3 shows that the noise level decreases less than amplitude, which clearly indicates a deterioration of S/N, however, comparing circuits 2 and 4 shows that noise level decreases (1.06dB) by nearly as much as amplitude (1.3dB), which tends to confirm that charge amps are less sensitive to Cgs variations.

Circuits are attached as pdf

 

[ricardo]

My understanding is that the difference between 1 and 3 shows that the noise level decreases less than amplitude, which clearly indicates a deterioration of S/N, however, comparing circuits 2 and 4 shows that noise level decreases (1.06dB) by nearly as much as amplitude (1.3dB), which tends to confirm that charge amps are less sensitive to Cgs variations.

Thanks for this Abbey.

Another way to think of the effect of Cgs in a Charge Amp circuit might be that though it acts as a voltage divider, it also increased the equivalent series 'source capacitor'

Alas, FETs with high Cgs always have high Cgd too so I still wouldn't use 2sk170 in a LN mike for this and other reasons.

I can think of a few circuits where it would be better than other FETs but these, IMHO, aren't good mike circuits.

 

[abbey road d enfer]

Thanks for this Abbey.

Another way to think of the effect of Cgs in a Charge Amp circuit might be that though it acts as a voltage divider, it also increased the equivalent series 'source capacitor'

Alas, FETs with high Cgs always have high Cgd too so I still wouldn't use 2sk170 in a LN mike for this and other reasons.

I can think of a few circuits where it would be better than other FETs but these, IMHO, aren't good mike circuits.

I agree 100%. I would never use a 2SK170 (or LS170) in really high-Z circuits. I use them in mic in-line mic booster where the source is low-Z.
I experimented with them and found out that when operated with high Vgd (30-40V), their input current starts to be a nuisance, even with medium source impedance, like 5-10kohms.

 

[Voyager10]

The effect of FET capacitance can be gauged from the drop in gain between 1nF and 68pF in the ADK circuit. (With 1nF the FET capacitance is too small to make a difference, so the gain is set just by the transconductance and the R3/R4 values. With 68pF the gain is reduced as Ciss/Crss are large enough to present a significant load). So low-capacitance FETs like the 2N3819, BF256B, etc. have a 15%-20% drop in gain, whereas with the 2SK170 and 2SK369 it's 50% or more).

The THD works in a similar way. The '1nF' figure gives basic nonlinearity without any capacitance effects being significant - basically how close to a straight line the I/V curve is. The '68pF' figure will be worse if Crss (or Ciss) changes as the drain-gate (or source-gate) voltage changes; this is sometimes plotted in the datasheets.

 

[Mr. Moscode]

Are those gate to drain caps really 1Farad in 2 of the circuits?

 

[ccaudle]

I'm trying to figure why it matters . . .
I thought this thread was about FETs for DIY condenser mics.

My logic was that someone asked why there were no tests of 2SK660. Thor had even noted several posts previously that 2SK660 was the only JFET he uses in microphones.
However it seems that 2SK660 is discontinued. Renesas still has the NEC datasheet on the web site, but it now has a stamp that is is in "Phase-out/Discontinued" status, and it is not available at Digikey or Mouser.
You mentioned 2SK596, but although it is in stock at Digikey, it has a note that "This product is no longer manufactured" and is only available shipped from Rochester Electronics (which I believe is a vendor who buys up obsoleted devices and resells them to service uses where it is impractical to redesign using a current part).

Usually when there is a market of ca. 1 billion units per year, someone will make a new near equivalent when it is no longer economical to continue producing an old design. All new devices that I have seen for many years come in only a surface mount package, so it stands to reason that even though 2SK660 is discontinued, and 2SK596 is discontinued, there should be a very similar device in SMT package available which are tailor made for use with small condenser microphone capsules.
And SOT packages are certainly usable for DIY, so a low noise transistor with on-die diode biasing would still be relevant.

Renesas still has a datasheet for 2SK1109 available, but you can't navigate to it through the web site that I have found, you have to know the part number to search, and not available at Mouser (although is available from Rochester Electronics), so probably not a market Renesas care to still be in.

Unisonic make TF2123 which is a SOT JFET with bias diode, but I'm not sure how easy it is to get outside of Asia. While poking around on that site I found that they also list K596 which seems to be a version of 2SK596 still in TO-92 package as well as several other transistors specifically called out as targeted for microphone applications (a total of 8 I found which have the bias diode and are called out as intended for microphones).

Unisonic have sales offices in China and India, so may be very difficult to get in Europe and North America.

It might be the case that all of the more recent devices are only sold in China (at least I am not aware of any electret capsule manufacturing occurring elsewhere).

Since as Thor pointed out these types of devices are made specifically to be low enough noise to use with small diameter electret capsules, they would be interesting to compare against the other tested devices.
I would be interested to know of any other manufacturers, it seems NEC (Renesas) and OnSemi are out of that business. I would not be surprised if all of the traditional Japanese, American, and European device vendors have completely ceded that market to Chinese vendors now.
The OnSemi TF252TH is obsolete, but Rochester Electronics apparently has a couple hundred thousand in stock, so might be viable for DIY for quite a few more years, but I would really prefer something which is currently manufactured, and preferrably available outside of Asia.

 

[soliloqueen]

I have tested the LSK170A regarding noise. Dont waste your money

in terms of my experience with the LS models, it's less the noise and more the superior breakdown performance. Especially in the 87 circuit. Neumann, for inscrutable reasons, specified the 2N3819, a FET that can only work in the circuit in a narrow sliver of its manufacturing tolerances without cutting out at high signal, and then spent ages selecting the 1 in every 10 units that could do it instead of just...selecting a better jfet for the application. The LS replacement for the 2N3819, at least, is produced in such a way that every unit can be used in a u87 amp without cutting out... So it has that going for it? But like, why not just use one of the better-suited jfets for the design in the first place? that way you can save time AND money. i read somewhere that neumann tried replacements, but it didn't sound right. I would need to test this kind of thing myself, but it makes me suspicious. maybe thor knows something about this

 

[abbey road d enfer]

Are those gate to drain caps really 1Farad in 2 of the circuits?

1f=1femtoFarad or 10E–15 Farad

 

[abbey road d enfer]

The effect of FET capacitance can be gauged from the drop in gain between 1nF and 68pF in the ADK circuit.

There is no doubt about it in the case of a non-global NFB circuit.
Just like there is no doubt that level, and distortion are smaller for the charge amp.

(With 1nF the FET capacitance is too small to make a difference, so the gain is set just by the transconductance and the R3/R4 values. With 68pF the gain is reduced as Ciss/Crss are large enough to present a significant load). So low-capacitance FETs like the 2N3819, BF256B, etc. have a 15%-20% drop in gain, whereas with the 2SK170 and 2SK369 it's 50% or more).

The FET in a charge amp sees a much higher source capacitance (Cgs+G.Cgd), which makes the variations of its own gate capacitance less perceptible.

The THD works in a similar way. The '1nF' figure gives basic nonlinearity without any capacitance effects being significant - basically how close to a straight line the I/V curve is. The '68pF' figure will be worse if Crss (or Ciss) changes as the drain-gate (or source-gate) voltage changes; this is sometimes plotted in the datasheets.

Exactly. The G.Cgd term is less variable than Cgs, making the circuit more linear.

 

[thor.zmt]

I've done a simulation of a typical Neumann circuit, in 4 configurations;

1. Common-source
2. Charge amp

Just a remark, "Charge Amp with EXTREMELY LOW loop gain".

Try bootstrapping the load, a cascode on the FET.

THEN we will have something that deserves the moniker "Charge Amp".

And the conclusion that extra capacititance in the charge amp deteriorates gain and SNR similarly is correct, the degree of deterioration is directly inverse to loop gain.

A "charge amp" with minimal loop gain, will behave almost like the common source circuit, one with infinite loop gain will tolerate infinite parasitic capacitance and show no degradation in Gain or SNR.

Thor

 

[thor.zmt]

Usually when there is a market of ca. 1 billion units per year,

They went MEMS many, many years ago. All the small electret mic's with build in FET for mobile phones and earbuds etc have long been discontinued. MEMS is a single chip, including everything and with excellent consistency.

Many MEMS microphones are even 3.072MHz PDM Output, straight into a CPU pin.

Unisonic have sales offices in China and India, so may be very difficult to get in Europe and North America.

I remember seeing their stuff at mouser, no?

Since as Thor pointed out these types of devices are made specifically to be low enough noise to use with small diameter electret capsules, they would be interesting to compare against the other tested devices.

There was an article in EE Times IIRC showing actual test results in microphone circuits.

The 2SK660 I got were a surplus reel when they were dumped by Chinese distributors. I think less than 1 cent US per fet, 1,000 pcs or so.

Thor

 

[ricardo]

Probably the easiest source of a FET with built in diode like 2sk596 & 2sk660 for DIY is to buy some PUI 5024 capsules from Mouser and cannibalize the FET. Even lower noise than the NECs and high gain.

I believe it's a MOSFET so take precautions not to kill it with static. The diode will protect it from -ve Vg but there is nothing to drain +ve Vg charge.

2n3819 is used in many legacy mike cos in Jurassic times, it was the ONLY readily available FET. These designs predate my time at Calrec.

BTW, PUI 5024 is an excellent Omni capsule

 

[MicUlli]

Probably worth to mention the concept of "equivalent input noise". Whatever circuit you use, EIN is the one and only figure that counts for noise performance.
If we want to isolate the JFET as primary noise source we can use a model of a source follower (see pic sfollower).
Input referred noise density is

nd = (1 + Cin/Ccapsule) * SQRT (8kT/(3 gm)

and describes EIN perfectly. gm is the transconductance at operating conditions.

There is no way to reduce nd whatever circuit you design (at equal JFET operating conditions)...

 

[thor.zmt]

If we want to isolate the JFET as primary noise source

But it ain't, that's point. For a complete microphone J-Fet as noise source in Mic's is tertiary or lower order.

As remarked, the Johnson-Nyquist noise of the resistors in the input circuit dominates at low frequencies, it is lowpassed by the capsule capacitance.

At higher frequencies the Brownian motion of air molecules impacting the diaphragm is the limit. It is usually higher than a relatively noisy J-Fet.

Putting a lower noise J-Fet into the circuit usually makes things worse, as low voltage noise implies large geometry and thus large capacitances which in most common microphone circuits causes things to get worse not better.

Thor