JFET shootout
- Thread starter MicUlli
- Start date
Help Support GroupDIY Audio Forum:
Ricardus
WILL SOLDER FOR FOOD
The classic 2N3819 was not in there. Would be curious to see how it stacks up.
rock soderstrom
Tour de France
Thanks again Ulli for your work and effort.Hi All,
many people are interested in this question: What is the best JFET for diy mic builders?
Find this useful info in the attachment and have fun
Could it be because it's discontinued by the brand-name manufacturers, and can only be sourced from so-and-so brands nowadays?
Ricardus
WILL SOLDER FOR FOOD
Maybe but that's not really true. Mouser has them in stock right now.
https://www.mouser.com/ProductDetai...3819-TRE-PBFREE?qs=W/MpXkg%2BdQ4rsn8OhSyZOA==
k brown
Well-known member
Seems odd that FETs specifically designed for mics were not included, like 2SK660 and 2SK596.
They require different test conditions?
They require different test conditions?
Last edited:
Like i said - so-and-so brands...
Ricardus
WILL SOLDER FOR FOOD
If Mouser sells it, it's fine.
It's not Toshiba/On semiconductors. Not to say they are fake, but many have tested these and they don't seem suitable for where the original part was used. They all seem to fall on the wrong end of specs.
Last edited:
These are good tests but Ein isn't the only criteria for noise in a Mike circuit. The reason many RF FETs are used in commercial mikes is that the input capacitance (and more importantly Cgd) reduces the 'gain'. This is one reason why K118 is often better (quieter) than K117 in a real mike.
In a FET mike, LF noise is usually completely dominated by the 1G/capsule_capacitance noise.
The type of circuit also influences matters. The common Schoeps circuit needs a FET with high Idss for low noise and some FETs are inadequate.
The FETs used in the Transounds & Primos like K596 are a different beast entirely.
In a FET mike, LF noise is usually completely dominated by the 1G/capsule_capacitance noise.
The type of circuit also influences matters. The common Schoeps circuit needs a FET with high Idss for low noise and some FETs are inadequate.
The FETs used in the Transounds & Primos like K596 are a different beast entirely.
A lot of comments, thanks for the hints Let me answer to some of these:
1. 2N3819 was not tested because of the poor forward transconductance of only 1,5 mA/V @ 1 mA Id which is comparable to 2SK118
2. 2SK660 and 2SK596S claim to be ideally suited for mics, that is only true for electret mics with 5..15 pF capacitance The data sheets note an output noise level of -110 dB(A) @ 15 pF which is 10 dB more than the winner..
3. @ricardo: Sorry, but EIN is THE one and only criterium for noise spec. because it includes the gain of the circuit automatically. No question, the gain itself is influenced by the JFET Cin, but also the output noise. But this may be better visible without the noisy 1G resistors.
4. @kingkorg: I did not test smd JFETs because they are not easy to handle for DIY people, but you are completely right. Good test candidates are 2SK3557, 2SK209, 2SK932 and a lot more.
The tests show that channel noise is the most important factor. Therefore devices with high forward transconductance gm AND low excess noise capacitance Cen are preferred. Unfortunately Cen has to be checked by measurement. Cen is correlated with the input capacitance Cin of the JFET but cannot be expressed analytically.
Let me answer to some of these:1. 2N3819 was not tested because of the poor forward transconductance of only 1,5 mA/V @ 1 mA Id which is comparable to 2SK118
2. 2SK660 and 2SK596S claim to be ideally suited for mics, that is only true for electret mics with 5..15 pF capacitance The data sheets note an output noise level of -110 dB(A) @ 15 pF which is 10 dB more than the winner..
3. @ricardo: Sorry, but EIN is THE one and only criterium for noise spec. because it includes the gain of the circuit automatically. No question, the gain itself is influenced by the JFET Cin, but also the output noise. But this may be better visible without the noisy 1G resistors.
4. @kingkorg: I did not test smd JFETs because they are not easy to handle for DIY people, but you are completely right. Good test candidates are 2SK3557, 2SK209, 2SK932 and a lot more.
The tests show that channel noise is the most important factor. Therefore devices with high forward transconductance gm AND low excess noise capacitance Cen are preferred. Unfortunately Cen has to be checked by measurement. Cen is correlated with the input capacitance Cin of the JFET but cannot be expressed analytically.
Last edited:
I wouldn't lightly dismiss Ricardo's comment. He has more experience in his little finger than any one of us in our entire body. He's just a little grumpy from times to times.
Your test jig, with it's very low input capacitance, due to the elimination of Miller capacitance, clearly favours FET's with a very low intrinsic noise.
Many mic head amps don't have this kind of circuitry and favour low capacitance (particularly reverse capacitance) because it doesn't load too much the capsule.
One may argue that, in the 21st century there shouldn't be any condenser mic with a circuit that doesn't eliminate Miller capacitance, however, many are still in production, and not necessarily from cheap manufacturers.
Somewhat unrelated: how do you explain "NTE457 is highly claimed as „best performer“, as it's apparently the worst in your study?
k brown
Well-known member
In actual use, my externally polarized mics using K660 have not displayed any problematic self noise.
Of course, I've never used them to record a softly-played mandolin in an anechoic chamber . . .
Of course, I've never used them to record a softly-played mandolin in an anechoic chamber . . .
I try to clarify my point:I wouldn't lightly dismiss Ricardo's comment. He has more experience in his little finger than any one of us in our entire body. He's just a little grumpy from times to times.
Your test jig, with it's very low input capacitance, due to the elimination of Miller capacitance, clearly favours FET's with a very low intrinsic noise.
Many mic head amps don't have this kind of circuitry and favour low capacitance (particularly reverse capacitance) because it doesn't load too much the capsule.
One may argue that, in the 21st century there shouldn't be any condenser mic with a circuit that doesn't eliminate Miller capacitance, however, many are still in production, and not necessarily from cheap manufacturers.
Somewhat unrelated: how do you explain "NTE457 is highly claimed as „best performer“, as it's apparently the worst in your study?
Suggest a capsule with voltage Vin and capacitance Cc and a JFET with input capacitance Cin. At the JFET gate you get Vgate = Vin*Cc/(Cc+Cin) because of trivial voltage divider rule. Now inject a noise current In at the JFET gate node. It will produce a noise voltage Vn = In/(jw(Cc+Cin)). Calculate
Vn/Vin = In/(jwCc). Cin is eliminated and therefore does not influence the signal to noise ratio..
NTE457 is highly claimed as best performer in the sensor industry. Used in several capacitive sensor devices, but not often seen in condenser mics..
It depends..
One example is Primo EM23 capsule witch was measured having an SNR of 84 dB(A). Sensivity is roughly -40 dBV @ 94 dB SPL. Self noise is 10 dB(A). The capsule related noise floor is therefore -124 dBV(A). If we use a preamp with -122 dBV(A) we sacrifice the good performance, the overall self noise would be 14 dB(A). A preamp with -130 dBV(A) would nearly preserve the excellent capsule SNR and would lead to 11 dB(A) overall self noise.
So what is wrong with getting the best out of a system? Or the other way round, what amount of self noise is acceptable?
k brown
Well-known member
My point was that mic self noise is something that can get overly obsessed about - in practical use it is, in the vast majority of situations, inaudible in the final recording.
I really don't mean to beat a dead horse. Ambient noise such as breathing of an artist will be orders of magnitude above the mic noise. Unless you are building some kind of sensor, or purely exercise electronics to get closer to the best theoretical performance this quickly stops making sense in real world application.
If you look at the dynamic range you are mentioning, it goes beyond any ad converter not including preamp. Somewhere in the chain you will clip stuf, either mic, or pre, or introduce noise.
Also for regular DIYers it is really difficult to pull off a super quiet build, because it would need to be flawlessly executed. Wire length, flux residue, component layout, parasitic everything, capsule tolerance issues, capsule dirt collectedover time, humidity... soldering smd would be a piece of cake compared to this.
