Choosing the right JFET for source follower.

[Fancy Lime]

Sorry for the silly question, I seem to be too tired for my brain to function properly. When picking JFETs for buffers, do I want to pick high Idss or low? High Vgsoff or low? Sure, it does not matter as much as for common source stages but still. And yes, I can set the bias point and source resistor after picking the device, so that does not limit my choices. My gut tells me I'd want low Vgsoff for closer-to-1 gain but low Idss for lower noise. My frontal lobe tells me me to get some sleep. But my lizard brain tells me sleep ain't happening until I know which JFET...

Thanks,
Andy

[Rob Strand]

Other requirements to noise and gain are: distortion, output impedance, expected signal voltage, expected current drive on the output.    Heaps to consider.

The most intuitive result I could give is if you imagine the output impedance of the buffer being low, the gain will be high.  The gain is slightly less than unity and the "lost" voltage appears across the gate-source, that's the voltage swing applied to the actual JFET.   If the gate-source voltage is low then you would also expect the distortion to be low.

The output impedance depends on the JFET and the bias point.   The JFET part is rds0 = 1/yfs0.      The bias point part gives the output impedance as ro = rds0 * sqrt(IDSS/ID).       Since rds0 = VP / (2*IDSS),   you would lower the rds0 part by using a low VP and high IDSS JFET.    It's still not clear where you stand overall with ro as sqrt(IDSS/ID) will rise when IDSS is high.   If we write ro out in full we get ro = (1/2) VP/ sqrt(ID IDSS) so large IDSS still helps.

Because we want roughly Vsupply/2 at the source, and if we fix RS then  ID = (Vsupply/2) /  RS is pretty much fixed anyway.

So that points to having high IDSS and low VP.    The low VP is interesting since usually large VP is associated with lower distortion - need to think about that one.

There are some conditions imposed on the design:

- For full output swing the JFET needs to supply a current at least Vsupply / RS.   Since the JFET current limits at IDSS we need IDSS > Vsupply/RS.   If we have an AC load as well then the JFET needs to supply more current than that.   That means IDSS needs to be largish, which doesn't go against the result above for the buffer.

- Another limitation is when the JFET drain source voltage approaches VP the JFET starts operating as a controlled resistance and the positive peaks will distort.     The above requirement for small VP helps maximize swing with a low supply voltage.    If you were to use a high VP JFET you might consider biasing the output at lower than Vsupply/2 to get a more symmetric swing.

- The AC load needs to be considered in general.  But that seems to be covered by the high IDSS.  It does mean you might need a lower RS which will change the gain and distortion.

[Vivek]

A) Is there a possibility that this beats an Opamp in terms of  input impedance, output impedance and noise ??

B) is there any real advantage of using a FET and Opamp together, in similar fashion as TC electronic Integrated preamp. One of the Elmers had earlier posted FET/OPAMP config.

[Fancy Lime]

A) Is there a possibility that this beats an Opamp in terms of  input impedance, output impedance and noise ??

B) is there any real advantage of using a FET and Opamp together, in similar fashion as TC electronic Integrated preamp. One of the Elmers had earlier posted FET/OPAMP config.

I have been asking those questions in the past.

A) noise *can* be better with a single JFET if you can get a good low noise type like 2sk170 or 2sk117. Beware of fakes, though. The real ones are increasingly difficult to get. I happen to have a few, hence my question. Impedances are always going to be better with a decent JFET-input opamps. Even a humble TL072 will outperform the best discrete JFETs in those respects by orders of magnitude.

B) there used to be when opamps were in their infancy but with modern opamps I see no reason to go to the extra trouble with strange biasing and whatnot. Modern audio opamps with JFET inputs are amazingly good. However, for guitar effects or amps, even an old TL072 is good enough so that the opamps choice is not the limiting factor for overall performance. You can use a better one but you will almost certainly not be able to tell a difference in a blind A/B test. I tried. Extensively. The only real difference was bandwidth way above audio, therefore irrelevant. There are also technically slight differences in low frequency noise but those are way too small for me to hear. So do as I did after extensive lecturing by PRR and R.G. and simply enjoy the fact that even modest past-1970s opamps are actually way better than we would need them to be to make any real difference. At least in guitar audio and as long as the circuit is designed according to the opamps specs.

@ Rob
Thanks! Well, we always want the lowest possible Vgsoff at a given Idss or the highest possible Idss at a given Vgsoff, don't we? That's what makes the 2sk170 and it's peers so interesting. But within any given type of JFET, Vgsoff and Idss have a fixed correlation. Correct me if I'm wrong. So my question was more, if I have to choose between lower Vgsoff or higher Idss, which do I prefer for a buffer? Your musings are a great guide to buffer design in terms of the relevant parameters and resistors, btw. Thanks! I did not even think about distortion until now.

Andy

[Rob Strand]

Thanks! Well, we always want the lowest possible Vgsoff at a given Idss or the highest possible Idss at a given Vgsoff, don't we?

Well at least for the buffer.   For the common-source type amp I'd have to think about it.

But within any given type of JFET, Vgsoff and Idss have a fixed correlation. Correct me if I'm wrong. So my question was more, if I have to choose between lower Vgsoff or higher Idss, which do I prefer for a buffer?

Based on the previous argument we can use the output resistance ro as a guide.

Since,

       ro = (1/2) VP/ sqrt(ID IDSS)

If we say for a given model of FET the parameters scale with 'a' as

        VP = kV * a^2  and  IDSS = kI * a^3

where, kV, kI are constants.

then

    ro = (1/2) kV a^2 / sqrt(ID kI) / sqrt(a^3)
         = constant * (1/sqrt(ID))  *  sqrt(a)

so for low ro we would choose a lower 'a' and that means the lower VP and IDSS.

This is a first order approximation.   IIRC the scaling of VP and IDSS doesn't quite follow powers of 'a' I've indicated for VP and IDSS, especially when VP & IDSS are at one of the extremes (can't remember which end  ).   The square root varies slowly so the gains for smaller VP's are small over other VP's.   Also, we don't want to choose an IDSS so low that we can't get full output into the load + RS - as per design constraints at the end of the previous post.

FWIW, it would be worth verifying this conclusion.
[EDIT:  In spice, I scaled three JFETs according to 'a' and the lowest 'a' had the highest gain and lowest distortion.
             So, assuming the scaling powers are correct, the conclusion holds that using the smallest VP (and IDSS) is best. ]

[PRR]

There's 12 different ways to bias a JFET. You think it matters?

[jonny.reckless]

For a self biased source follower running on a 9V supply I always use a J112. Using a 4k7 resistor, it will typically self bias around 4V, half of the supply rail, 1mA quiescent current, give or take. Plenty of headroom.

It's pretty quiet, and adds nice second order harmonic to add that elusive "warmth" when you push it a bit hard. My go to input stage for any solid state guitar amp is a 1M input resistor, 10k gate stopper, and a J112 with a 4k7 source resistor. It also works surprisingly well as a "warmer" for your DAW, at 0dBV you get around 0.5% to 1% 2nd harmonic. I have a couple of these I use with aux buses when doing masters if I want an old school analog warmth. A simple self biased JFET source follower sounds better than any plugin I have found so far for this purpose.

[iainpunk]

I have a couple of these I use with aux buses when doing masters if I want an old school analog warmth. A simple self biased JFET source follower sounds better than any plugin I have found so far for this purpose.

nothing beats real analog of course, but i like to use fruity waveshaper if im working with samples or other computer only instruments in a song/beat (mainly hiphop, please keep quiet about that guilt pleasure) i like to set the input gain to the smallest setting and then turn up slightly until the warmth comes in enough. sometimes i even use it on the master bus to add slight compression and warmth to really smooth out the mix


cheers, Iain

[bushidov]

Jonny, you would add a decoupling capacitor between the input and the intersection of R2 and R3 of your schematic, correct? What value would you typically use? 10nF? Just curious, as it sounds like you've used that specific set of values and are familiar with its audio qualities.

It's pretty quiet, and adds nice second order harmonic to add that elusive "warmth" when you push it a bit hard. My go to input stage for any solid state guitar amp is a 1M input resistor, 10k gate stopper, and a J112 with a 4k7 source resistor. It also works surprisingly well as a "warmer" for your DAW, at 0dBV you get around 0.5% to 1% 2nd harmonic. I have a couple of these I use with aux buses when doing masters if I want an old school analog warmth. A simple self biased JFET source follower sounds better than any plugin I have found so far for this purpose.

[Rob Strand]

The two cases for the buffer:
- a buffer that's just a buffer, like in a buffered bypass, where you don't want to change the signal.
- a buffer that's part of and effect where you might allow the buffer to add some character to the signal.

They are quite separate goals.

[Eb7+9]

nothing beats real analog of course, but i like to use fruity waveshaper if im working with samples or other computer only instruments in a song/beat i like to set the input gain to the smallest setting and then turn up slightly until the warmth comes in enough. sometimes i even use it on the master bus to add slight compression and warmth to really smooth out the mix

very cool to read this Iain,

I posted a hardware-emulation of this mechanism years ago which is based on my Triode modeling work, and what you're saying very much mirrors how I and others use my circuit in a guitar chain (last in line just before a tube amp) ... not surprised to see this appearing as a non-linear wave-shaping plugin as I've been selling my Tube Amp book to digit-emulation heads more and more lately ...

take a look at the DC-transfer curve that accompanies my short article (taken from the Triode modeling part of IFMTA) :

http://www.lynx.net/~jc/transferCurvature-TubeSimulation.html

Part of the reason why I see people perhaps not getting the idea fully is because I didn't publish exactly what I'm using in practice, but rather the theoretical equivalent in order to demo the idea. Using a jFET is only one way to get that curve, I've since discovered two other (also simple) methods.

When done with correct polarity, the wave-elongating properties of the circuit causes the front stage of a tube amp to bias shift and hence drop its gain as a function of signal dynamics ... in fact, we've found it works especially well with older low-gain designs to "wake" them up.

cool to see this in plugin form and to hear about you using it on mixes ...
~jcm

[jonny.reckless]

Jonny, you would add a decoupling capacitor between the input and the intersection of R2 and R3 of your schematic, correct? What value would you typically use? 10nF? Just curious, as it sounds like you've used that specific set of values and are familiar with its audio qualities.

I don't normally bother with an input cap when building personal projects, but I would probably use 22nF simply because I use a lot of that value in various inter stage coupling caps. The input cap tends to make the whole preamp slightly microphonic especially with high gain preamps. If you are going to use one, and you have 3 or more gain stages (60dB of gain or more) I'd suggest coating it in hot melt glue to dampen the microphonics.

[jonny.reckless]

The two cases for the buffer:
- a buffer that's just a buffer, like in a buffered bypass, where you don't want to change the signal.
- a buffer that's part of and effect where you might allow the buffer to add some character to the signal.
They are quite separate goals.

I agree. You'd probably want an op amp for the first type. Anybody using a JFET source follower in an audio circuit is likely after the warmth and character of the "analog sound".

[iainpunk]

The two cases for the buffer:
- a buffer that's just a buffer, like in a buffered bypass, where you don't want to change the signal.
- a buffer that's part of and effect where you might allow the buffer to add some character to the signal.
They are quite separate goals.

I agree. You'd probably want an op amp for the first type. Anybody using a JFET source follower in an audio circuit is likely after the warmth and character of the "analog sound".

well, boss/roland use Jfet buffers and they are super transparent, its all up to the specific Jfet type and the design applied. jfets sound more clear/transparent with higher source resistors, but i agree, the Jfet warmth is a good function for it as well.

cheers, Iain

[BJF]

Hi There,


I could see two routes for low voltage applications such as 9V operated circuits both about equal in terms of signal handling and noise performance

In my experience J Fets make excellent buffers and the only drawback is voltage gain is always lower than 1 times but a couple of % loss is often of no concern

A, one could use say J113,J112 that would bias automatically fairly suitable or for J113 one could use a voltage divider at grid to raise voltage at source to half powersupply ( for a 9V supply that could mean just a bias divider at gate of 5M6 and 1M at a total  IZ of≈ 850K).
Nifty with the very simple buffer is that bias is stable and with J113 consistent through 100 devices mounted. J112 will bias slightly higher and is also fairly consistent between devices.


B. one could use say PF5102, J201 and hook gate up to midpoint supply if there is one or make a bias divider at gate to set source at half potential and for a 9V supply that could be roughly 2M2 and 1M8 at IZ≈990K
If you need to set precisely or cannot get tightly specified transistors then gate leak,a resistor can be hooked up to a trimmer divider that allows 0V-4V5

There are ofcourse a couple of thousand devices but above ones are fairly easy to find and consistent

Most Japanese 2SK XX have very pure materials and as a consequence very low noise-and they usually come in selected IDSS groups. Some commercial types are not screened for noise and it can be worth checking several suppliers to find some with guaranteed low noise. Also some like 2N3819 is typically low noise but IDSS and UP are
not defined and in 400 units some can fall into use B and some into use A and the rest inbetween  just as easily as you can get 100 units fairly consistent the latter being rare

Regardless of route in can be worth considering that PSRR is very poor and for 9V operation one could use a 7806 regulator to feed Drain giving the final buffer a PSRR somewhat closer to that of an OP amp. Even at 6V output swing can anyway be close to 3V pk-pk which would be close to typical outputswing of an OP amp operated at 9V.
PSSR can become important if the circuit the buffer sits in is daisychained with other devices and in difficult cases hum and digital noise can find its way to the drain and enter circuit.
At minimum  a lowpass filter in supply feeding the drain can be used at say 100 Ohms and 100uF.

Anyway have fun
BJ
BJF Electronics
Sweden