Jfet matching

[lowell]

Do jfets need to be matched if they are being used as a basic voltage amplifier common source arrangement? More specifically, if I'm making a bunch of amps or pedals that have one jfet for voltage gain, how important is it that the jfet of each amp or pedal be matched from unit to unit? I realize it's very important in modulation effects, but is it also important in this setting? I'd like the gain staging to be consistent from unit to unit. Thanks!

[Enzo]

Pedal to pedal???

Will anyone be using two of them side by side, and if so will they notice one is set to 4.63 and the other 4.64 on the knob?

Matching tends to come up when more than one part has to work together. Line up a half dozen new Twin Reverbs, and see how consistent they are, knowing the 12AX7s in the preamps are not matched in any way.

[lowell]

I expected that response cause I was thinking the same thing. I'll rephrase the question.

The jfet is the first gain stage in the amp...then a few tube stages. I am just curious how much the GAIN will change from fet to get. The amp has a great overdrive sound and I don't want to have one that has tons of gain and then one that is weak on the gain. How much does quiescent Id affect gain? Or does it not?

[Enzo]

Build one unit. Put a socket in there for the JFET. Stick a bunch of them in and see how much they vary.

What socket? I use DIP sockets. I grab a 14 or 16 pin one from the drawer, and snip it with my dikes to leave three holes, and install that. Shove the transistor into the three holes.

[loudthud]

JFETs can be all over the place from batch to batch and from vendor to vendor. If you go by the data sheet you can see a five to one variance in some parameters. In practice most parts you can buy through normal distribution channels vary about two to one. Outside the normal channels, surplus stores and Chinese Ebay sellers, you can expect a wider variance. Some parts can be bought with a letter suffix that will narrow the variance.

Many production circuits that don't count on closely selected parts use lots of Source degeneration to stabilize gain and operating point. Such circuits are usually within 1 or 2 dB for gain and overhead. Some of the circuits you find on the internet have a trim pot of every JFET. You can expect a wider variance in these circuits, up to 6dB, but without the trim pot some circuits might not pass a signal with all JFETs.

[darkfenriz]

The jfet is the first gain stage in the amp...then a few tube stages. I am just curious how much the GAIN will change from fet to get. The amp has a great overdrive sound and I don't want to have one that has tons of gain and then one that is weak on the gain. How much does quiescent Id affect gain? Or does it not?

Not much. Maybe twofold change for bias extremes and much less than that if you use source degeneration resistor.

[J M Fahey]

Fets are all over the place, so much so that they ceased to be a mainstream component, go figure , and most "audio" (MPF102 / 2N5457 / etc.) types are being obsoleted,that´s why they ask 5 or 7 bucks each for remaining ones.

Only types which remain in massive production are switching types. (J11x series) which are terrible for Audio, very low gain.

Please check old 70´s and 80´s Yamaha and Roland schematics, notice they
a) use Factory selected parts, indicated by a colour dot applied to case.
With the colour letter added at the end of type number, such as xxxx(y) or xxxx(g) for (y)ellow or (g)reen ... or any other colour.
b) since Vgs varies a lot, they "force" extra voltage on the gate so compared to it, actual Vgs becomes less important, the point being able to assemble amps by regular assemblers instead of needing to have each and every amplifier adjusted by a Tech or Engineerr.

In my nightmares I dream of a factory turning out amplifiers at high speed ... only to bottleneck them at some adjustment desk where a Tech clips his measuring stuff into each one and spends 10 minutes trimming trimmers.

I sometimes use them, but I buy 50 or 100 at a time, measure Vp (cutoff negative bias) or, more useful in practice, how much voltage do they drop across a 4k7 source resistor, on which I normalize.
Once I know that voltage, I separate them in 3 little boxes.

* Those which drop around 2.5V into 4k7, are good general purpose ones, and with a load resistor of 22k (across which some 12V will drop) they are fine being fed by some 24/25V and provide some 20/25X gain.
Perfect for those Randall preamps.

* the few which drop around 1.5V , are saved for higher gain stages (mind you, only about 10% Fets are that good) , used with 33k to 47k load resistors and 24V supplies for very good 30X/40X gain OR can be used in pedals with much lower load resistors, thing 10k or less, whatever is needed to drop about 4V across them.
Gain sinks a lot.

Many forget that JFets are NOT "triodes" but "pentodes".

Now triodes are quite self adjusting so designing with them is quite forgiving, while pentodes are stubborn beasts which are hard to tame ... if possible that is.

Almost forgot: about 40% of modern Fets will drop around 4V into 4k7 ... will provide terrible audio preamp specs or designs, but are fine as "cathode followers" wher the extra Vgs is actually an advantage, as it improves headroom, *OR* signal switchers , the actual purpose modern ones are designed for .
Similar to pedal in/out switchers as used in classic Boss or Ibanez and most Japanese pedals.

So I suggest you buy a bag of modern ones and select them.

[mozz]

I would say buy a transistor checker that also tests fet's, they often measure gain, Ebay has them less than $20 but read the specs first.

[lowell]

Thanks guys. So it may be better to go with a jfet input opamp? The other issue I'm having is the white noise. I only want a gain of 2 from this thing. Very subtle but creates the amount of gain I'm looking for in subsequent tube stages.

I have been toying with Rd and Rs values to try to mitigate the white noise. I read a post on some forum a while back with a reference to RG and his "happy fets." It does seem like reducing those resistor value and allowing for more quiescent current reduces white noise.

My jfet setup:
2n5485

Rd 10k
Rd 4.7k
Rinput 100k (gate series resistor)
1M input z resistor
24v supply

I also then read to shoot for no more than 75% of max spec Id. So I calculate a Rd of 2.2k and Rs of 1k to get "closer" to that. It seemed to help...but there is still a good amount of white noise.

What are your suggestions in mitigating the noise?
Would an opamp cure both issues?

[eschertron]

What kind of input headroom do you expect? Using a jfet with a larger Vgs(off) may play nicer with pedal-level inputs. I'd reserve the 2N5485 for microphone inputs where I want lots of gain.

[Enzo]

And how much white noise do you HAVE now?

[R.G.]

It's much simpler to combine an N-channel JFET with a PNP bipolar in a feedback pair. The feedback will make the gain predictable from unit to unit even with variations in the devices.

That's what feedback is for.

[J M Fahey]

Very true.

But then, it makes no sense to use a discrete FET at all.

Justification behind Fets is that they are non linear and add presumed "good" even order harmonics.
Once you get them inside a NFB loop, and a strong one at that, they get not only stable but linear ... no fun in that.

You might as well use a TL072 which costs 50 cents, and has two very predictable and accurate Fet input Op Amps.

Did I say "low noise"?

[lowell]

RG does your last comment also come with linearization?

Enzo. Tons.. but that's after 3 more mostly undivided tube gain stages.

I just fixed it. The noise that is. I had a 470k resistor from the last B+ node. That feeds the 24v zener and filter. I replaced that 470k carbon film with a 24v zener and a 100k metal film resistor. Killed it. Does that make sense to you all? Was it just lotsa johnson noise from that 470k? Seems that a zener to dump a lot of the voltage and a smaller resistor was the key.

[R.G.]

Yes, the feedback setup does linearize the FET. Pretty much any time you see "feedback", you should mentally substitute in there "linearizes and stabilizes everything", at least for negative feedback.

Knowing more about your circuit now, some things are clearer. Yes, the noise was coming from the 470K resistor's interaction with the rest of the circuit. Maybe it was all thermal noise, maybe not. Maybe part Johnson and part RF oscillation.

Single ended stages have essentially zero power supply noise rejection. Noise on the power supply flows right into the stage and gets combined with the output signal in proportion to the passive components' values. Ditto for ground noise rejection - it's zero. Worse than zero, actually, since input ground noise is amplified. So any noise coming from the power supply is amplified. In your case, that's whatever comes from the 470K and is not filtered by the "filter" you mention.

But the filter needs to be better for a JFET than for a tube. A 12AX7 struggles to have gain above 100kHz. Pretty much any signal-type JFET will happily run at a couple of hundred MHz. That's significantly higher than anything other than specialized o'scopes will show. A more common 20MHz scope can't see signal up there at all. If your filter wasn't good at filtering things well over MHz, the stage >could< have been singing at a frequency you can't see, and the phase jitter of this oscillation been being heterodyned back down into audio as hiss.

If this was the case, the 470K was acting like a great big load resistor, the filter was not in evidence, or if it was an electrolytic only, was acting like an inductor and tuning the output of the JFET at RF with the stray circuit capacitance. Changing the 100K for a zener and a 100K >could< have cut the RF gain enough to quell the oscillation.

This is a little bit of a "here be dragons" tale, not knowing the specifics. But I've been burned this way a couple of times, with JFETs, opamps, and bipolars. It's one reason I sweat blood over JFET layouts and put places for 0.01uF ceramic bypasses on every opamp I put on a PCB. They're just too good in some instances.

A lot depends on the quality of the local bypassing. I'd be very interested in what happens if you put back in the original 470K (which could have been an outlier itself, with high excess noise) and do a really bang-up job of filtering right at the JFET load resistor, with a small dipped epoxy ceramic cap of 0.01uF or so from the top of the load resistor to the bottom of the junction of the input pulldown resistor to the bottom of the source resistor.

I'd also like to see what happens to your noise if you ditch that series input resistor. It's thermal noise is directly added into your signal, no matter what you do.