Non-Phool jellybean audio-frequency JFET

A few BF862s in parallel, for choice. About 0.8 nV 1-Hz noise in the flatband, 1/f corner around 1 kHz.

Cheers

Phil Hobbs

ay.

OPA2134? ~$2 at the 100 quant. 8nV/rtHz, (only 8MHz)

George H.

ay.

Hey, I was leafing through electronic design at lunch, and thought of you when I saw an add for Jfets from linear systems. (linearsystems.com)

George H.

BF862s, really. They're the cat's pajamas.

Cheers

Phil Hobbs

BF862s parallel just fine. But for audio, even just one is better than good enough. There's no need for subnanovolt noise in audio.

Cheers

Phil Hobbs

se

eap

ts

s D

se

lly

do

or

o away.

t -

Geesh Phil, I don't know. Mostly I can't tell the difference between a Jfet and a J stroke.

(They've got BF862's at newark for ~$0.30/100) ... I might have a use for a 0.8nV Fet. Where's the 1/f knee? At the moment it's a ~10kHz resonace, but I could move that up, some.

George H.

oise

BS,

cheap

its

t -

You 'pay' 10pF of input C for each one.

(According to the 'one' npx spec sheet I looked at.)

George H.

cheap

Right, but this is audio after all. I use them in boatloads to make photodiode bootstraps that work up to ~20 MHz. You put a good-quality current source in the sources, which gets rid of Cgs pretty well, and then bootstrap the drains to get rid of Cdg and Cds. You still get the voltage noise differentiated by Cdiode+Cin, but since they're so quiet, that's still a big win.

I'm experimenting with using BF862/pHEMT combos with diplexers to get low 1/f noise and extreme bandwidth.

Cheers

Phil Hobbs

away.

It's about 1 kHz. They're firmly in the flatband at 10 kHz.

Cheers

Phil Hobbs

Hey Phil, Do you have a compliment to the BF862? Mikek

Ok, ok, Hello BF862, you've got some really nice parameters there. Do you have a complement for the BF862? Mikek

Don't I wish. Try a BF862 in an inverted cascode with some nice quiet PNP transistor, e.g. a 2N5087 at low frequency or a BFT92 at high frequency.

Cheers

Phil Hobbs

Ya, most of this is over my head, I was just thinking about replacements for an AM radio antenna preamp circuit I know of. Is this a physics problem building a P type to match an N type? Thanks, Mikek

They fit the mold of what I was thinking might be the case - an "RF" part with application elsewhere. I had actually found Phil's earlier postings praising them (and verified that they were still available) before I posted, but was not at all clear after looking at a datasheet that cut off frequency specs at 0.1 Mhz on the low end if they'd work reasonably for audio, given that they were being touted for RF in car radios and a lot of what Phil does goes up there a ways.

Quite a price variance between digikey/newark/mouser, with mouser winning strongly (27 cents for 1, 23.3 for 100.)

Given a lack of directly comparable graphs I have a hard time telling if the slightly more expensive (43/29.9) and smaller (smaller not being an advantage for me as a one-off tinkerer) 2SK880 has a _slightly_ lower corner frequency or not really, but I suspect it's not enough of a difference to actually matter (not going phoolish), given that I'm mostly looking for a reasonable part to have on hand and play with in the same way I have a pile of 3904/6, LM833, etc.

Now to see how I do with Larkin's x-acto/copper foil prototyping technique, since I can't see getting boards made for every iteration of goofing around I want to do, and these things are seriously small...

I bought a reel of BF862s a year or two back for $650, or about 22 cents each.

Using SOT23s in protos isn't too hard. You can mount them easily on the pad-per-hole style of perf board, or get some prototyping adapters, e.g. the Bellin Systems ones.

Cheers

Phil Hobbs

Yes. The hole mobility in silicon is low, which makes the transconductance low, which makes P-channel devices noisier.

BJTs don't have the same issue since the base is so narrow and the transconductance is very high, independent of device polarity. (PNPs used to be slightly quieter than NPNs, but not any more.)

Cheers

Phil Hobbs

noise

r BS,

g cheap

ut its

fet

class D

these

no

o be 5

ginally

s to

, and

t may

ets

ely

e I do

Khz, or

e for

ll go away.

of

Great, that's perfect. A colleague was chattering about active damping (as was used to damp torsional fibers back in the day.) And I mentioned that we could try active damping to kill some of the johnson noise in a high Q RCL circuit. A nice low noise jfet might be perfect. But I need to think about it some more.

George H.

t -

"Ecnerwal"

The sample to sample parameter spread is massive - so much so that it is normal to select devices for a given circuit so that bias / operating point conditions will be met.

If you need diff pairs with low input offsets - then be prepared to waste a lot of FETS.

FET input op-amps and matched FETs on a chip are the way to go.

.... Phil

That's all too true of every JFET that I know about except one: the BF862. Have a look at the datasheet--they're magic. I'd never use a JFET in anything if it weren't for these ones. Their transconductance is very high, so the action is all over in about 400 mV. They're very predictable for a JFET, comparable to a pHEMT, and almost as good as a BJT. You just parallel them up and away you go.

Cheers

Phil Hobbs