JFET input stage low noise amplifier

[Gerhard_dk4xp]

Unfortunately, discrete JFETs are no longer as fashionable as they used to be.

There are new JFETS from TI:  (interesting thread)

<        https://www.diyaudio.com/community/threads/ultra-low-noise-jfets-from-texas-instruments.375079/        >

Gerhard

[CurtisSeizert]

I was in need of a weekend diversion, and was inspired by this thread... see attached simple amplifier schematic with NSD chart, using 4x JFE2140 and OPA210.

Might be useful for someone needing to put something together in a hurry?

Nice. I have played with these a lot, and I think the sweet spot for drain current is around 0.5 mA per JFET if you're interested in low frequencies. That said, the zero tempco drain current is in the neighborhood of 7 mA for the JFE2140, so the tempco may be a bit worse at lower drain current. Note that this is not properly reflected in TI's model, which has VTOTC=-0.45m. I haven't run the simulation, but I think that would put you in the neighborhood of 1.2 nV/rtHz NSD. Also, the offset and drift averages for each pair are not centered around zero, so you can get a little better matching if you flip half the pairs around. This could help with the increased tempco from running at lower drain current.

[sdouble]

The BF862 from NXP is long dead. But linear supposedly released, end of last year, an "upgraded version" of the BF862. It's named LSBF862.
Did someone ever put thei hand on this new beast ?
any feedback?

[mr ed]

Interesting thread. 
I have an old PARC 113 low noise amp that I am restoring (mostly battery issues).
Mine gets 6nV/root Hz at 1kHz which is Ok for its age.
I am looking at replacing the old front end jfets 2N5521 with modern LSK389A jfets.
The original jfets have tiny gates and are really noisy under 100Hz.
As I understand the circuit so far, all I need to do is rebias the new pair from 0.12 mA to around 1-2mA.
In theory 15dB less flicker noise should be easily possible. 
I will make a new thread on that if successful.

[Kleinstein]

With a significantly higher current one likely also needs to change the drain side resistors, not just the current source. If present, also resistors at the source side would need adjustment (both for the absolute value and offset). With a higher output current also the compensation may need a change (correspondingly large cap).  So more than just a small change.

[TimFox]

Those old Princeton Applied Research amplifiers are well-worth restoring!

[mr ed]

I took the time and went through the PARC 113 first stage to understand it properly.
At least 6 maybe 8 resistors will need to be changed as the cascode tail has its feedback and common mode nulling circuit there.
Four resistors must be 0.05% to keep the diff. Inputs close to being in balance at high gain - which are EXPENSIVE. Plus the lsk389 and the ls311 to sub for the ancient bjt pair ad811 (of which this ID is reused for an opamp). Not sure its worth it.

P.s. Those who have 113's or similar-  please check internally for battery leakage.
My unit had original NI-CAD batteries from 1992 that still worked but were just starting to leak.
Disaster averted.

[Kleinstein]

The amplifier may want the precision resistors for accurate gain, not so much for low drift. The dirft part is separate and would want adjusted, selected resistors to trim the JFET pair. If the input FETs are not broken, I would still leave the amplifier as is.
The 2N5221 are supposedly not so bad JFETs when it comes to noise. The actual performance can still scatter. The high noise may also have other reasons, like a supply issue. They use the batteries as filters in the supply.

Instead of a modification I would consider a purpose build preamplifier for the task that really needs lower noise.

[mr ed]

Agreed.

[David Hess]

The PARC 113 schematic looks a lot like the discrete instrumentation amplifiers shown in AB-103 (Page 25) and figure 6 of AN-104 (Page 224) in the old Precision Monolithics Incorporated (PMI) application book.  I think the difference is that PMI relied on current sinks to set operating current of the differential pair.

I do not immediately see where the operating current of the input stage is set with the PARC 113, unless it is through Q111, in which case adjust R140, and then the Q113 current mirror?  Then as Kleinstein points out, drain resistors R102 and R122 need to be lowered.  Maybe that is enough, although the amount of frequency compensation suggests that this will lead to trouble.

An alternative may be to add matched current sinks which attach to the source sides of Q100, and then lower the value of the drain resistors.

It seems like a lot of trouble compared to a modern implementation based on the two PMI examples shown below.

[mr ed]

This  is my understanding of the front end circuitry, i can and do mess up...i made some measurements at 10k gain, shorted inputs.

In the parc 113 q102 supplies constant current of 0.340mA which goes down to r108 and then  q103 .
The top of r108 is 5.4v and the bottom of r108 is 1.6.  This clamps the jfet voltage across drain to source at about 4v.
With the emitter of q102  at 1.6 - BE drop of 0.6 the jfet source is about 1v so gate to source is -1v setting the jfet current to about 0.200mA through the whole cascode (R105). The tail circuitry has 4 precision $$$ feedback resistors, the balance pot and the common mode cancellation feedback.

Bottom  line, r102, r122, r108 a bit and the 4 tail resistors r123, 126, 128, 130 need to change significantly. Maybe r110 a bit too.
As the current to get a lsk389a in in its happy zone is around 1mA and a bias of 0.15v, its a lot of precision rework thats more than i care to tackle.

Better off just using  new modern parts in a new front end. Btw, the whole amp only takes only 11mA, that would go up too.

[Kleinstein]

The FET curent is set / regulated via the drop at the drain side resistors and Q111 -> Q107/Q108.
One should get some 4 V  (12 V - 8 V) at R102/R122.
This is a bit similar to the old µA741 OP-amp circuit.

For a modern replacement I would not worry so much about the resistor part - the tricky part is more the capacitive part to tune the higher frequency CMR and keep the amplifier stable.

[mr ed]

Right. The manual addresses adjustments in case its unstable even as designed. Adding higher gain parts adds more interactions.
One could try a 2N5564 which is available in case of a needed repair but it too may trigger unstabality.
One can get build boards from China with modern transverter chips on them for 40 to 60 dollars, like this one:
ADA4530-1 fA-Level Electrometer / Transimpedance Amplifier.

[EC8010]

Beware that the ADA4530-1 can be quite finicky to use. I have used it at the 1fA level using a 100G current conversion resistor and one PCB had gentle oscillations < 1Hz. The problem turned out to be that the ground planes were not solidly bonded to chassis and allowing the four chassis mounting holes to connect to the ground planes fixed the problem! I would also point out that the ADA4530-1 has quite a bit of voltage noise, I would have thought you'd be better off with a JFET front end. I've paralleled five pairs of 2SK2145 successfully, although I had a suspicion that something wasn't quite right when I tried measuring the self-noise of an inductive source.

[David Hess]

Taking a page from the PMI examples where they used current sources to bias the input differential pair, I wonder if you could add current sources to both the source and drain sides of the differential pair to increase the operating current, lowering the voltage noise, without affecting the existing common mode and differential connections to the differential pair.  Imbalance in the current sources would be corrected by the existing biasing.