Anyone have more than one 3457a and compared their DC readings?

[rigrunner]

Would either of these be a reasonable part to replace Q111 with?

2n5565 http://www.farnell.com/datasheets/908607.pdf
U440 http://www.farnell.com/datasheets/910106.pdf

[Kleinstein]

The trouble is more likely Q111 not U111. U111 is just the output stage of the compound amplifier. The main properties like offset, noise, input bias are set by the JFETs Q111.  At low frequencies U111 amplifies the signal between pins 5 and 6 by a large factor, so it is normal to see noise a lot more at pin 7. Pin 5 and 6 should be very close to each others and a common signal there only shows how much the current from the current source around Q114 is fluctuating. So not visible fluctuation at pins 5 and 6 are good, but no surprise.

The U440 looks like having quite a lot of 1/f noise, though likely still acceptable. The 2N5565 looks better. However this is still a HF type FET - so it might work, but could show similar popcorn noise as the old ones.
As the modern replacements are usually 2 dies, one could also try two separate, matched audio JFETs like sk30, SK117 or SK170. One might need to test a few to find reasonable U_GS matching so the RV113 has enough trimming range.

[rigrunner]

Matched a pair of jfets in place of Q111 and R113 adjusted to suit. First short test looks ok.
Will need to log for a longer period of time to see if it deteriorates when warmed up.

[rigrunner]

It is still looking good after warm up. It looks to be less noisy than my original 3457a now too 

[Kleinstein]

The curves from meter 1 look like they have lower resolution. So maybe there is some setting different / wrong (e.g. faster reading or wrong range ?).

The data look good, though there is still some popcorn type noise visible, though much smaller amplitude.

With changed fets one should also test the input bias current, to see if is acceptable.

[rigrunner]

I don't think there was a difference with the settings on meter 1, but it was late in the evening when i did it so i can't rule it out. 
I'll run the tests again to be sure.
The meters have been warming up for a little while to do the bias test..

[Kleinstein]

The old curves from the good meter look similar to that of meter 2 now. Though more low frequency noise without AZ - that was expected (more thermal effect) from using two separate JFETs instead of a pair in one case.

When warmed up, the noise looks quite a lot worse than when cold - so maybe cooling it not working that well.

[rigrunner]

If you tell them what you are doing, it might be interesting enough for them to send you a free sample to see what the results might be...

I'll drop them an email and see if they are interested

I've tried a few different jfets and have settled on a pair of bf862 for the moment.

[rigrunner]

I'm having a hell of a time trying to get stable bias readings at the moment .
Meter 1 is giving between 4µV and 25µV, meter 2 between 4µV and 50µV 

I'm set up as usual - I've got a 10Mohm across high low and a piece of paper towel covering the meter terminals to try and shield against air currents.
We've got thunderstorms in the area at the moment could these be causing this?

[Kleinstein]

Thunderstorms usually go with high humidity and this might cause increased leakage if the meter is not warm. Pressure fluctuations from gusty wind might also cause some variations, though this should not be that much. Also capacitive coupling could cause trouble if not shielded.

Still 50 µV over 10 M is only 5 pA - this is still a very good value.

The 100 K protection resistors at the input already give quite some noise. So no need to go for absolutely low noise it would just be nice to have FETs with little 1/f and popcorn noise. Low input capacitance is not important, it's more like low popcorn noise and low thermal drift, if the non AZ mode is important. At least for the AZ mode, the FETs should be less noisy than the 100 K resistors - so something like 40 nV/Sqrt(Hz) at about 1 Hz is critical mark - this not that low to need really high end FETs. Its more like finding one that meets typical specs.

[rigrunner]

Here is the bias data for the two meters. There was quite a time difference between the two sets of data collected from meter 2 and the reading had shifted in that time.

Meter 1 R=9.59Mohm
Meter 2 R=9.71Mohm
Humidity changed from 52% to 56% as the electrical storm passed by.

[Vgkid]

Looks good.

[rigrunner]

Here is the data from the re run of the noise test.

[rigrunner]

Just a bit of an update on progress with the 3457a.

I've tried a few different jfets and changed a couple of other components in the input amplifier section to arrive at what seems to be a lower noise level.
The bias on meter 2 has gone up in the process. M1 currently reads 19µV whilst M2 reads 62µV.

[Kleinstein]

The bias current comes from semiconductor leakage, mainly the FETs and possibly some protective devices, but also from just leakage on the board. Leakage on the board can go up with contamination but also from just humidity / alcohol that diffused inside. So directly after cleaning it might be higher and it might need a few days to come down.

The noise looks good, or as good as one can get from this design - quite some noise is just from the input resistors used for protection, and also the ADC itself it not that low in noise, as you can see in the 3 V range.