3458A, fixing Error 202 - Slave test: convergence

[manupthehills]

I'm currently restoring a nice unit with the 202 error, and I found an issue on the A1 board: the area involved is the input amplifier.
The X10 and X100 gains are fine, but when in the X1 configuration the amplifier circuit is in open loop. Q106, a 2N4338, is dead.
This J-FET is a little hard to find nowadays, but I should be able to source a few low risk parts.
Looking for alternative parts with comparable performances is proving a bit difficult, and here lays the question: what is Keysight using on the Black Edition of this instrument? From pictures found on xDevs (many thanks for the formidable amount of information) I can see that the current J-FET, in a TO-92 package, is marked LS304, but I can't find anything more than that. Is that a custom part?
I'm re-posting a detail of the picture found on xDevs

[Villain]

My guess would be its a part from "Linear Integrated Systems" (their parts have LS model numbers), it seems not publicly available though as you can not find it on their product page.

[Kleinstein]

I would also suggest a linear systems part, possibly an equavalent to the J304 which is relatively similar to the 2N4338, though lower voltage specs.

Other possible 2N4338 substitutes may be SK208 or J201 in some variant / case. One may want to select for low gate leakage: a fixed leakage is likely compensated as an offset, but the leakage tends to go up with temperature.

For finding JFET replacements the Fairchild AN-6609 is a nice resorce.

[manupthehills]

Thank you for having introduced me to Linear Systems! 
They are on my way to the office, I might inquire them directly (will ask about what they might have unpublished in their drawers)

[manupthehills]

Temporarily bypassed the first issue I think I now came to a full stop: Board A3 very likely has a defective U180.
+12ref is good, -12ref instead is -11.3V and as a consequence +5ref is at 5.7V ... 
Any suggestion beside replacing the whole A3 board?
Being in the Bay Area, would be great if someone with a Saint in Paradise (Keysight) could share a miracle and help me find a replacement part   

[manupthehills]

Update on the repair:
Had to go trough some hoops, but I was able to get a brand new A3 board (latest RoHS compliant version). That fixed all the remaining errors.
An approximate performance check against a calibrated 34401 shows no concerns, and the instrument is now undergoing the "service note 18A" test.
This 3458A had its first and last calibration (standard calibration) in 2006 and still agrees, in the 10V range, to the last digit with my 34401 most of the time. It was not used for quite a while I guess, so I'm expecting some settlement in the next weeks. It will be checked again against a 34465A that is currently being calibrated in Keysight.

There's one question regarding the input leakage current: it's a positive 2 to 4 pA, perfectly in specs, but searching around the threads looks like it should be negative instead. Personally I'm not concerned, but, am I missing something?

[Kleinstein]

I don't think there should be a given sign for the residual input bias. The current is the sum of several contributions that partially cancel out. There may be a general tendency for one direction, but this can change with parts used. JFETs are known to have quite some variability.

In addition to this in the AZ modes there is a charge pumping effect, that depends on the trim  (? factory cal ) via the DACs. The effect would be larger for low PLC settings.  I don't know when and how the settings are trimmed, but chances are they would be adjusted to get a rather small overall bias in the common cases (e.g. 1 and 10PLC). There could still be small differences (at which temperature or the algorithm) in how this is done. Because of the temperature dependence I very much doubt this would be done at power up - maybe with ACAL. It may be relevant to know if ACAL could also effect the input bias in the AZ modes (the result may still be relative consistent and not much change).

[manupthehills]

Agreed, it's a kind of tug of war among reverse polarization currents with several elements pulling on each side.
This morning I measured timed it again at -0.3pA; temperature is probably slightly different as well as humidity, but didn't keep track of that.

The test was always performed @ 10V range and AZERO OFF

[martinr33]

The 3458A should be used with AZERO active for the most sensitive measurements. The polarity of the offset can go either way.

What you should NOT do is null out these zero errors when testing the meter.

 KS also recommends running the appropriate ACAL before every test when calibrating the instrument.

[manupthehills]

Hi Martin,
The AZERO OFF is only for this specific test, that is to assess the input impedance of the instrument in the low ranges, where it should be >10 Gohms.
This is done by observing (timing) the voltage that develops across the input capacitance of the instrument, that is estimated in 200pF (2x 82pF, plus some more due to traces and all other "stuff").
In order to be sure to observe the effect of the whole switching and amplification chain, up to the A/D input, and in a static configuration, the AZERO has to be off. The range should not affect the measure (as it's affecting the feedback path of the last amplifier on the A1 board) but it's easier to read the voltage that is measured if it doesn't change.

If AZERO is on, the signal chain is getting severed for the time it takes to perform the zero measure. From the input perspective that means that the input current would jump back and forth between two values, plus the charge injected every time the J-FET switches are operated. This makes impossible to estimate the input bias current, at least with this method.

In the end is the effect of the bias current on the source we're measuring that we care most. This test covers all the lowest ranges.

[Kleinstein]

The case with auto zero active is the more important one. Essenially all high accuracy measurement are done with auto zero on. It still help to know that the AZ off case is not contributing much, but the switching part should still not contribute much. Getting > 10 Gohm in the non AZ mode is relatively easy. The more difficult part is getting this also with auto zero active in the 1 PLC case, as the switching spikes can depend on the voltage.

The 3458 has quite some adjustement (AFAIK 3 DACs) to effect the switching spikes. I still don't know how the settings are determined / at what time the best values are found (could be during ACAL or factory calibration as the 2 extremes).

To measure the average bias with AZ one could add external capacitance, like 1 or 10 nF to get a much slower drift of the votlage. The 1 PLC mode would give more effect.

[manupthehills]

My original goal was to check if, beyond having a defective A3 board, the meter was also somehow damaged by the previous owner rather than assessing the real performances of this  unit.

Of course the charge injection associated with AZERO is important in sensitive measurements. Without a direct reading of the input bias current vs time vs input voltage (oscilloscope plus very high impedance amplifier?) I would try to isolate the various contributions this way:
1) bias current with AZERO off (whole switching/amplification chain)
2) bias current with AZERO on and high NPLC (minimal charge injection effect plus first-section-only bias current considering 50% duty cycle)
3) bias current with AZERO on and low NPLC (maximum charge injection effect)

and then combine the three observations. The input capacitance looks being high enough to integrate the contributions and still get good observations.

In my case the total contribution @ 1 NPLC is -14pA around 0V
Meanwhile temp went up a bit since this morning, and the AZERO off bias current changed sign. So I'm now running ACAL ALL to see if it compensates the charge injection contribution:
Some minutes later, and the answer is... the same

[manupthehills]

Further update:
A week long "service note 18A" test with the new A/D board gave as a result a drift of 0.0023 ppm/day. Not required, but ACAL ALL was run always when internal temp was 37.2C +- 0.1C
The instrument was then disassembled and deep cleaned. Caps, line filter, fan, DRAM, and VFD, all was replaced. Found a couple of dubious solder joints that were fixed. Added OPT 001

Meanwhile the 34465A came back calibrated (and fully re-adjusted), and they go from agreeing to the 7th digit (Mx+B trick on the 34465A), when room temp is same as 34465's TCAL, (ACAL on both instruments) to up to 20uV difference at higher temperature, this when measuring a 9V battery. Not bad for being calibrated back in 2006!

The new fan is a Noctua NF-A6x25 FLX with a 51ohm resistor in series. This arrangement allows the fan to run @ 11.6V, with a big difference in noise to respect the original Papst. The internal temperature is a little higher than with the stock fan, at this point I'd estimate in +2.7C, at around 22C ambient temperature.

The instrument is now undergoing SN 18A test again, and I'm planning in keeping it running for a couple of months before sending it to Loveland.

Still haven't decided if to lower the voltage reference's temperature. Now it looks like I'll not be able to get the 2006 calibration certificate from Keysight, and this is suggesting to make the modification. But it would be nice to be able to accurately measure how much it drifted in all these years. I'll decide along the way.

For now there's a new toy on the desk to keep me entertained for a while: 3245A OPT002

[Villain]

Keysight Loveland Standards Lab moved to Roseville.

[manupthehills]

Keysight Loveland Standards Lab moved to Roseville.

That's actually good news. Close enough to drive and drop it in person (after the last round of calibrations I have mixed feelings about the door to door pickup and delivery service Keysight has in the Bay Area)

[manupthehills]

Further update:
Months ago I finally decided to tweak the reference board A9 with a 100k resistor (Vishay S102K) to lower the LTZ1000A working temperature.
The meter then underwent 10V and zero CAL (3245A with the help of the recently calibrated 34465A). A couples of weeks ago checking the zero stability with input shorted I noticed that it was out of specs in the 1V and 100mV ranges. STDEV for the 0V reading was around 130E-9 (100NPLC, 100 samples, not catastrophic actually) in the 10V scale, while the 2W ohm measure of the short was around 6 ohms !! for both inputs...
The culprit turned out being the short coaxial cable connecting the F/R switch to the A1 board (see pic). On one side the crimp was looking a little dubious so I pressed it again with pliers and gave the Deoxit treatment to all the connections in the meter. The switch didn't show any issues so I didn't bother it.
Results: the 2W ohm measure went negative 25 milliohms (so better than the previous CAL 0) for both inputs. After CAL 0 everything is in specs and the STDEV in the 10V range went down to 97E-9 (100NPLC, 100 samples).
I searched in all possible ways without success; anyone knows the PN of those collet contacts? The meter has been stable for a couple of weeks now, but I'll feel better after replacing that cable.

There are also a couple of things about the A9 and the modification, but I'll post that in the specific tread

[martinr33]

Details on those wretched socket connectors on the 3458A switch.

It is an obsolete Molex part, and I have not found a suitable replacement. The sockets are designed to go in a housing, but HP used them free hanging on a wire. I have attached the Molex datasheet that covers them.

PN per manual (old)    Updated P/N per datasheet
02-05-5204 (4)           2107-2C5N1  Phosphor bronze, selective gold, 18-24AWG, .040 ID, chained, Specified for .045 round pin, but used with square pin.
02-04-5216 (1)           2107-2C6N1  As above but 24-30AWG

The .040 ID is very close to 1mm
Phosphor-broze is important as it makes the terminal springy.
The gold plating uses the least gold offering in the datasheet (others are thicker gold or have overall gold flash)

The terminals are specified for round pins with a 0.044 OD, but Keysight used them with .044 square pins. The pins are 0.004 bigger than the socket for a tight fit.

The square pins are listed as 1.14 MM-BSC-SZ-SQ p/n 16-06-0034. (closer to .045", but that's likely a conversion issue).

The part no. for round pins is 02-05-5221 / 2107-2C5P1 but I would say the existing part is good enough.

So if anybody finds them let us all know! Or even a decent replacement.

[manupthehills]

I'm looking into the ground current compensation around the A9 board, but I can't find anywhere the value of RP400 (nor I can open the instrument and measure it). Anyone knows or can easily measure that resistor network?
Moreover, anyone has a picture, or can please take one, of the backside of the A1 board near the area that covers the A9 socket (P400 - 401) and the ribbon cable toward the power supply board? (P3 connector). I took the picture of all the boards, front and back, before and after the repair/refurbishment, except the back of the A1 (And Xdevs docs are not accessible...)
Thanks!

[martinr33]

A1 Front:
https://photos.app.goo.gl/eucXZgV14rKwBckDA

A1 Back:
https://photos.app.goo.gl/rsRESApB3PATg6oaA

[manupthehills]

Awesome! Thanks, Martin!