HP 3458A repair - Convergence error with wierd workaround

[IanJ]

Hi all,

I acquired a faulty 3458A and set about a repair, there's a couple of repair videos on my channel.
It had been a wee bit abused (seems so common these days!) which I set straight, the only strictly electronics related repair I had to do as such was replace a blown fuse (F701) on the A2 (AC) board.

It powers up with a "Error 202 Hardware Failure Slave Test: Convergence" error.
The display is blank except for the units.

I can set TARM AUTO which brings up the digits but they only update once per sec irrespective of NPLC etc.
If I set TIMER 0 then it speeds it up per normal.

All modes on the meter work and appear accurate.

ACAL DCV = passes
ACAL OHMS = passes
ACAL AC = fails with Error 209 Timeout Unable To Read A/D 13 (number has varied)

So, eventually I looked at the A3 board vrefs and measured:
TP160 = +12.152
TP165 = -12.171
TP151 = +5.0918

Thinking perhaps the +/-12 are not matched closely enough I added a parallel resistor across the gain resistor of U165 (-12v) such that TP160 & TP165 are almost exactly the same, but the meter still boots with a convergence error.

Now it gets curious!:
I started looking at ZR_LO2 which I believe is the low-side zero-reference summing node, long story short if I connect +5vdc through a 1k resistor so that 5mA is injected into ZR_LO2 then the meter boots up with no errors whatsoever, and it appears to work fine.

ZR_LO2 without injection = -0.25mV
ZR_LO2 with injection = 2.7mV

The injection also brings the +/-12 vrefs slightly closer together, which was my original reason for the injection in the first place!
The injection throws the reads off a bit, i.e. 1.000000V input I get 1.0001V in DCV mode.....etc.



But, If I disconnect the 5mA injection after it's booted then accuracy returns I get 1.000000V on the display from my 1V source.

At this point:
ACAL DCV = passes
ACAL OHMS = passes
ACAL AC = passes
but TEST = I get error "Error 202 Hardware Failure Slave Test: Convergence"

I am no U180 expert, but I think I have fooled the startup convergence/self-test, and once past that all is ok.

One other thing which appears it may be relevant, ZR_LO2 has a 0.5ohm resistance down to the main A.Gnd on the A3 (the path back to the A9 board & LTZ1000 and catching a few friends on the way).
I have pulled the boards and cleaned all the connectors (Kleinsteins suggestion), but from what I can see the 0.5ohm is built up from long traces through the A1 board, down through the chassis interconnects and back to the A3 board. I'd be interested to hear what other people's 3458A measure here.
Easy way to measure is like this (yes, the 2-core isn't re-connected yet):


Obviously I want to get the meter fully working, so looking for suggestions/ideas what to look at next before I start thinking the U180 is knackered and then trying the NU180 (which might be fun).

In the meantime I am running it to test for drift.

Thanks,

Ian.

[Kleinstein]

The power on self test likely has a test that checks the the +12 and -12 V ref weights are balanced. This is needed to get good DNL from the ADC.
The test would include U180 internal 80 K resistors (or the 20 K resistors) and with drifted resistors the balance can be a bit off exact symmetry for the +-12 V. One can likely also trim the -12 V in a similar way to also pass the test.

If the error in U180 is still small one might get away with just that trim. I should also correct for at least the main DNL error related to the wrong voltages.

0.5 ohms as resistance for just the trace still looks surprisingly high. Ideally there should be only little current and thus only 0.25 mV measured.
There is the resistor R168 towards -12 V to compensate much of the current to Zr_lo from the A3 board. There may also be a little drop from the A9 boad to start with

[vf33184]

Hello Ian,
I measured the resistance of ZR_LO2 to analog ground at the same points you showed in the photo, and it reads ~0.55 Ω on my meter.

My meter is also faulty (I'm troubleshooting that in my own thread), so I'm just flagging this in case the resistance turns out to be an issue later on.

Vladimir

[IanJ]

Kleinstein, R168, nice spot!…..I’ll take a look at that……could have big implications on my A3.

Vladimir, thanks for that….might have another path for you to measure.

Ian

[qzh3887896]

I measured the resistance values at the corresponding positions on the two 3458A units I have on hand, which read 0.64 Ω and 0.95 Ω respectively.

[IanJ]

I pulled R168 the 10.2k resistor that's between -12vref and ZR_LO2.
I then connected a 3.3k resistor between ZR_LO2 and +12vref. This results in a net 3.8mA injection into ZR_LO2.....matching what I had before.

The 3458A powers up with no errors, and all seems normal.
However, having the 3.3K resistor affects the reading, it pulls it from 1.000000V to 1.000205 (+205uV), so likely tempco is a big issue here.

This got  me wondering about the original R168 10.2k down to -12vref, it's a normal 1/4 resistor, and so I reverted back to it and tried varying it's value and sure enough, it affects the readout.
Doubling the value to 20K took the readout from 1.00001V to 1.00003V (+20uV).
I can only assume this would not be the case on a fully working U180 otherwise the 3458A's tempco would go out the window! Wondering if my U180 has a stuck or leaky switch?

So, looks like this can only turn out to be a curious experiment.

Ian.

[Kleinstein]

The idea of R168 is to compensate the current from the 7 to 12 V amlifier part in U180. So ideally have essentially zero current to ZR_lo. This is how it should be.

There can still be some residual voltage difference between the ground and ZR_lo from current flow on the A9 board itself and how good the ground current is compensated on the A9 board.

So while in this example it works to shift ZR_low, it is not a good way to trim. With current towards A9, there would be additional TC and more effect of contract resistance.

The better way would be to tweak the 12 to -12 V stage, or maybe R154 (offset for the ladder style DAC for the fine slopes).

[IanJ]

Those golden 3458A's that have a better than normal, somewhat unusually good tempco (doesn't Xdevs have one?)......a contributing factor might be virtually zero current through ZR_LO2.

Ian.

[Kleinstein]

The current through R168 and from U180 should be in the 1.2 mA range. With roughtly 1-2% accuracy for the compensation the residual current should be only some 10-20 µA. This would be only some 5-10 µV of drop at the 0.5 ohm resistor and not much TC from there.
The issues are more the TC of the LTZ1000 ref. board and the resistor / switch matching inside U180. Some units just have better luck than others.

A part if the TC can be from the Zjump correction. This could give a relatively constant contribution, but still some variation with the switch R_on.
https://www.eevblog.com/forum/metrology/3458a-worklog/msg6262738/#msg6262738