Keithley 2010 Repair

[KungFuJosh]

my guess you havent adjust current for 1N4579A to TC=0 ?

Not yet. I'm getting the ADR1399 tomorrow, and I assume that will change it anyway.

Is it correct that adjusting to TC=0 should improve measurement stability? If so, I'm tempted to do that before and after changing the reference for a more accurate stability comparison.

Thanks,
Josh

[Kleinstein]

The 1N4579 TC depends on the current and it adjusted that way. It would not be effected by changing the LM399 to a ADR1399.
The improvement from a more stable 1N4579 with an adjusted current could be less noise from temperature fluctuations or during changing temperature.
How much depends on how bad the initial TC is - after all this are already low TC zeners, even without a trim of the current.

There would be a tiny effect on the longer time stability, as the zener also set the current to the LM399/ADR1399. However the ADR1399 has low R_out (0.04 ohms typ) and there is thus quite some ( ~ 30000 fold ) attenuation of the zener drift for this path.
The issue is shorter time (~ 1 minute) for the times when the gain is adjusted.

The effect of zener TC would be visible as a kind of saw tooth when measuring a stable reference during initial warm up after turn on.

[KungFuJosh]

R359 provides the current for this 6.2 V low noise zener diode. The TC of this zener diode depends on the current and if the TC is high one gets some extra noise from thermal variations.  Once things are repaired, one could check the drift of the 6.2 V or amplified +-12.4 V and if needed replace R359 with a slightly different value.

Are you sure about R359 being the one to change? @GigaJoe said R459 (I haven't found this in the schematics), which does connect to the Zener:

ignore adc ...

so schematic a bit ( tiny) incorrect .. VR122 - actually TC=0 zener , not a basic one. bottom orange diode. Vishay resistor R459 actually low TC and defined current. ( R359 on schematic )

I'm not sure I understand. R359 is 3.01kΩ in the schematic, and on the PCB (bottom side). R459 is a 1.28MΩ CMF series resistor (PCB top side), and measures about 51k in circuit. It does connect to VR122. Next time I have the PCB out of the case, I'll see if I can determine the relationship between those two resistors...if I remember.

When replacing the LM399 with a ADR1399, one would need to increase the current to this reference, so change R216, that is in the original configuration some 2 mA with some 1.7 K to get 3 mA there.

There is averaging for the LM399 reference (at least in 10 PLC and slower AZ mode). However the LM399 popcorn noise also has slower components, that are not full filtered out. So the LM399 reference noise can still matter. This may not be as important with the 2001, but it could be with the lower noise 2010.

Is there any reason I would not want to increase the current higher than 3mA? The datasheet says 3 to 13mA for the ADR1399. It seems it would be easier to target the middle of the range, and not worry about being exactly ~3mA.

The 1N4579 TC depends on the current and it adjusted that way. It would not be effected by changing the LM399 to a ADR1399.
The improvement from a more stable 1N4579 with an adjusted current could be less noise from temperature fluctuations or during changing temperature.
How much depends on how bad the initial TC is - after all this are already low TC zeners, even without a trim of the current.

There would be a tiny effect on the longer time stability, as the zener also set the current to the LM399/ADR1399. However the ADR1399 has low R_out (0.04 ohms typ) and there is thus quite some ( ~ 30000 fold ) attenuation of the zener drift for this path.
The issue is shorter time (~ 1 minute) for the times when the gain is adjusted.

The effect of zener TC would be visible as a kind of saw tooth when measuring a stable reference during initial warm up after turn on.

I'll test the diode current before making other changes. However, I assume after I get it right, I will need to replace the resistors I use with higher quality/lower TC versions. From the datasheet, it looks like I'm to target 2mA for the zener TC=0.

Stability is an issue currently (see attached). It could be the TC=0 is too far off, could also be the LM399 is not great anymore. I assume it's more to do with the Zener change, but stability wasn't great before the repairs either (which I suppose would be obvious ).

Thanks,
Josh

[GigaJoe]

that idealistic current curve far from the real thing , i got TC=0 mA as 1.34 (was 2.2) and 1.43 ( was 2.4)  ,  a good thing it not so steep curve, so steps of 100 ohm for resistors works well.

yeah that Vishay resistor ...  you can ring connectivity what resistor directly attach to zener and +12 volts , right? ...

see ... LM399 not involve directly to measurement ...  it a reference point to recalculate exact value periodically, for example 2001 do it every 90 sec (approx.) but ,it basically take a set of signals such as: reference , input, and az-point that measured every cycle connecting it by ADG switches to adc input, and then recalculate it. and correction of AZ + reference ( I personally believe it averaged value) change value periodically ...  as it not a direct measurement it will up and down ...
noise, spikes..  averaging , a few days :  https://i.ibb.co/V0RSqNYM/voltage-deviation.png

btw..  when mount adr instead lm ,  i would personally do not cut legs ... but bend it accurately in the way to make some sort of sprig, the way of mechanically separate package from the board.  add an additional enclosure , and expect to drift it for 1 year at least ... in hope it will stops ...so maybe i wouldn't do it ...

[Kleinstein]

The ADR1399 current is not that critical. It should also work with a little less than 3 mA. Much higher current would need more power and it also make the copper trace more important. So I would aim for 3-3.5 mA. The resistor to set the current should not be super critical.

For testing if the 1N4579 TC is OK one could looks at the measurement when starting from a cold meter. This should give some kind of saw tooth - the better the TC the smaller the amplitude.

For the short time noise there are also other parts contributing, not just the references.

[KungFuJosh]

Assuming I'm supposed to measure the current across the diode...the issue is pretty clear. It started at ~0.45mA and is dropping slowly as it warms up. 0.5mA is below the curve in the datasheet.

[KungFuJosh]

I don't know about R459. If that was the current drop resistor, shouldn't we see some stability there?

[Kleinstein]

The current through the 1N4579 zener should be set by R359. With some 6.2 to 6.4 V and 3 K this should be a little above 2 mA.
It is not about measuring the current, but checking of the zener voltage or the amplified 12.8 V ref. changes a lot with temperature.
If the changes are small, I would leave R359 unchanged. If it changes significant, one could consider changing R359. The 1N4579 is supposed to be already within +-5 ppm/K at the norminal 2 mA, which would suggest that the TC zero should be within about 20% of the nominal 2 mA. A different zener type may be further off. So it would be more iterative guess-work, based on the measured change in voltage (e.g. in the first few minuted after turn on).

The schematics could be wrong with R359 - maybe a typo and this is R459 ?

[KungFuJosh]

The current through the 1N4579 zener should be set by R359. With some 6.2 to 6.4 V and 3 K this should be a little above 2 mA.
It is not about measuring the current, but checking of the zener voltage or the amplified 12.8 V ref. changes a lot with temperature.

I don't understand. If it's not about measuring the current, why is TC=0 shown as a curve based on test current of the diode? Is that not supposed to be measured across the diode?

If the changes are small, I would leave R359 unchanged. If it changes significant, one could consider changing R359. The 1N4579 is supposed to be already within +-5 ppm/K at the norminal 2 mA, which would suggest that the TC zero should be within about 20% of the nominal 2 mA. A different zener type may be further off. So it would be more iterative guess-work, based on the measured change in voltage (e.g. in the first few minuted after turn on).

I will test the voltage stability over time of VR122.

The schematics could be wrong with R359 - maybe a typo and this is R459 ?

R459 is a 2.58MΩ CMF THT resistor on the top of the board (it's not in the schematic at all). One end does connect to the zener, the other end connects to Q178 (see attached).

R359 is 3.01kΩ SMT on the bottom side of the board, and that matches the schematic at least in value (I've not test the connections yet - I will eventually).

Thanks,
Josh

[GigaJoe]

for zener TC=0 , no it not about current over diode at all ... don't believe what do you see in a datasheet on that diode ... it far from reality
what do you need to do is measure voltage  across the zener , or 12 volts as derivative from the zener , that actually using by ADC ... ADC do use +12 and -12 , not lm399

you have to measure it right , of course,  cold to hot with fully closed\assembled device.  ideally 10-20 microvolts cold->hot condition ... large span - worse state ..

[Kleinstein]

At least measuring the current over the zener should not have done damage. The measurement messes up the voltage and this way cause too low a voltage and thus less current than in normal operation.

R459 seems to add a tiny bit off offset to the input. Not sure why - it could help a little avoiding exactly 0 V for the AZ case as the 0 V point (exactly 50:50 ration in the run-up) is one of the worst points of the ADC.

[KungFuJosh]

Here it is with the case open- voltage measured across the diode.

I could run the test again with the case closed too if the probes will fit.

[KungFuJosh]

Here's VR122 with the case closed.

[GigaJoe]

yep .. same thing i saw in my 2001 ... so imagine 2 millivolt X2 to make 12 .. 4 millivolt difference has to be compensated by software.
and because it far off from TC=0 , it a bit wobbling in microvolt range and that also compensating by software ....

So i add decade resistor in sequence to defined current resistor , closed case and monitor voltage while zener heats up ,  adding resistance up as device up, till the voltage stops going up .  than wait, double check, replace resistor as a total sum of chain ,   close case double check .. adjust again .. .  The final result i have 20 to 40 celsius , +20 microvolts...   

as well i raise the zener on wire from the board and use some fluffy paper towel to make heat insulation,  together with current resistors, ( it appears 3 resistors instead one, as i need such precise value)

btw, im lookin now at  https://www.eevblog.com/forum/projects/multislope-design/   , where some discussion was to mod ADC for K2010 ...

[KungFuJosh]

I'm going to try focusing on R359 first, as I think @Kleinstein is right about that being the current source for the zener.

The question is, how can I accurately measure the desired resistance without 6000 resistors and trial and error? Should I solder in a 2k - 5k potentiometer, or is there some fun math we can do?

[Kleinstein]

One can now use the curve from the zener datasheet as a help. One still has to estimate how much the temperature changes. With lets say a 15 K temperature rise for the warm up one has about +0.1 mV/K as drift for the zener.
The means the TC is at the upper limit (or a bit beyond that) for the 1N4579  (maybe it's only 1N4578 ?) and would need to reduce the current with a larger value for R359. To change the TC by -0.1 mV/K it would need something like a 20% reduction in the current, or some 3.6 instead of 3 K.

As we don't know the temperature rise and the curves for TC vs current are only typical with multiple manufacturers for the diode we can not expect the correction to be very accurate. It may take another 1 or 2 iterations. For the next step one can than do a kind of linear interpolation - so it would help to do a very similar test (like the warm up). Interpolation works usually better than extrapolation and one may thus try a bit larger step first (e.g. 3.9 K).

[KungFuJosh]

Would it be better for me to monitor the voltage and temperate at the same time rather than guesstimating?

If I install a low ohms resistor between R359 and VR122, would that be useful to test the supply current to target the datasheet value? Or should I ignore that for some reason?

[Kleinstein]

If one has a good way to also monitor the temperrature, it would help a little, but not absolutely needed. The warm up after power down is usually pretty reproducible. The difference could be between 2 or 3 rounds to get low TC.

[KungFuJosh]

Here's the first 5 minutes with 3.6kΩ for R359. It seems the sawtooth effect is reduced? I noticed the voltage is also lower.

I restarted the log/reset the readout panel for the Second 5 Minutes.

Am I correct to assume that as I approach TC=0, the voltage will get closer to the nominal 6.4V value? Or is that not that accurate?

This was measured with the case open.

Thanks,
Josh

[KungFuJosh]

Same tests with 4.7kΩ.

[KungFuJosh]

Same tests with 6.2kΩ.

[Kleinstein]

The zener voltage has quite some tolerances. Even the initial 6.2 V of the old diode was not so obvious off. So getting closer of further away from the nominal 6.4 V has no meaning.

With less current it looks like the voltage still drifts up, though a little slower. A slight problem for the comparison could be the time zero points - a few more seconds at the start my make quite some difference. So one may have to look more a the later drift like 2 minuted to 10 minutes or so.
It looks that at least the low frequency noise got better with lower current - a good part of the LF noise can be from thermal effects.

The voltage still looks rather noisy - the expected noise should be rather low. The whole point of the extra zener is to have a noise level well better than the LM399. There would still be some very low frequency noise from thermal effects.
So the limiting noise should be to a large part from the LM399 reference in the DMM6500. The meter is not a good choice for the measurement, but I would expect better.
Given that one aims for good noise, at absolutely makes sense to have the TC somewhat trimmed.

p.s.:
The curve is still going up, even with only half the nominal curent. This makes me think that the diode may not be a true 1N4579, but maybe a lower grade or lower current type. The diode could still be OK, but a much lower current would also cause more noise.

[KungFuJosh]

You're right- I should have used my 34465A to test this. It's not ideal anyway, so comparison is what matters...and I should have been monitoring temperature too, but I think that's been pretty stable.

I haven't posted them yet, but I'm also doing 20 minute tests for each, as the stability is more important IMIO than the startup change. I definitely don't have a good zero point for any of these tests.

So far, 4.7k seems to be the winner. I'll posts the 20 minute results soon.

Thanks,
Josh

[KungFuJosh]

Here's the 20 minute tests. 4.7kΩ had the smallest PP.

Testing order was 3.6k -> 4.7k -> 6.2k -> 5.6k -> 4.3k.

[Kleinstein]

If the noise is really from the zener, it would not make much sense to do all the complications with the extra zener.
If the noise is from the DMM6500, this would be surprosingly poor performance. Maybe there is an additional EMI issues with the cables bringing RF to points where it should not be.
To make sense the noise should be really good - better than also the 34465.