Keithley 2010 Repair

[KungFuJosh]

Keep in mind the case is also open in those tests. I'll do a test with a fresh 4.7k at R359, case closed, 34465A for the meter (which is usually more stable for short tests compared to my other meters).

Besides that, stable warm up time is like 2 hours with the case closed. I expect it to look a little shitty in the first 20 minutes.

[KungFuJosh]

I'm monitoring the test with the case closed using the 34465A. It's less noisy, but also clear that the tests would have taken 10 times longer with the case closed. Either that or they were a waste of time lol.

Once the temperature & voltage get more stable, I'll reset the readout and see what it looks like.

Thanks,
Josh

ETA: It's probably also worth noting that I'm using cheapo generic SMT resistors from my 805 book in place of a higher quality 1206 resistor. I would guess that will also effect VR122's stability, especially with the case closed.

[KungFuJosh]

Here it is with the case closed (not screwed in) after a 2-hour warmup.

This is the best PP of the resistors I tried. However, the measurement stability/PP is 2.5 times worse than before making the change. I assume that's due to the replacement resistor not being good enough quality/TC.

Calibration also needs to be adjusted after that change, but I also assume that shouldn't affect stability, only accuracy.

[KungFuJosh]

I reinstalled the stock parts for VR122 and R359. Tested this time with case closed only. The xlsx file has a true zero starting point.

This is a completely different story. Is that replacement 1N4579A defective or fake or what?

Thanks,
Josh

[KungFuJosh]

I think at this point it's pushing the limits of the 34465A...but I'm not patient enough to wait longer. I'm putting it back together now.

[Kleinstein]

As the voltage was still drifting up, even with a much lower than normal current, the new Zener is not meeting the 1N4579 specs. With a real 1N4579 one should have seen a change in sign of the drift already with 3.9 K and already the 3.6 K test would be boarderline.

The old Zener is drifting down. One might consider  using a slightly small resistor than 3 K for R359 to get less drift. This could also be a 2nd resistor in parallel. A guess for the TC is from the first 5 minutes: this would be some 50 ppm change in voltage with maybe some 2 or 5 K of temperature rise. This would suggest that some 5% higher current could be about right.

It is normal that the Zener noise is pushing the 34465 to it's limit. The extra zener only makes sense if it is really low noise. The 6.4 V range compensated zeners like 1N4579 / 1N829 are really low noise, if operated to a current for near zero TC. They are just not super low drift and not all units are the same. Measuring them directly to see the noise would even push a 3458 to it's limits. For checking the TC/drift the 34465 and likely even the DMM6510 are OK.

[zepto]

It's possible Keithley did some selection of those parts since they weren't very expensive and their performance is critical.

With it out you should be able to find the minimum TC of the diode itself using a precision current source and a source of heat.
The TC will be negative at say 80% of nominal current and positive at 120% and somewhere between will be a minimum. The TC minimum will be much closer to nominal if that is the highest grade of that family.

You do have to be careful not to heat one side/lead of diode more than the other.

[KungFuJosh]

It sounds like it might be worth a small adjustment to R359 to see if I can get the stock VR122 slightly better.

Is it reasonable that I want this meter to be more stable in the 10V range compared to my 6.5 digit meters?

The datasheet claims 1.2uV noise in the 10V range (2min), but that's not what I'm getting...yet. See attached (tests performed with over 12 hours of warmup). The ref is a battery powered VREF10-001 r9 (which appears to be more stable with the batteries vs. external power brick).

The other meters' datasheets don't give a noise spec in such clear language. Bastards.

Thanks,
Josh

[iMo]

.. with a single opamp (I did with OPA189 or ADA4523) with set A=10000, BW=10Hz (0.1 to 10Hz) made of simple RC, you will get the noise of the zeners into XXmVpp region, easily measurable with any scope or DMM. Has to be put into a metal cookie box, powered by batteries, coax output.

[Kleinstein]

It is not absolute clear how much effect the zener TC has on the noise. The big unknown is how stable the temperature around the zener actually is on real world use (warmed up and with closed case). It could still make sense to do 1 or 2 tries for a lower TC at a slightly higher current.

The problem with most keithley meters (including the K2010 and DMM6510) is that the noise in AZ mode is not white noise. So just an RMS noise claim alone is not worth much. The actual noise depends on the time window / averaging time. To judge the performance of the meter one would want the allan deviation curve, e.g. for 10 PLC converion mode. This would be both with a shorted input (still the 10 V range) and a really low noise reference. Not sure of the VREF10 would be good enough for this.
The shorted input in the 10 V range would mainly test ADC noise.
The 7 or 10 V ref. test would test the external reference and ADC internal reference part.
A shorted input in the 100 mV range would mainly test the amplifier (and related) noise.

p.s.
Measuring the reference noise with an AC coupled amplifier can be an option. It still needs a bit of care with the OP-amp and filter capacitor. Some OP-amps have quite some current noise and this would require a large capacitor to also get low frequencies like 0.1 Hz. Electrolytic capacitors can cause issues with settling and leakage current.

[KungFuJosh]

I'm going to order some better quality 1206 thin film resistors to try adjusting the current. I don't want to use my cheapo 805 resistors, as I'm sure that will introduce their own instability.

Beyond that, my primary concern is seeing how the 10V test compares with other meters using the VREF10. I don't have anything better to test with.

After the long warmup, it is now a tiny bit better than the DAQ6510. That's something. What bugs me is that both the 34465A and the SDM3065X appear to be performing better than the Keithley meters. I suspect that the SDM3065X has some type of averaging applied that can't be adjusted, despite all the meters being set the same (in theory) with no controllable filter applied.

Anyway, I made a thread to discuss the stability comparisons: https://www.eevblog.com/forum/testgear/bench-dmm-stability-comparisons-with-testcontroller/

And I'll focus more here on the repair/adjustment of the K2010.

Thanks,
Josh

[Kleinstein]

The resistor quality for the Zener current is not that critical. The zener part only needs to be short time (like 15 min maybe 30 minuted - don't know the details) stable. So no worry about long term drift or humidty effects. The TC is also not that critical - the Zener is way lower impedance and resitor TC would be attenuated something like 100 fold. So chances are the choice of value would be more relevant then the resistor TC.
So no need to overdo it with the resistor - 25 ppm/K grade should be well good enough.

The meters in the test are working quite a bit different:
The SDM3065 has a SD ADC chip. This may give a tiny bit of extra filtering / soft start and end of the aperture. However this would only be more minor (more in the ms range). I don't know if the meter relies on the resistor ratio or does an in between gain measurement like the K2010.
The KS34465 has a contineous time SD ADC build with parts like a MS ADC. AFAIK the ADC gain is not corrected in real time, but only by explicte ACAL calls.
The K2010 uses a seprate reference for the ADC from the long term stable LM399 reference. Besides the actual measurements the meter spends some time on reading the LM399 and link the zener/ ADC to the LM399 ref. This acts as some digital low pass filtering of the LM399 reference, but takes up some time (like 1/3 the time). However it also adds some noise and the implementation of the AZ step is poor: they seem to use some averaging for the zero reading and this causes extra low frequency noise down to some 0.02 Hz to effect the result.
For the dmm6510 I know less details, but it has a similar extra LF noise and thus likely also does zero averaging. I don't think the DMM6510 uses the extra ADC reference, but may still correct for ADC gain drift in the background. This helps with very long time drift and TC, but adds short to medium time noise.

[KungFuJosh]

Unfortunately, replacing the stock VR122 brought back fail 101.2 (no 7V at A/D), and now there is also 305.2 (bias current).

I assume/hope that adjusting the current as you suggested might cure both.

In case that doesn't cure it, replacing VR122 with a higher voltage diode did remove error 101.2. Is there a high-quality replacement TC=0 zener you could suggest that would be similar or better than a real 1N4579A? Would 1N4578A-1 be good enough, or is there a better choice?

Thanks,
Josh

[Kleinstein]

The higher current would increase the voltage, but only a tiny bit. So chances are small that it would be enough.

The 1N4578 may work, but there is a larger tolerance in the current to get zero TC. For the 1N4579 this should be within some 15% of the nominal 2 mA, with the 4578 this may be 30%. A somewhat higher current should be OK, much lower would come with more noise.
Mouser / Digikey still also have the 1N4582 - that is the 4 mA nominal version with even more tolerance (~1.5 - 10 mA for zero TC). It may work, but could need rather high current.
Both are a slight gamble, like maybe a 50% chance. Getting more from the same may not help much if they come from the same batch.

One could take an even larger gamble with a 1N828: nominally higher current (would be OK if zero TC on the low side) and a large tolerance in the voltage (6.2 - 6.9 V, nominally 6.55 V).

[KungFuJosh]

Comparing +/- %/oC from the datasheet:
1N4579A is 0.0005
1N4578A is 0.001
1N4582A is 0.002

I'll get 2 or 3 of the 1N4578A, and I'll get a 1N4582A as a backup plan.

What do you think about using a decade resistor box, or a 5k potentiometer to dial in the current in circuit?

ETA: Also, what do you think about using heat shrink tubing on the diode? Joe mentioned insulating his diode for better stability.

Thanks,
Josh

[zepto]

With the original zener back in place the meter is stable and within a couple ppm of your other multimeters? You believe there is some significant issue with AD reference voltage?

Besides a self-test that is not officially documented, is there any evidence the meter is not performing correctly?

[Kleinstein]

I don't think heat shrink at the diode would make much difference. Most of the heat conduction would be via the wires and the PCB.
The idea of the TC adjustment that small temperature fluctuation would no longer matter that much.

I would see the TC specs in the datasheet as tolerances on how close the nominal current is to the zero TC current. So with the 1N4579 this would be someting like 2 mA +-15%, for the 1N4578 is should be 2 mA +-30% and for the 1N4582 it would be 4 mA +-60% ( maybe more ?).
The point with the current trim would be to aim for something like +-5-10 ppm/K or 5-10 x better than the maximum 1N4579 specs or some +-2% for the current away from ZC zero crossing.

There would be one more option: replace the LM399 with an ADR1399 and use the ARD1399 also for the ADC. With a lower noise long term reference there is less gained from the extra ADC reference. The ADR1399 would be a bit more noisy than the 1N4579 on the short term, but having the same reference also for the ADC also takes away some noise in the gain measurements (reading the 7 V reference). So there should be less noise for the gain adjustment steps (e.g. every 100 s or so).

One could give it a first try with the LM399 to see if the DMM accepts a 7 V ref. for the ADC. The hardware side should be OK with this as this is how the K2000 works with the same TF245 resistor network. The question is where the self test limits are: 6.2 V seems to be just a bit low and 7 V may give an error too.
One would bridge the LM399 ref. to where the zener was.

If one has a decade resistor box one could use it to dial in on the zero TC current. A pot may not be that good as one can not really read the value. The TC measurement of some kind is not super fast and an easy adjust thus not that important. Knowing what vaule was tested could be more helpful.

[KungFuJosh]

With the original zener back in place the meter is stable and within a couple ppm of your other multimeters? You believe there is some significant issue with AD reference voltage?

Besides a self-test that is not officially documented, is there any evidence the meter is not performing correctly?

Are you suggesting that a $6,130 7.5-Digit professional level bench meter being marketed as "high stability" is acceptable to perform with 55% worse stability than a Siglent 6.5-digit meter that costs over $5000 less?

Are you also suggesting that those "undocumented" tests that accurately describe the issues the meter has, are not evidence in and of themselves?

[KungFuJosh]

I don't think heat shrink at the diode would make much difference. Most of the heat conduction would be via the wires and the PCB.
The idea of the TC adjustment that small temperature fluctuation would no longer matter that much.

I would see the TC specs in the datasheet as tolerances on how close the nominal current is to the zero TC current. So with the 1N4579 this would be someting like 2 mA +-15%, for the 1N4578 is should be 2 mA +-30% and for the 1N4582 it would be 4 mA +-60% ( maybe more ?).
The point with the current trim would be to aim for something like +-5-10 ppm/K or 5-10 x better than the maximum 1N4579 specs or some +-2% for the current away from ZC zero crossing.

There would be one more option: replace the LM399 with an ADR1399 and use the ARD1399 also for the ADC. With a lower noise long term reference there is less gained from the extra ADC reference. The ADR1399 would be a bit more noisy than the 1N4579 on the short term, but having the same reference also for the ADC also takes away some noise in the gain measurements (reading the 7 V reference). So there should be less noise for the gain adjustment steps (e.g. every 100 s or so).

I already bought an ADR1399, but I wanted to get the meter stability in spec without errors before swapping out the refs to do a before/after comparison to see how valuable the reference change really is in this circuit.

I would guess that installing the ADR1399 might solve the 101.2 error anyway, but I'd prefer to fix VR122 first.

One could give it a first try with the LM399 to see if the DMM accepts a 7 V ref. for the ADC. The hardware side should be OK with this as this is how the K2000 works with the same TF245 resistor network. The question is where the self test limits are: 6.2 V seems to be just a bit low and 7 V may give an error too.
One would bridge the LM399 ref. to where the zener was.

If one has a decade resistor box one could use it to dial in on the zero TC current. A pot may not be that good as one can not really read the value. The TC measurement of some kind is not super fast and an easy adjust thus not that important. Knowing what vaule was tested could be more helpful.

I might get a decade box, we'll see what happens with that.

A precision potentiometer (preferably multi-spin screw type like Vishay T93YA502JT20) could dial in the resistance and then be measured when removed from circuit. That should be close enough to find an available resistance value.

Ideally I would want to connect it with leads long enough to reach outside the meter, so it can be adjusted without needing to disassemble the meter. What do you think of that?

Thanks,
Josh

[Kleinstein]

With the original zener back in place the meter is stable and within a couple ppm of your other multimeters? You believe there is some significant issue with AD reference voltage?

Besides a self-test that is not officially documented, is there any evidence the meter is not performing correctly?

Are you suggesting that a $6,130 7.5-Digit professional level bench meter being marketed as "high stability" is acceptable to perform with 55% worse stability than a Siglent 6.5-digit meter that costs over $5000 less?

Are you also suggesting that those "undocumented" tests that accurately describe the issues the meter has, are not evidence in and of themselves?

 

Like many Keithley meters the 2010 has issues with low frequency noise. This is unfortunate, especially as there should be a relatively simple (for those who have the source code) fix in the firmware to get quite some improvement. This effects just such stability measurements or DC measurements with a time scale of some 10 seconds to 1 hour or so. With there unfortunate way of averaging the AZ data they can get a small (like 10%) improvement on the short time (like < 0.5 s) noise, but it could be a huge penalty (like 5-10x) the noise at longer avearging times (like 10s - 10 min). With a simpler AZ mode without fancy averaging the Allan deviation curve should continuing down like square root time until thermal fluctuations from the input take over at maybe some 5-30 minutes. I don't have much hope they would fix it for the 2010, but maybe for the DMM7510 and DMM6510.

So yes for some time scales the K2010 could be less stable / more noisy than something like SDM3065 or HP34401, maybe even a Hantek 3065.

A strength for the K2010 could be with the 100 mV range and current measurements - it still depends on how much 1/f noise from the amplifier come through.
Another point can be the long term ( days and months) stability - here the measurement of the ADC gain in the background can help and can allow the relatively good accuracy specs (24+4 ppm)  for the 10 V range. It is doing some kind of ACAL for the 10 V range in the background. With a well aged LM399 the meter long term stability could be even quite a bit better.

The current pricing is more for a legacy product some still to replace an old one. At this price it is hardly competitive for new users. There are now better performing meter like the DMM7510 and KS34470 available.

Having the current setting resistor external for the tests could be a good idea. A pot and maybe the decade would want some 1 or 2 K series resistor from protection against accidently too low a setting.

[KungFuJosh]

Great call on the series resistor! I'll start with the pot at 3k, and I'd like to think I won't overturn it...but safer is better.

The stability is purely speculative without other 2010 owners sharing similar test data. I could be wrong to assume it should be better, but I doubt it. As you said, the current needs to be adjusted to improve the stability. And the fails don't help either. We'll see how it does when it's corrected.

Thanks,
Josh

[GigaJoe]

a new ADR1399 would be magnitude less stability to original 399 that selected and has many years of active hours.

[KungFuJosh]

a new ADR1399 would be magnitude less stability to original 399 that selected and has many years of active hours.

Initially that's likely (assuming the LM399 isn't the cause of the instability/fails).

It's a fanless meter, so I won't mind leaving it powered on 24/7 to age the new reference.

[Kleinstein]

The aging of the reference is a good point. The LM399 is normally reliable and rarely causing instability, besides the know popcorn noise.
Very likely from the software side (poor averaging in AZ cycle), the meter includes quite some low frequency (1-0.01 Hz) noise despite of the AZ mode that would ideally suppress it. It would still make some sense to fix the zener issue, to start with less of the LF noise from the reference side.
Unless one somehow hacks / fixes the software side the meter will stay somewhat noisy for just the time frame that often matter.
However this would be a whole new can of worms.

If the kind of digital reference filtering part works one may not see that much improvement from a lower noise long term reference like the ADR1399.
The shorter time (like up to a few minutes) noise is not directly from the LM399, but from the zener and ADC.

With the old LM399 one would at least keep the good long time (over weeks and years) stability. It may take years for the ADR1399 to get that stable.
There is a good chance the old LM399 reference is screened and better than average. After all this was a kind of premium meter at it's time.

There is at least a measurement of the stability with a shorted input in the long noise comparison thread.
https://www.eevblog.com/forum/metrology/dmm-adc-noise-comparison-testing-project/msg1366061/#msg1366061

[Cyclotron]

I don't know if its appropriate to start a new thread. The K2010 I have was claimed to have faults with reading DCV when I bought it, seemed fine during tests but I had pushed the front/rear button a couple times.

I've now put new caps in it and the new LED display cleaned the parts and boards.  I also upgraded the firmware from 08 to 15. 08 may have been the original? It's old.

Now I get errors when I run a self-test, but they are inconsistent and seem to be related to my messing with the front/rear switch.

The errors I'm getting most consistently are 305.2, 600.1, and 600.2.
Sometimes I get others like a 400.2 once.  But it is getting fewer as I mess with it and time goes by.  I'm guessing that the front/rear switch has some corrosion.
I thought Josh said you could push a button and it would display some text for these errors, but I can't find that post again.