Keithley 2002 Repair

[martinr33]

I have a Keithley 2002 which, although it completely passes self test, is noisy on the 200mV, 2V and 200V ranges (amongst others) By noisy, I mean the reading varies over a fews seconds by hundreds of counts.

I like to characterize noise using 100 samples and the calculated SD. Not as good as TiN's charts, but a quick way of checking performance.
 
On the 20V range, I get about 0.3uV SD, which seems quite good. This is about 6 - 8 counts movement in the least significant digit. On a 2001, the same methodology yields about 3uV, consistent with 7.5 digits.

From this performance, I assume that the front end and ADC system is largely OK. Neither of these subsystems worries about the range. Gain and switching is handled by  those little MOSFET switch chips that are scattered throughout the system.

It seems likely that the problem is in one of these switches, so my first job is finding the likely candidate and seeing how the voltage behaves across it.
These little FET switches seem to be common points of failure.

As switching one of these out is going to blow my calibration, I am also thinking of upgrading the signal path a bit, following TiN's lead. Over time, many of the components have improved. For example, the DG444 now has a plus version, which claims improved balance and lower  leakage.

So, a few questions.

        - Are there more op amp upgrades to be done on the A/D board?
        - How noisy is the analog section, anyway?
        - Is it worth upgrading other parts in the analog section, specifically the integrated switches but maybe also the front end amps.
        - Anybody have the newer ROM images - seems that there's a B02 now.

My goal is to get the thing stable across all ranges, and then send it off for cal.

[Kleinstein]

Formally opening the case is invalidating the calibration. However even replacing some of the switching MOSFETs is not expected to give a significant change in the readings - so the old calibration factor can still be valid. If at all leakage currents change a little and they tend to give offsets, that are corrected by a kind of auto-zero anyway.
Before sending it to calibration one should use / test it for a while, just after a repair there is a increased chance for failure.

With an older Keithley meter it might be worth a look at the supply filter caps.

The first thing would be to really measure the noise in different ranges to see where the noise is too high. It might be worth to take a little more than 100 samples to calculate the SD. It might even help to have a look at the Allan deviation, which takes a rather long time series. The 20 V readings with shorted input are mainly due to the ADC itself. The 200 mV range noise on the other side is to a large part due to noise of the input amplifier.
If the switching FETs are suspected, one should also check the input bias current.

Many of the switches are not that critical and thus not worth an update, even if better versions are available. The art in building an high end DMM is having a circuit that does not depend so much on many parts. There are not that many parts that really determine the performance.

There is the DMM noise comparison thread, that shows a lot on the noise of the KE2002 and also the 2001. Even if the resolution is 8.5 digits, the noise performance is not 100 times better than typical 6.5 digit meters. The 2001 is rather high noise - its strength is more in good stability / linearity.

[TiN]

I can run specific tests (e.g. 2002 vs LTZ) if you like on my 2002's. They are constantly on, and will be ready to do some sampling since tomorrow as VK5RC's LTZ TC finishing tonight. Overall my tests on DCV 20V range show noise about same as 3458A, same on both my units (one is manufactured in 2007, no parts replaced but zero opamp on ADC), second is repaired (everything on mains earth-referenced PSU is replaced, all electrolytic caps replaced), manufactured in 1992. I have A10 on newer and A05 FW on older unit. Saw Bxx on ebay photos, but have no firmware.

[martinr33]

That could be useful. I am partial to SD as a basic measure - 100 readings  / 10 NPLCAlthough 1000 readings at 1 NPLC might be a bit more insightful.

To Kleinstein's point, this approach won't catch drift, but it does show noise.

I'd appreciate some SD readings, as they give me some quick ideas for troubleshooting.

This unit is about 8 years old. Local cal quote is $240, same as a K2001. They will do a 2010 for $70, which seems like a deal. But I need to fix the front end drift - I would guess, it took a high power hit on a low voltage range. So I need to look at shared pathways.

Also, I think that the 20V pathway goes through the amp at unity gain, rather than a bypass. Which would further cast suspicion on the switches.

[ap]

Just a heads-up, often when they offer cal, they mean cal only, no adjustment, which is usually a factor 2 or so. So you may wish to recheck.

[Le_Bassiste]

martin,
not wanting to hijack your thread: can you please have a look at this one https://www.eevblog.com/forum/metrology/kei-2002-high-offset-at-cold-start/ and tell me how yours does behave?
as for noise figures, this is what i got so far from my 2002, after about 48 hrs. up-time. internal temp-reading was 2 degrees above cal-temp. this is important to note, as i found that a fully warmed-up instrument has some 10% increase in  noise.
all readings are SDEV on short-circuit, at 10 NPLC, no FILT, 50 samples
200mV: 90 nV  (0.45ppm)
       2V: 116 nV ( 0.058 ppm)
     20V: 580 nV (0.029 ppm)
   200V: 13.8µV (0.069 ppm)
     1kV: 30 µV (0.03 ppm)

my K2001 is slightly better than the 2002 at 200mV range (SDEV 0.375 ppm), but almost double as bad on all other ranges.

[nikonoid]

Martin,

I am also in the process of fixing my K2002 and will be shopping around for calibration. Can you please tell us who gave you the quote for $240 and if this is a calibration with data. Thanks a lot.


PS. I have used 4gte.com before to calibrate K2000 meter. They did a great job and adjusted the meter where it was out of spec. Tech guy actually called me and explained what was out of spec. I also got detailed calibration report with data after adjustment.