Keithley 2010 low noise multimeter noisy.

[Jwalling]

Have a Keithley 2010 on my bench that is somewhat noisy. When set to slow update rate, voltage readings seem to fluctuate up and down then back up again over a period of about 10 seconds or so. Happens whether warm or cold.
When reading the following voltages from my EDC 521 DC voltage calibrator:

Set to 100VDC, about 7mV of noise.
Set to 10VDC, about 700uV noise.
Set to 1VDC, about 70uV noise.

So noise scales with range.

It's not the calibrator, My Keithley 2000 reads fine.
The LM199/399 voltage reference is rock steady.

Worst case AC ripple across the bulk capacitors is about 200mV. Don't think that's it...

And of course no schematics.  I think this would be a fairly easy repair with them.

[Kleinstein]

The data so far suggest that the noise is not from the input amplifier (noise from the amplifier would not scale with ranges). The prime candidate would be the DMMs reference, but it could also be the ADC itself.  Noise from the reference would not show up with shorted inputs.

A next step could be a more quantitative noise measurement. If possible record some time series (e.g. 1-5000 points) and calculate something like the Allan variance to get an idea if the noise is more white noise or 1/f noise type. It also makes sense to record the noise with an relatively high external voltage (e.g. 7 V in 10 V range) and with a short. The interesting ranges are usually the lowest (e.g. 100 mV) and the best (e.g. 10 V). Usually just one relatively short PLC setting (e.g. 1 PLC, no extra averaging/filtering) is enough. Usually one uses with Auto zero - the non AZ mode might show some extra LF noise, just like going up and down over 10s of seconds.

[mimmus78]

It can also be one of the switching circuit with some leakage. I repaired 4 keithley by changing one of this analogue switch.

Try to check noise with other functions ... but I guess because you have noise at 10V you will end having noise in every function.

Inviato dal mio Nexus 6P utilizzando Tapatalk

[mimmus78]

It can also be one of the switching circuit with some leakage. I repaired 4 keithley by changing one of this analogue switch.

Try to check noise with other functions ... but I guess because you have noise at 10V you will end having noise in every function.

Inviato dal mio Nexus 6P utilizzando Tapatalk

[Jwalling]

The data so far suggest that the noise is not from the input amplifier (noise from the amplifier would not scale with ranges). The prime candidate would be the DMMs reference, but it could also be the ADC itself.  Noise from the reference would not show up with shorted inputs.

A next step could be a more quantitative noise measurement. If possible record some time series (e.g. 1-5000 points) and calculate something like the Allan variance to get an idea if the noise is more white noise or 1/f noise type.

I don't know how to do that... I'm guessing via GPIB and some software?

It also makes sense to record the noise with an relatively high external voltage (e.g. 7 V in 10 V range) and with a short. The interesting ranges are usually the lowest (e.g. 100 mV) and the best (e.g. 10 V). Usually just one relatively short PLC setting (e.g. 1 PLC, no extra averaging/filtering) is enough. Usually one uses with Auto zero - the non AZ mode might show some extra LF noise, just like going up and down over 10s of seconds.

In my initial post, those measurements were taken with the cover off. Additionally, when I was applying 10V, the calibrator was in the 100V range, which was really dumb of me!

So with the cover on and after a 1 hour warmup:
On the 100mV scale with a direct short I get readings that vary between 5.07uV and 6.24uV, so about 117nV of noise.
On the 10V scale with 7 volts applied from the calibrator, I get readings from 6.999710V to 6.999823V so 113uV noise.

[Kleinstein]

Depending on the integration time used, the noise in the 100 mV range (300 nV_pp and thus likely something in the 50 nV_RMS range) looks relatively good. Also the noise for the 10 V range now looks much better, especially if one has to consider that some of the noise might be from the voltage source. Still the noise is nearly 100 times higher, which suggests that either the 100 mV range is really good, or there is something to the 10 V range. So there might not be a problem with the meter at all.
Noise data only make sense if the voltmeter settings (integration time, filtering, auto zero setting) are known. Usually the RMS (=  standard deviation) values for a reasonable long time series (e.g. 100 points, no digital filtering/running average) is a good fast way to compare.


To record the data, one could use GPIB or maybe the RS232 or LAN interface.
Calculation of the Alan variance curve is than a second step. There is a DMM noise comparison thread that should have links to suitable programs and if needed someone from that thread could likely do the calculations. Alternatively to the Allan variance plot one could also use FFT to get a noise spectrum if needed.

[Jwalling]

Just getting back to this now...

Figured I'd try the same test using a battery pack comprised of 3 AA Alkalines as a source. This would eliminate the calibrator as a noise source.

Measurements ranged from 4.729505V to 4.729633V over a 5 minute period, so 128uV of noise.
I stopped short of covering the whole mess with a grounded tinfoil shield. 
I temporarily swapped in an LM399AH from a HP 34970A. No difference in behavior, which is not surprising as the original seems nice and stable.

I don't think this is easily repaired without schematics. Thanks for you thoughts, though! I appreciate it.

[mimmus78]

This is bad decisions. Never touch a reference, especially with a soldering iron. After all this heat stress it may take years to calm down, and stop drift like it was before. And this is now for both meters.

Inviato dal mio Nexus 6P utilizzando Tapatalk

[Jwalling]

This is bad decisions. Never touch a reference, especially with a soldering iron. After all this heat stress it may take years to calm down, and stop drift like it was before. And this is now for both meters.

Huh. I didn't know that. No big deal on the 34970A DMM board. It has a bad ASIC, 74X errors (corrupted FRAM) and I've been scavenging parts from it for awhile.

Thanks for the heads up, though.

[Kleinstein]

Can you show the curve from measuring the batteries ? Depending on how they are handled the batteries may also show some drift or even jumps. It is not only noise that can change the readings.

Have you checked behavior of the resistance ranges, especially an easy one, like 10 K ?

As defects quite often happen to the input section, it might be a good idea to also measure input leakage. It is relatively easy with just a 1 M resistor of a low leakage about 1 nF capacitor at the input. Chances are not very high to have excess leakage, but the test is relatively easy.

Even without a schematics one can localize some errors quite well, by comparing different ranges and input voltages. However this needs quantitative values and more info (e.g. integration time).  It also makes a difference if there is noise or a drift or jumps between two values. It really help to record the data to the PC.
Many of the meters are rather similar in the basic building blocks. Once the problem is localized to a building block, it might be able to get a diagram from the area in question by reverse engineering. So it is only a small part of the circuit that really needs to be looked at in detail.

[picburner]

A while ago, when I repaired my 2010, I did reverse engineering of part of the circuit but I stopped it when I found the failure.
If it can help, I can give you the schematic I've made.
It is not complete and I can't guarantee it is error free but better than nothing....
The original files are made with Altium but I can also convert them to PDFs.
What is your firmware version?

[Echo88]

For anyone interested: the partial schematics of the Keithley 2010, made by picburner and the K2010 A10-Firmware, are also now available with the following link.
https://doc.xdevs.com/docs/Keithley/2010/2010_reversed_partial_sch_and_fw.zip

Many thanks to user picburner! 

[Kleinstein]

The reversed schematics give a first crude idea. As expected there are a few errors inside, but for most of the circuit one gets a pretty good idea on how it is supposed to work. Much looks like rather similar to the Keithley 2000.

So DC volts starts with a bootstrapped AZ OP as a buffer and than an DG408 mux and amplifier (this time with discrete JFETs ).

The ADC reference seems to be separate 6.2 V zener which must be rather low noise and the LM399 reference only for scale factor adjustment.
The ADC really looks very much like in the K2000 - so the obvious question is what was wrong in the K2000 that essentially the same circuit could perform so much better. It looks like it is not the obvious change (a better OP for U138).

[saturnin]

The main difference between K2010 and K2000 is the low noise DCV input section. Instead of quite noisy LCT1050 (1.6 uVp-p) used in K2000, K2010 features a composite input amplifier based on LT1124 (70 nVp-p). Also 100 mV, 1V, and 10V DC ranges are routed directly to the input amplifier (and not through 9.9MOhm resistor as in K2000).

The output signal from the input amplifier is routed to the gain amplifier which is *very* similar to one used in HP3458A. Based on my findings, the scheme of the gain amplifier in the attached reversed schematics is not 100% correct.