Keithley 2010 noise

[martinr33]

So I fixed the display on one of these, put in a shorting plug and fired it up to take 100 readings on the 10V range.

To my surprise, it only varies by one count. The internal stats consistently report a SD of 200nV across 100 readings. That's set to slow, which is 10 NPLC.

This seems like a remarkably good number. I can't get this out of my K2002, which I have never seen do better than 400nV - although it is often double that. I see similar results from other K2002 units here. The K2002 does have double the range (20V not 10)

You can see this kind of stability on a 6.5 digit meter, but this is a whole class  up.

Keithley makes claims that this is a low noise meter, which appear to be true. It is also a lightweight box, feels almost empty compared to the K2001 and K2002.

[TiN]

You are not the first one to report low noise of K2010.
But you indeed know what to do .

Maybe reverse-engineering of 2010 ADC would give some hints as well. 

[Mickle T.]

Wow! It's a nice and silent meter! It's even better than my 6581 with 280 nV rms noise on 10PLC.

[Kleinstein]

With the SD readings on such a short time series one has to be careful. There may be additional noise at lower frequencies.
There is some indication from the noise of the DMM7510 and K2002 (to a lesser extend) that Keithley uses a kind of running average for the internal AZ and this way reduces the higher frequency (e.g. 1-100 Hz) noise at the cost of additional low frequency noise (e.g. 0.01-1 Hz band).
 
A full noise test should do a longer time series (e.g. more than 10 minutes) and preferably show the results not just as a single SD number, but as a Allan deviation plot or a FFT spectrum. 

Slow mode on some meters could be running average - this gives a misleading low SD number. For a noise test digital filtering or averaging mode should be turned off. Once you get close to the limit where only the last digit changes by one from time to time there can also be trouble from quantization that can lead to both too high of too low noise numbers. At least I would expect the internal SD number not to do this mistake.

[saturnin]

Maybe reverse-engineering of 2010 ADC would give some hints as well. 

Actually, K2010 uses the same ADC topology as K2000 (see attached schematics, taken  from bbs.38hot.net). The only difference is a higher grade (faster) integrator in K2010 (AD744) instead of AD711 used in K2000. In K2010, they likely implemented a different low-noise input stage (located under the black plastic shield).

When a measurement of real signals comes into play (like 7V or 10V reference), I really doubt K2010 can compare to 8-1/2 digit DMMs since it uses LM199 reference, which will be the biggest noise contributor.

[dr.diesel]

You can compare your results to mine posted here:

https://www.eevblog.com/forum/metrology/dmm-adc-noise-comparison-testing-project/msg810831/#msg810831

[Andreas]

Hello,

but why did you do only one measurement all 5 seconds  on 1 or 10 NPLC where you could do easily more?
This way the results are worse than they could be.

with best regards

Andreas

[e61_phil]

What resolution do you get from the 2010 in your test setup?

[Andreas]

Hello,

his data looks like 60nV / step resolution (16 values per 1uV in 10V 10NPLC range).

But with 5 seconds (instead of 0.4) per measurement there could have been also be some averaging function active.

with best regards

Andreas

[dr.diesel]

But with 5 seconds (instead of 0.4) per measurement there could have been also be some averaging function active.

No averaging if you refer to my dataset linked above.  That was compiled quite a while ago, so I don't recall the exact reason for 5-sec, probably just to make graphing one step easier.  Instrument setup data is at the beginning of each run.