Do I have exceptional Keithley 2000 or poor Keithley 2001?

[mimmus78]

Hi guys,

I'm running some monitoring of my LM399 reference based board. I'm doing this by one Keithley 2000 and one Keithley 2001 multimeter but I have unexpected results.

I was expected the Keithley 2001 to be far less nosier than the Keithley 2000, but this is not the case.

I'm using the same configuration for both meters:

  10 NLPC
  10 averaged readings
  autozero sync
  10 V range for K2000 - 20V for K2001

This K2000 is really stable, never expected it to be as noise free. Results don't change too much even at 10 NPLC without averaging.

Attached readings graph (ignore fist hour and more, meters where still cold) ...

So my question is do I have exceptional Keithley 2000 or do I have to go hunting inside Keithley 2001 for some problem?

Anyway now I think I really need one 3458A ...

[Kleinstein]

Both curves still show quite some noise. So for me it is more like a rather noisy K2001. To check for the noise source one could do a simple test with shorted inputs, to separate in noise of the input-stage / ADC and that of the meter internal reference. You have 3 lm399 references involved: the DUT, and one in the K2000 and K2001 each.

Besides true noise there is also possible RF interference, that might look like noise but can influence the meters to a different degree.

[Andreas]

Hello,

looks plausible to me.
My K2000 has around 7uVpp or around 1uV std deviation when measuring a 7V LTZ1000 reference in 10V range.
When I bought my K2000 the seller had hooked it together with his K2001 to a SMU.
Noise on K2001 was significantly higher.

With best regards

Andreas

Edit: has the averaging an effect when reading out by interface?
My measurements are with 10 NPLC and no averaging.

[mimmus78]

Well I cannot exclude RF noise, but to me it's not the case.

I run also some tests with a battery and got consistent noise figure similar to that of the reference board.

With shorted leads at 20V range I got:

  12uV p2p noise in 1 minute at 1 NPLC (no averaging)
  7uV p2p noise in 1 minute at 10 NPLC (no averaging)

From what I read in the "noise test" thread this is ok for a K2001 ... but maybe I misinterpreted it.

Starting thinking that K2000 is fair beast for 10V measurements.

Just thinking to sell my K2001 to someone that needs all those special features it has and get a more modest and new 34465A with fresh calibration.

[saturnin]

Hi,

I have exactly the same experience. Surprisingly, K2001 is noisier than K2000. It is confirmed by their specifications (see attached screenshots from user manuals):

K2000 10V range: rms noise at 1NPLC is < 4uV (based my experience, it is lower actually), while
K2001 20V range: rms noise at 1NPLC is 6 uV typical (0.3ppm of 20V is 6 uV)

I think it is not so well known fact and I was personally quite disapointed when I found out...

[mimmus78]

has the averaging an effect when reading out by interface?

Yes but not as I expected ... fast test, run for only 10 minutes and a total of 47 readings.

STDEV with AVG = 3.22 E-06
STDEV with NOAVG = 3.40 E-06

seems to little difference.

PS: time was different for the two tests but for the purpose of producing the chart I put only one timeline

[mimmus78]

Surprisingly, K2001 is noisier than K2000. It is confirmed by their specifications

ohh I searched this noise specs in K2001 manual and for some reason I din't catch up ... double disappointment 

[VintageNut]

My K2000 is stable and quiet measuring my 2X Fluke 731B. My K2000 was calibrated by Keithley and the data shows 1ppm deviation from 10V. I have adjusted both of my Fluke 731B to match the 10V of my K2000.

I am also borrowing a DMM7510 and it is much better than the K2000. I can see the 731B drift 1.5 ppm over the course of the day and then the 731B returns to its morning value the next day.

This drift of the 731B is below the resolution of the K2000. As good as the K2000 is, the DMM7510 is much better. The DMM7510 measures a few ppm lower than the K2000 but that is not important to me.

[Andreas]

has the averaging an effect when reading out by interface?

Yes but not as I expected ... fast test, run for only 10 minutes and a total of 47 readings.

STDEV with AVG = 3.22 E-06
STDEV with NOAVG = 3.40 E-06

seems to little difference.

PS: time was different for the two tests but for the purpose of producing the chart I put only one timeline

Ok that is no difference. (only statistical stray).
With averaging 10 readings the STDDEV should go down by a factor of 3.
I guess that the averaging only works for the display on the instrument.

With best regards

Andreas

[martinr33]

I have a 2001, 2010, 2002 sitting on my bench right now.

I measured a LTZ1000 unbuffered reference, 100 readings, NPLC 10, digital filter. Noise is the SD reading, only maybe 10 mins warmup so there might
be some drift.

2001:   5uV    noise    7th digit
2010:   1.5uV noise    7th digit
2002:   1uV    noise    7th and 8th digits
 
the 2010 is a bit of a surprise. It is a stripped down machine with a segment display.

Also - on the 2000, 1uV of noise on the 10V range is below the sixth digit, which represents 10uV increments. So maybe  there's a bit of quantizing going on, chopping off some of the noise.
 
I see the same kind of thing with my 6.5 digit 34401A.

[mimmus78]

I 2010 is very similar to 2000, I think it has new firmware ad improved better selected references.
Anyway stdev of my K2000 seems way better than 1uV ... that's nice.

This things that AVG is not working is strange ... meter really slow down when enabling averaging.
Maybe we should read better the manual and read the averaged vale with some other command.

[mimmus78]

Well I used "READ?" to read measurement.
This command even if slow down measurements according to AVG setting just returns last measurement and makes no AVG.

[TiN]

Time for some bit of RTFM
2001/2002 has two modes for averaging, moving and repetitive.
Default is moving, with amount of samples = 10.

Also READ? commands resets triggering and math systems. FETCH? with proper sample window setting would be better solution, and maybe provide less noise (I did not test this myself though).

I can try it in next few days, if you guys want. 

[mimmus78]

Anyway I don't understand why "READ?" slow down with AVG enable if it just return one reading.

I tried half a dozen solutions but I never managed to make it working, some examples just uses GPIB
functions my library seems not to implement, and understanding all the details to avoid those special
commands and settings need some more RTFM.

Last attempt was to read 10 or 100 samples and do AVG by myself, but neither this worked and got
only an empty reply.

TiN if you have some "spare" time to check this it will be helpful, if not I will try it on next weekend
and let you all know :-)

========================================================================

   # Setup Keithley 2001
   l.ibwrt(ud, "*RST")
   l.ibwrt(ud, ":SYST:AZER:TYPE SYNC")
   l.ibwrt(ud, ":SYST:LSYN:STAT ON")
   l.ibwrt(ud, ":SENS:FUNC 'VOLT:DC'")
   l.ibwrt(ud, ":SENS:VOLT:DC:DIG 9; NPLC 1; AVER:COUN 1; TCON REP")
   l.ibwrt(ud, ":SENS:VOLT:DC:AVER:STAT OFF")
        l.ibwrt(ud, ":DISP:WIND:TEXT:DATA \" \";STAT ON;")
   l.ibwrt(ud, ":DISP:WIND2:TEXT:DATA \" \";STAT ON;")

   l.ibwrt(ud, ":FORM:DATA ASCII")
   l.ibwrt(ud, ":TRAC:POIN 10")
   l.ibwrt(ud, ":TRIG:COUNT 1")
   l.ibwrt(ud, ":TRIG:DELAY 0.000")
   l.ibwrt(ud, ":TRIG:SOURCE EXT")
        l.ibwrt(ud, ":SAMPLE:COUNT 10")

        l.ibwrt(ud, ":TRACE:CLEAR")
        l.ibwrt(ud, ":INIT")
   #time.sleep(20)
   l.ibwrt(ud, ":TRAC:DATA?")

[martinr33]

Turns out that the 2001 has a couple of settings in the setup menu.

Autozero - should set to synchronous. Autozeros every reading.
Power line sync - should also be synchronous for high precision.

Once I set these two parameters, I lost triggering - the K2001 defaults to high speed measurements, not high precision. Now, the K2001 is down at 1.57uV SD, measuring a 7.11V

https://goo.gl/photos/X81xs1nazapSfSMN8

Not sure how this will translate into GPIB commands. But it seems like this is static setup.

My 6.5 digit HP 34401A shows 7.11374, with the occasional flicker to 5. But I tweaked it to be closer to the 2001. The 2002 is notionally in cal.

[TiN]

Stacking 200x's like that is enough to get 2002 out of cal though. .
I have day off tomorrow, so will run the test. I'm done with fairy-tale LTZ module, so can use it as reference.

Here's initial reference data as base. Reference is in DIY TEC thermal chamber, so there is no TC other than DMMs.

RAW DSV data. First hour or so had temperature setting change, that part removed from graph below.



Here's code used for sampling on 2002 (2001 is exactly same):

:SYST:AZER:TYPE SYNC;
:SYST:LSYN:STAT ON;
:SENSE:FUNC 'VOLT:DC';
:SENS:VOLT:DC:NPLC 10;
:SENS:VOLT:DC:DIG 9;
:SENS:VOLT:DC:AVER:STAT OFF;
:TRIG:SEQ:SOUR TIM;
:TRIG:SEQ:DEL 0.01;
:READ?

No need for any special commands, just writes and reads.

[mimmus78]

Turns out that the 2001 has a couple of settings in the setup menu.

Autozero - should set to synchronous. Autozeros every reading.
Power line sync - should also be synchronous for high precision.
Once I set these two parameters, I lost triggering - the K2001 defaults to high speed measurements, not high precision. Now, the K2001 is down at 1.57uV SD, measuring a 7.11V
Not sure how this will translate into GPIB commands. But it seems like this is static setup.

Meaning                            GPIB
=========================================
Autozero synchronous =  ":SYST:AZER:TYPE SYNC"
Power line sync =             ":SYST:LSYN:STAT ON"
High precision =                ":SENS:VOLT:DC:NPLC 10"

I used "NPLC 1" to get more ADC noise and hopefully accentuated averaging effect ...

[mimmus78]

>> Stacking 200x's like that is enough to get 2002 out of cal though. .
Unless you use some audiophile grade oxygen free copper OFC super compensated leads and some black insulation box for power supply. 

Calculated stddev to almost 1uV@10 NPLC for the 3458A, like in spec ... 10X better than 2000 and way better than 2001.

[doktor pyta]

I also have problems with K2001. With 1NPLC and 100 AVERs it has much higher noise than HP34401A set for 100NPLC.
Have anyone tried to make upgrade similar to TiN (https://xdevs.com/article/k2002_nadc/). Does it make sense in case of K2001.
PS. Are there official schematics of K2002 available?

[lukier]

I also have problems with K2001. With 1NPLC and 100 AVERs it has much higher noise than HP34401A set for 100NPLC.
Have anyone tried to make upgrade similar to TiN (https://xdevs.com/article/k2002_nadc/). Does it make sense in case of K2001.

That would be interesting. I remember from the K2001 thread that TiN once tried replacing LM399 with LTZ1000, but it made things worse 

PS. Are there official schematics of K2002 available?

In general I don't think so, but TiN did some rev-eng of the ADC board AFAIR.

[saturnin]

I was also thinking of replacing LM399 with LTZ1000 in K2001. Unfortunately, you simply can't desolder LM399 and connect +7V ref from LTZ1000. It is because 2mA current is fed to LM399 from -10V ref source located in the A/D board. Moreover, the negative reference terminal of LM399 is connected to a virtual ground created by U328 (AD707).

A possible solution is shown in attached drawing. It is not a complete design - it rather shows necessary modifications in the original reference circuitry to support LTZ1000. I ceased to implement it when I realized that main contributor to K2001's noise performance is not the voltage reference, but rather the A/D converter...

[mimmus78]

Yes K2000 is 2 or 3 times less noiser than K2001 ... maybe K2000 can be a nice project, but I don't think you will get noticeable improvement on noise.
You would improve the 10V accuracy but the other ranges will continue to be as good as a 6.5 digits multimeter.

[Zucca]

I was also thinking of replacing LM399 with LTZ1000 in K2001.

Wow! I got the same idea. TiN inspired me with by doing that in a HP 3245A (see article here). It this case things are much better than in a K2001, see below how the LM399 is used there:



In a K2001 the LM399 has no reference to ground...



shall we discuss this in a new thread? Oh let´s see... I think some Voltnut Dragon will destroy our idea pretty soon. It THEY didn`t manage the upgrade yet, there will be a good reason behind.

[Kleinstein]

Before thinking about an update of the reference one should check where the noise is really coming from.
There are 3 main sources for noise:
1) the input amplifier
2) the ADC circuit itself (might depend on voltage and integration time)
3) the reference

So the first step would be a careful characterization to see where to start from. Using different integration times also helps to separate sources, as there can be 1/f type noise.
Measuring noise at short circuit takes out the reference for most of the part. Low ranges are often mainly the input amplifier. Knowing the circuit one can usually get a good estimate on how much noise in the 20 V range comes from the amplifier.

Measuring noise with a low noise voltage adds the LF noise of the reference.

p.s. I just looked at the available diagram for the 2001: it looks like they have two references inside: The LM399 for long time stability (and ohms source) and a second one (VR801 = 6.4 V zener) as a working reference for the ADC, that is one the ADC board itself.

[Zucca]

Interesting...

VR801 = 6.4 V zener

The popular 1N4565 thru 1N4584A-1 series of Zero-TC Reference Diodes
provides a selection of both 6.4 V nominal voltages and temperature
coefficients to as low as 0.0005%/oC for minimal voltage change with
temperature. Four different operating currents are available for selection at
0.5 mA, 1.0 mA, 2.00 mA, and 4.00 mA. These glass axial-leaded DO-35
reference diodes are optionally available with an internal-metallurgical-bond
by adding a “-1” suffix. This same “-1” bonded Zener package construction
is also available in JAN, JANTX, and JANTXV military qualifications.
Microsemi also offers numerous other Zener Reference Diode products for a
variety of other voltages up to 200 V.

http://www.datasheetarchive.com/dlmain/2645466039e35e26f019f39d2feadc007b3f14/M/1N4571A

EDIT:

So did some homework... this is the VREF generation circuit on the ADC K2001:



On R845 there is the voltage produced by VR801, let´s call it VD... the current flowing in R845 must go through R843 and R847... so it means

Vref+=VD/2K*1.2K+VD=1.6*VD
Vref-=-VD/2K*3.2K=-1.6*VD

So Vref(VR801) is about 10.24V.... In theory we can replace the VR801(VD=6.4V) with a LTZ1000 (let´s say 7.15V). And by changing R843 with 900R and R847 with a 2,9KR we should get a new Vref (LTZ1000) at about 10.36V, rock stable.

Ah didn´t check any DC current, heater stuff yet... just want to point out the mod looks possible here...

Just don´t know how much this mod will impact the performance...