DMM Noise comparison testing project

[Jorn]

Now continuing with measurements on my own reference. A ltz1000 reference with vishay VSMP resistors (once available from Digikey, in single pieces and custom values) and a LTC1150/LT1010 buffer. It has been running 24/7 for almost 2 years now.

A 2-year old 34470A is used to measure the output of the voltage reference. To minimize temperature variations the 34470A was placed in a cooling chamber (actually a cheap wine cooler) at 18 degC  which is able to hold the temperature inside the dmm within one third degree. (Within 24 hours the temperature varied from 20.28 to 20.59 degC, perfect!).



The reference and the cooling chamber was placed in a 15m2 room where the radiator was set to 22 deg. This ensures relative stable temperature between 21 and 23 degC during the hole day (outdoor temperature was between -6 and 2 degrees). The reference has a TC of 0.05 ppm/degC (meassured in the range from -1 to 25 degC). Expected voltage variations from temperature is now limited to

LTZ1000 reference < 0.1 ppm 
34470A dmm < 0.3 ppm

After 6 hours to stabilize a 20 hour dataset (18000 samples) was collected and the overlapping allan deviation was calculated. Result is shown here:



Conclusion:
Best stability is achieved with five 100 NPLC measuremens (= 20 sec average) unless you have time to wait an hour or more. 

However beware that the uncertainties for high tau values is very high. Look what happens when we extend the period to 87 hours of measurements:



The first 20 hour, for some reason, was remarkably stable.

Attached is the raw data file if you want to play with Overlapping Allan Deviation plots yourself

-Jorn

 

[maxwell3e10]

Its interesting to note that under these (admittedly more controlled than typical) conditions, the performance of 34470A meter is on par with HP3458 and K2002 meters in being able to maintain better than 2x10^(-7) V  stability for 1000 sec.

[Kleinstein]

Comparing the stability curves at 0 voltage and with an voltage close to full scale, it is obvious that that with many of the meters the reference noise sets the limit for the full scale measurements. The references in the 34470 / 3458 and Ke2002 are rather similar (LTZ1000 based) - so not a surprise.

The measurements are also a combination of the internal and external reference of similar quality. So a lot of noise could be due to the external reference too. Not all of the Ltz chips are equally good.

[maxwell3e10]

Has anyone tried to measure directly the difference in voltage between two LTZ1000 references? This would be a good way to check their intrinsic noise and stability without the effects of all the signal conditioning circuits present in the DMMs

[enut11]

Here is an XLS file of my HP3456A with input shorted, 10NPLC, 10v range.

After a short time the readings settled down to 0.00000v and stayed there??

Cannot believe that there is no noise??
enut11

[Kleinstein]

AFAIK the old 3456 has a display oriented ADC - so the output steps directly corresponds to ADC steps. Also compared to the resolution the noise is relatively low. So it can happen, that quantization is hiding the noise, just it often happens with 3.5 digit DMMs.

[Andreas]

Cannot believe that there is no noise??

Hello,

look harder: the sign is toggling. (+/- 0.00000 V)
so there is definitely noise.
Only the output resolution is too low.

with best regards

Andreas

[enut11]

Repeated test on HP3456A with 1 NPLC. This time I recorded some noise.

The unit is operation on 240vAC and I changed the crystal to 4.875MHz

BTW, to attach file here I had to reduced size and save the spreadsheet as XLS  with reduced functionality.

The original spreadsheet from @IanJ and modified by @Bud:
https://www.eevblog.com/forum/metrology/3458a-logging-via-excel-macro/
the HP3456A version is at Reply #55

[maxwell3e10]

The autozero problem with Keithley DMM7510 meter is still not resolved. I wrote a little script to force autozero measurements and it works better than default settings, although still worse than Keysight 34465A.

I am attaching the script (change the extension from .txt to .tsp). If anyone has a stable voltage reference and a DMM7510, I am curious how it will perform at finite voltages with the custom zero algorithm.

[cncjerry]

I have two 3456a meters that will run for hours and hours with the inputs shorted without flipping the last digit or sign.   I ran one for 10,000 NPLC 100 samples once at 9.99999V displayed and recorded without changing.  The same reference on my 3457a using the HIREs register wanders all over the place but it has the extra digit of resolution.  I don't think this says anything about the 3456a or the reference other than within its range it is pretty stable but you can't compare the stability of a 6.5 meter to a 7.5 or higher.

Lately one of my 3456a meters seems to be drifting higher so I'll have to get under that issue.  I recently purchased a bunch of LM399s thinking I would make a parallel reference for it as another member did but again, within 6.5 digits how good does the reference need t be?

I love my meters! 

[Kleinstein]

With the input shorted, there is usually not much noise from the reference (at least it is avoidable). The shorted test is more testing the input amplifier (in lower ranges like 200 mV or 2 V) and the ADC circuit itself.

The 3456 seems to have rather low noise, but a coarse quantization. This makes is hard to see the noise. One could get more meaningful results with a shorter integration time, so that random noise is not swamped by quantization. A fully stable result is just not saying very much, it gives only an upper limit for the noise. Having an external reference and than measure is a test for the two references involved.

[blackdog]

Hi,

I need some help please... 

I want to start controling my Agilent 3458A with a measuring computer.
I Installed EZGPIB en loaded the "EZGPIB script for HP 3458A ver.1.01" from de website of Illya.

The software is running afther i changed the gpib_address to 11 for my 3458A

I espected it would change range and NPLC, before i run the script i put the 3458A in 10V DC Mode and 100 NPLC.
Afther starting the script it started measuring, stil on the 10V DC range and it swicht to a low NPLC value.
And now 15 minutes later nothing is changed?

Did i do something wrong?

Kind regards,
Blackdog

[pelule]

Hi
is the "END ALWAYS" command in the script. This is needed for the 3458A to respond.

/PeLuLe

[blackdog]

Hi pelule,

It has just switch range, maybe i'm to impatient. 

Kind regarts,
Blackdog

[ManateeMafia]

blackdog,

I believe the delay is intentional. I can't remember if there was an ACAL ALL in the script but it may have been there in case it was performed.

[Echo88]

Hi,
im trying to get the python-script "pars_noise_sd.py" which TiN wrote to work, but with the downloaded Python Interpreter 3.6 its just shows me the error:

Python 3.6.1 (v3.6.1:69c0db5, Mar 21 2017, 18:41:36) [MSC v.1900 64 bit (AMD64)] on win32
Type "copyright", "credits" or "license()" for more information.
>>>
========== RESTART: C:\MyMeasurements\HP34420A_1\parse_noise_sd.py ==========
| Multimeter    | Range      | NPLC   |          Counts   |    Sq.sum   |  Average   |    SD      |    SD,uV   |
Traceback (most recent call last):
  File "C:\MyMeasurements\HP34420A_1\parse_noise_sd.py", line 15, in <module>
    o.write ("|Multimeter|Range|NPLC|Counts|Sq.sum|Average|SD|SD,uV|\r\n")
TypeError: a bytes-like object is required, not 'str'

Then it produces an empty output.txt-file.
I use the HP34420A-example from the xdevs-DMM-Noise-website to log my Keysight 34465A-DMM and the produced .csv-files all look good to me.

Maybe someone can take a look at the problem? I want to contribute to the DMM-Noise-Database with the mentioned 34465A and also with my VXI-based HP E1410A 6.5Digit-DMM.

Attached is a produced example file. The file-ending .xls needs to be changed back to .csv (forum-upload-restrictions  ).

[fluxamp]

Hi,
im trying to get the python-script "pars_noise_sd.py" which TiN wrote to work, but with the downloaded Python Interpreter 3.6 its just shows me the error:

Python 3.6.1 (v3.6.1:69c0db5, Mar 21 2017, 18:41:36) [MSC v.1900 64 bit (AMD64)] on win32
Type "copyright", "credits" or "license()" for more information.
>>>
========== RESTART: C:\MyMeasurements\HP34420A_1\parse_noise_sd.py ==========
| Multimeter    | Range      | NPLC   |          Counts   |    Sq.sum   |  Average   |    SD      |    SD,uV   |
Traceback (most recent call last):
  File "C:\MyMeasurements\HP34420A_1\parse_noise_sd.py", line 15, in <module>
    o.write ("|Multimeter|Range|NPLC|Counts|Sq.sum|Average|SD|SD,uV|\r\n")
TypeError: a bytes-like object is required, not 'str'

Sounds like the code was written for Python 2.7. Try using that one or bug TiN to fix it 

[Echo88]

Nope, Python 2.7 also doesnt work, but at least produces "|Multimeter|Range|NPLC|Counts|Sq.sum|Average|SD|SD,uV|" in the output.txt-file.
Error Message:

Traceback (most recent call last):
  File "C:\MyMeasurements\HP34420A_1\parse_noise_sd.py", line 28, in <module>
    sum += float(row[setname[0]])
KeyError: 'HP34420A'

Guess ill bug TiN.

[TiN]

Don't use _ in the name of the insturment or filename. Underscore used as separator for filename/data parsing.

After rename your file into HP34420A_DCVOLT10_NPLC100.csv and correcting first line to date,HP34420A script work fine:

Code: [Select]

c:\__\code\py_parsesd>parse_sd.py
| Multimeter            | Range         | NPLC  |    Counts     |    Sq.sum     |  Average      |    RMS        |    RMS,uV     |
| HP34420A              | 10            | 100   |       74      | 0.00000000    | 1.051549E-07  | 1.698011E-07  |    0.17       |

Surely you can easily modify script to process different separator symbol, if you want keep underscores.

[Echo88]

Thank you! I will upload the DMM-noise-results of both my multimeters when theyre done.

[fonograph]

Here is the comparison of DMM7510 and 34465A for finite input. I used a 10.24 V 7Ah (almost dead) lead battery from a UPS supply and a 3.13 V (primary, moderately used) Li CR123A battery from a wireless alarm sensor. For lead battery, I removed linear self-discharge trend. For Li battery, just found a quiet temperature spot.

As might be expected, at finite voltage 34465A does not perform as well, probably due to limitation of the voltage reference. It would be best to compare 34470A to DMM7510 in this case.

@dvdput: Thanks for posting the data! Looking at the file on 10V scale the digital resolution is 3 uV. This is much worse than 34465A. If there is no way to change this setting it points to a clear superiority of 34465A at a smaller list price. It would be nice to add some of this information to the DMM comparison table, so people can decide which meter to buy with  information that is often not stated on the datasheets.

Does that mean that the Keithley 7510 problem where its inferior to keysight 34465 when doing 10 NLPC measurements with auto zero only happens when its shorted,but when its actualy measuring real voltage then its superior?

I mean,at the start of this thread there was this idea that to test the multimeter noise,the standard procedure was to short the inputs,then it was found that 7510 shows weird behaviour where the noise doesnt drop with higher NLPC like expected,now your graph for measuring real voltage,real electronic device shows that it works better than Keysight?

Does this mean its not a problem with normal use? Becose normaly people measure various devices and not the meter itself with inputs shorted.I think it doesnt matter how the noise measures with input shorted,only thing that matters is how it performs in normal usage.

I was considering buying it, is this real problem?

[dr.diesel]

I was considering buying it, is this real problem?

Yes, at a minimum the 7510 has a *somewhat* flawed AutoZero implementation.

[fonograph]

Is it software or hardware problem? Didnt Keithley fix it with new firmware patch or something?

Doesnt that graph I quoted show the problem only exist with shorted inputs?

[Kleinstein]

The unexpected extra low frequency (e.g. 0.1 Hz range) noise of the DMM7510 and some other Keithly meters seems to be a software problem, as it can be lowered a little with different triggering mode. So it does not seem to be a HW problem (e.g. temperature fluctuations). The funny thing is that also other Keithly meters like the K2002 seem to show a similar effect, though less obvious.
The Keysight meters show that there is no need for the extra bump in the noise spectrum / Alan deviation curve.

My suspicion is they use the average of several zero readings (over something like 10s) in stead of direct subtraction for the AZ mode. It might seem to be a clever way to reduce noise - but it does not work well, because it would let more 1/f noise through - so they trade in noise from the 100 Hz range to higher noise from the 0.1 Hz range. The AZ mode not only suppresses drift but also low frequency noise, just like in a chopper amplifier. The noise form the non AZ range suggests that the HW could do better, even if we take into account that continuous integration (like used in the HP34401) favors the non AZ mode a little.

Another possible reason could be a slow working numerical compensation of temperature effects. If they measure internal temperature every 10 s (or so, but more often if in triggered mode) and do some corrections based on this, it could have a similar effect.

When measuring a not so small voltage (e.g. more than 50% of FS), the noise of the references makes the excess noise less visible, but it is still there, though less obvious. The bump at about 15 seconds it still there for the 10 V input. The reference noise can be both from the signal measured and the DMM internal reference. It looks like the LTFLU reference in the DMM7510 is really good for a not heated reference. The very long times seem to be dominated from the source, thus making the LM399 in the 34465 look similar in the curves below.

The bump in the noise curve does not look that large, however the extra noise is just at the time (1-30s) scale where low noise is most wanted. For precision measurements this is about the time scale used for manual switching or external polarity reversal. So it is a kind of embarrassing at 10 s aaveraging and 3 V the DMM7510 is not better than the cheaper 34465 and for a short even a factor 5 higher in noise.

There where a few SW updates, but AFAIK they where addressing mainly software crashes of the GUI and not the measurement itself.

[HighVoltage]

Is it software or hardware problem? Didnt Keithley fix it with new firmware patch or something?

Doesnt that graph I quoted show the problem only exist with shorted inputs?

As others, I have pointed this problem out to Keithley directly without any real valuable response to this particular issue.
All other issues that I had got addressed in the many FW updates that followed.
This leads me to believe, it might be hardware related and can not be fixed by software.