DMM Noise comparison testing project

[dr.diesel]

This leads me to believe, it might be hardware related and can not be fixed by software.

I doubt the AZ issue is HW related.  The software issues were fixed (They still haven't fixed the web interface issues, at least it wasn't noted in the changelog) fairly quick because once IDed and reproducible that stuff is usually easy.  The AZ issue on the other hand would require the *right* guy at Keithley to research and study what is going on.  The difficulty here might also be in the fix, it would alter the behavior of the instrument and might be an issue for some customers, requiring re-certification or production line changes.

I suspect we will never get any traction on this issue.   

[maxwell3e10]

This plot shows that with 10 V input one can remove the noise bump by using a custom trigger script on DMM7510, at the expense of higher short-term noise:
https://www.eevblog.com/forum/testgear/keithley-dmm7510-smu-2450-2460-problems/msg1194829/#msg1194829

I was going to do a more fair comparison of Keysight 34470A with the DMM7510 at finite voltages because 34465A is limited by poorer voltage reference.
Also, I am wondering if the 1/f noise without AZ is better in Keysight meters, so maybe Keithley has to work harder to suppress it.

In terms of practical impact, the 1/f noise in DMM7510 is embarrassing if one has to show Allan deviation plots or even time domain plots for a low-noise DUT. When someone asks why does your DUT has 1/f noise, try to convince them its actually the meter.

[fonograph]

Indeed at 0V it shows Keysight is better,at 3.13V its match,then at 30 second+ the keysight pulls ahead,but at 10V the Keithley is better,both short and long term!

Just look at the graph! The keithley in black is under the red keysight curve! So at 10V keithley is better and this autozero problem doesnt make it inferior like when measuring 0V,whitch is short.

My point is,as a person that want best dmm to measure 10V,I should not be put off by the poor 0V noise of 7510 since at 10V its better than keysight,or I am wrong?

[VintageNut]

Anecdotally, I have used my DMM7510 to measure one of my Fluke 731Bs in an environment where the temperature fluctuates p-p 10 degrees F or more. The long-term measured standard deviation of the 10V output of the 731B is around 1/4 ppm.

In the long-term graph, you can see the rolling hills of the temperature in my house.

I consider the DMM7510 to be stable, repeatable and very low noise measuring my 10V Fluke 731B. I do not need any better performance.

[maxwell3e10]

My point is,as a person that want best dmm to measure 10V,I should not be put off by the poor 0V noise of 7510 since at 10V its better than keysight,or I am wrong?

If your application is specifically to measure 10V and if you are comparing DMM7510 to 34465A, which is about 1/3 of the price, then indeed DMM7510 is better.

I recently got a 34470A, so I will run at some point a direct comparison DMM7510 to 34470A. A few other people here can also do it at 10 V. Would be good to throw in a 3458A as well.

[fonograph]

My point is,as a person that want best dmm to measure 10V,I should not be put off by the poor 0V noise of 7510 since at 10V its better than keysight,or I am wrong?

If your application is specifically to measure 10V and if you are comparing DMM7510 to 34465A, which is about 1/3 of the price, then indeed DMM7510 is better.

I recently got a 34470A, so I will run at some point a direct comparison DMM7510 to 34470A. A few other people here can also do it at 10 V. Would be good to through in a 3458A as well.

That would be great! I am currently deciding between 34470A and 7510,any test comparing the two directly would be very appreciated.I would also like other measurements tested,not just DCV,also ACV and DCI,ACI,resistance 4 wire and capacitance

[Kleinstein]

This leads me to believe, it might be hardware related and can not be fixed by software.

I doubt the AZ issue is HW related.  The software issues were fixed (They still haven't fixed the web interface issues, at least it wasn't noted in the changelog) fairly quick because once IDed and reproducible that stuff is usually easy.  The AZ issue on the other hand would require the *right* guy at Keithley to research and study what is going on.  The difficulty here might also be in the fix, it would alter the behavior of the instrument and might be an issue for some customers, requiring re-certification or production line changes.

I suspect we will never get any traction on this issue.   

Fixing the AZ issue should also be relatively easy, as the weak point is relatively well localized. However it will take a different guy than the user interface part and this might be old code they have to touched for decades (the K2002 seems to have a similar issue, though less obvious). I don't see why there should be an important change in overall behavior. If they are really afraid of this, they might even add an optional choice of AZ mode (so not just AZ off / on), since there are a few possible good choices depending on the situation:
1) the old Keithley mode with averaging for compatibility (not much other advantage)
2) using the zero before and after the signal with half weight - should be the obvious solution
3) using the zero reading before the signal only - to get less latency for reading
4) use less time on zero correction  - useful for relatively noisy signals.
The slightly different behavior when changing the AZ mode might cause some changes and this can include a few negative ones

[HighVoltage]

That would be great! I am currently deciding between 34470A and 7510,any test comparing the two directly would be very appreciated.I would also like other measurements tested,not just DCV,also ACV and DCI,ACI,resistance 4 wire and capacitance

I have both instruments and like them a lot.

Here is an interesting startup comparison between the 34470A and 7510
Actually I have one 7510 and two 34470A hooked up to a good and reliable warmed up 10V source.
Then turned all three instruments on at the same time.

The 7510 has a very interesting warmup graph and is much faster at reliable numbers than the 34470A.
Here are some comparison graphs.
Essentially the Keithley 7510 has reached reliable values after 3 min and the Keysight 34470A at 30 min
 

[maxwell3e10]

The 7510 has a very interesting warmup graph

Yes, I remember noticing the steps in the voltage when the temperature is changing quickly even when its reading zero. I am not sure if its an ideal behavior, but they probably do a more aggressive temperature correction than Keysight.

Can you leave the instruments recording 10 V for a few hours at 10 PLC and post the data?

[fonograph]

omg yes! I love you HighVoltage! Thats the kind of information I need,I am real jealous that you have 2x 34470 and 7510,any kind of comparsions and performance observations would be highly welcome.

The fact that it hits operational temperature in 1/10 time is significant.But what about ACAL? Both 7510 and 34470 have automatic calibration that compensates for temperature,or is it not important when it comes start up time required for stabilization?

How long does the 7510 ACAL last anyway,34470 is around  15 seconds if I remember correctly.I have read that 7510 fan is loud and 34470 is quiet,is that true?

[dr.diesel]

How long does the 7510 ACAL last anyway,34470 is around  15 seconds if I remember correctly.I have read that 7510 fan is loud and 34470 is quiet,is that true?

Should probably move further discussion to one of the 7510 threads, or a new one, but:

7510 ACAL takes 7min 4sec.

I've not been utilizing the temp based ACAL scheduling option of the 7510 because I haven't tested/verified it, maybe a project for the near future. 

[HighVoltage]

Can you leave the instruments recording 10 V for a few hours at 10 PLC and post the data?

Here is a first test of my 7510 at a stable 10V input with 10PLC setting and AZ ON.
I have the instrument running over night and will post a full buffer file tomorrow.

[maxwell3e10]

This looks good, similar to other measurements. The long-term stability is quite good.
What would be most interesting is to take 7510 and 34470 data at the same time to compare them directly.

[dr.diesel]

This looks good, similar to other measurements. The long-term stability is quite good.
What would be most interesting is to take 7510 and 34470 data at the same time to compare them directly.

I wonder what improvement could be gained by proper screw type banana jacks, similar/same as the 3458A...

[Kleinstein]

The very low frequency noise that might come from thermal EMF and similar is not only coming from the terminals. There a quite a few internal points (e.g. relays, the front / back terminal switch, PTCs for protection, temperature gradients at chips/fets). So I won't expect a great improvement from just different plugs.
For really low level signals the special meters like the 34420 have the better input amplifier and you don't need a high end ADC and reference at the nV level anymore, as other errors are usually larger.

With an external source there are also variations from the source. Even a good 10 V reference (except JJA) is usually less stable than the thermal EMF limit. One of the older curves comparing the 34465 and 7510 showed this when measuring both simultaneous - the very low frequency parts was from the source as both showed the same variations.

[maxwell3e10]

Here is my test of the Keithley DMM7510 and Keysight 34470A directly against each other with a 9.6V battery input. So the noise bump in the Allan variance at 20 sec in DMM7510 is still quite pronounced.



The 34470A meter seems to have extra noise on the 100-500 sec time scale. This can be compared with tests by Jorn, https://www.eevblog.com/forum/metrology/dmm-adc-noise-comparison-testing-project/msg1136604/#msg1136604, where there is no rise in the noise and the Allan variance stays below 0.2 uV. Perhaps it is due to better ambient temperature stability.

[branadic]

Question: What noise would one expect, if DMM measures its own voltage reference in 10V range?

I'm currently observing the reference voltage on my Prema 5017 by itself and I'm the opinion that the noise is rather high. Attached is a picture directly after powering the unit up with capturing data with 0.5Hz repetition rate. I will go one measuring and will then measure the reference voltage with 3458A for comparison. Maybe there is some room for improvement within the circuit by changing some opamps by lower noise types?

-branadic-

[Cerebus]

Question: What noise would one expect, if DMM measures its own voltage reference in 10V range?

That's a bit of a "how long is a piece of string?" question.

It's going to depend heavily on the ADC architecture in use.

In something like an HP 3458 or 34401 where the reference voltage is going to be applied simultaneously down the 'measurement' and 'charge balance' routes then it's going to have correlated noise that will, to some extent, either cancel or reinforce itself. Whereas if it's a classic double slope integrator then the reference voltage will be applied serially in time and there will be no cancellation. Both types will, of course, band limit the noise to ¹/_2t where t is the aperture time.

And so on for other ADC architectures.

I think you can only come up with an expectation after an, at least moderately, rigorous trip over a detailed block diagram of the meter in question with some good guesstimates to hand of the noise characteristics of each stage in the block diagram.

[maxwell3e10]

Here is a bit more data on direct comparison of DMM7510 and 34470A. In my lab there is often a pronounced temperature oscillation of a few tenths of a degree with a period of 1000 sec. One can see nice Fourier peaks in the temperature and the voltage. For these tests the meters are placed on a lab bench with plenty of space around each one, while the battery is well thermally insulated. Under these conditions, Keysight 34470A has about a factor of 2 higher sensitivity to temperature changes compared to Keithley DMM7510. That is the origin of the peak in the Allan deviation around 500 sec which drops down at 1000 sec because the temperature changes are fairly periodic.

[Kleinstein]

When measuring it's own reference the ADC will see some of the reference noise, but not all. What exactly depends on the ADC type. Especially the very low frequency part is well suppressed. There may be a significant contribution (could very well be amplification)  from some higher frequencies (where the modulation in CB ADCs happens). Of cause the amplifier part is still there.

AFAIK the Prema meters use an AZ OP at the input. This could be a significant contribution to the noise in the lower voltage ranges, but it usually is not for the 10 V (or similar) range. Here changing the OP might be possible  (e.g. LTC1050/ICL7650 to LTC1052/ICL7652). However with AZ OPs it is usually a compromise between bias current and voltage noise. The very lower noise AZ OPs usually have to much bias and maybe to much current noise too.

The Prema meters use an special ADC chip - so nothing to improve there. There might be a small chance to improve reference filtering (to reduce the higher frequency noise part). At least a few meters (e.g. Keithley 2001 AFAIK) seem to have skipped on this rather simple option. However I am not sure on how much that contribution is - it may not be that much. Looking in detail at the noise source could be interesting, but would be more likely worth a separate thread - if it leads to significant results they would be interesting in this thread too.

[branadic]

When measuring it's own reference the ADC will see some of the reference noise, but not all. What exactly depends on the ADC type. Especially the very low frequency part is well suppressed. There may be a significant contribution (could very well be amplification)  from some higher frequencies (where the modulation in CB ADCs happens). Of cause the amplifier part is still there.

AFAIK the Prema meters use an AZ OP at the input. This could be a significant contribution to the noise in the lower voltage ranges, but it usually is not for the 10 V (or similar) range. Here changing the OP might be possible  (e.g. LTC1050/ICL7650 to LTC1052/ICL7652). However with AZ OPs it is usually a compromise between bias current and voltage noise. The very lower noise AZ OPs usually have to much bias and maybe to much current noise too.

The Prema meters use an special ADC chip - so nothing to improve there. There might be a small chance to improve reference filtering (to reduce the higher frequency noise part). At least a few meters (e.g. Keithley 2001 AFAIK) seem to have skipped on this rather simple option. However I am not sure on how much that contribution is - it may not be that much. Looking in detail at the noise source could be interesting, but would be more likely worth a separate thread - if it leads to significant results they would be interesting in this thread too.

Thanks, but don't you think that those ~30µVpp in 10V range are a bit much? I would have expected something significant smaller.

-branadic-

[Kleinstein]

30 µV_pp are too much for a 0.5 Hz update rate. There is a slight chance that measuring the internal reference also gives a path for RF noise to enter the system. So the fist test would be the simple readings on the shorted input. The second test would than be reading a independent stable voltage (e.g. 9 V block).  Ideally reading the own reference should give a lower noise than an external stable voltage, down to the level of the shorted input, but usually a little more.

[branadic]

I will start a new thread on this topic.

-branadic-

[Andreas]

Question: What noise would one expect, if DMM measures its own voltage reference in 10V range?

I'm currently observing the reference voltage on my Prema 5017 by itself and I'm the opinion that the noise is rather high.

Hello,

if I look at the datasheet of the ADC the noise is explainable for me:

https://www.ohh.de/5610.htm

2 seconds integration time correspond to N1=30 and N2 = 10 thus around 23.6 bits.
Input range is +/-30 V so 60V / 23.6 bits gives 4.7uV.

If I assume that the 3 sigma has to be multiplied (and not divided)
we get around 15uVp / 30uVpp noise alone from the ADC.

So I fear that the analog path with 2uVpp estimated maximum noise is not the mayor source of the noise.

with best regards

Andreas

[fonograph]

Here is a bit more data on direct comparison of DMM7510 and 34470A. In my lab there is often a pronounced temperature oscillation of a few tenths of a degree with a period of 1000 sec. One can see nice Fourier peaks in the temperature and the voltage. For these tests the meters are placed on a lab bench with plenty of space around each one, while the battery is well thermally insulated. Under these conditions, Keysight 34470A has about a factor of 2 higher sensitivity to temperature changes compared to Keithley DMM7510. That is the origin of the peak in the Allan deviation around 500 sec which drops down at 1000 sec because the temperature changes are fairly periodic.

could you please please please do 34465,34470 and 7510  all at once? You did 7510 vs 34465 with various voltage ranges and then 34470 vs 7510 at 10V only.If you could do all three with variety of voltages at same time that would be awesome.

If you could also test them side by side cold start on warmed up reference to see how they stabilize with respect to time.