Tektronix DMM 4050 shows exceptionally good performance

[quarks]

Hello all,

as already mentioned in several posts, the Fluke 8846A / TEK DMM 4050 are great meters (probably the very best you can buy).
Since I bought my 4050 I am very happy with it and alway had good confidence in the accuracy. Against my very first Geller SVR I measured within a few ppm, but I did not know the real absolut values. Since then I added a lot of better and better calibration lab gear. And the 4050 repeated to show way better then spec results. At least for one year by now I have readings around 1ppm against my very best gear at 10VDC.

Today I finished a many hours STATS reading against my Fluke 732A and compared it over many hours with my HP3458A.
I have no clue how this is possible, but both meters agree in 7.5 Digit mode to the last digit without any tweaks/adjustments (see att. picture). 

Here is my setup info:

4050 running in STATS mode over night
3458A turned on this morning, 100NPLC, ACAL after warmup of 4h
 
I already know from robrenz that his 8846A is equally good, but I really wonder if others have similar exeptional results to report.

Bye
quarks

[robrenz]

Very nice work quarks!

If you don't mind this looks like a good thread to post some of my correspondence with Fluke tech support.

This is an email response to a question I asked by phone to Krista at tech support.

Hi Krista,

If I understand the situation correctly, the main cause of the difference in resolution is that the remote port has higher resolution than the front panel display.  I'm guessing that Statistics mode doesn't use the full remote resolution, and it may not do internal math the same as Trend Plot, but someone would have to dig through the firmware code to know that for sure.  In both cases, the resolution is higher than the advertised 6-1/2 digits, which only gives 10 uV display resolution on the 10 V range.

As for stability, we don't promise anything better than the 24 hour specifications.  But, the instrument will usually perform better than specified if the environment is better controlled, for example; the air temperature is kept very constant, air flow around the 8846A is kept low to none, AC line voltage is kept constant, and humidity is kept constant.  Of course, low thermal emf connections and wires also need to be used.  Using 100 PLC helps with noise too, but it will be really tough to keep the environment around the 8846A constant enough to avoid all drift.

Let me know if more is needed.
Best Regards,
Ray


Thanks Christa/Ray,

I am a big fan of the 8846A so no complaints here just trying to understand nuances of the meter. I didn’t state my question very well the first time so I hope this is more understandable. For everything stated below the digital filter is on. I am very happy with the stability of my 8846A I think it is extremely good with shorted inputs at 100 NPLC giving a total span of 1µV with a SD of 73nV over 152hrs with a 2.4 degree C ambient temp variation over that same time.  But I am doing some stability testing on power supplies that have sub 30µV drift over 8 hours at 10V output.  If I use trend plot directly I only get 1mV resolution so no good.  If I use stat mode I get 10µV resolution so better but not great. So I zero out the 8846A and go into trend plot and I have 10nV resolution of the trend plot readings of the power supply drift.  If zero/offset is done purely mathematically I think that would require 30 bits to get 10nv resolution of a 10V signal. If that is the case  then I understand the stability would be basically the same as my shorted inputs testing. But if zero/offset is an internally generated analog offset, is it as stable as when zero/offset is not used.  Do I need to add an additional uncertainty to my characterized shorted input stability above when using zero/offset in this manner? I understand how with averaging you can get increased effective resolution. But trend plot zeroed with a 10V signal applied gives 10nV resolution on all resolutions/NPLC. Is the meter working internally at 30 bits or more without averaging all the time? Relative to remote port vs display: at values lower than 100µV trend plot gives 1nV resolution where stored values to usb stick only read 10µV resolution
Thanks for your time, Robin

Robin,

While the details of the math and resolutions being used are a challenge to fully determine, we do know that the hardware is capable of very high resolution, especially at 100 NPLC.  There could very likely be 30 bits (100 million counts) of hardware measurement resolution at 100 NPLC.  As measurement rates are increased (e.g. 1 NPLC), the hardware measurement resolution for each measurement decreases.

When the ZERO button on the front panel is pressed (or the remote Calc: Null command is used), the readings are offset
purely mathematically. 

All of the readings of a 10V input are still being taken on the 10V range.  Fortunately, the 10 Vdc range is the highest performance range in the instrument - it has the lowest noise, lowest drift, and best linearity. 

So, the internal trend plot math is giving additional resolution, possibly partly as a result of averaging multiple readings, and
at 100 NPLC because of  the additional internal measurement resolution. 

To get a better idea of what the 10 Vdc range stability and noise is, the shorted input test can be performed with the meter locked (manually ranged) in the 10V range.  The 10V range performance
should be mostly independent of input level, whether 0V or 10V.

Hoping this helps, Ray


Hello Ray,
 
Thank you in advance for your patience.
 
Regarding hardware resolution I find that when manually ranged to 10V, with a 10V input signal and zeroed in math mode, Trend plot yields 10nV resolution regardless of NPLC setting. Since zero is purely mathematic that implies the hardware resolution is that high (1 part per billion) even at the shortest integration time. This is in 4 digit .02 NPLC mode, 10V signal, manual 10V range, Zeroed in math, trend plot


Regarding the 10V range being the highest performance I did some testing based on your input and found a curious response shown below. Even though the noise on the 10V range is unbelievably low its value is compromised by the spurious spikes. Any clue as to their cause? I have a metrology friend with a tek4050 and he gets the same thing but his spikes are much more frequent about every 5-8 seconds and are 2.9µV.
 
10V manual range shorted inputs 100NPLC digital filter on. (not zeroed using math)
Amazing, since the resolution is 10nV that is sub 10nV noise on the 10V range over a 3 minute time frame.



Confirmed in Stat mode over a 3 minute time frame.


But when I go longer I get a consistent 1.79µV spike.  Notice the exponential decay shape of the return to zero. Is that a clue to the cause of the spike?


And both polarity over an hour but predominantly negative.


Again confirmed with Stat mode


I suspected power line influences but here is the same testing but manual ranged to the 100mV range. (basically same results on 1V range just not shown)
That is a 189nV min-max span compared to the 3.58µV min-max span on the 10V range over 22 minutes.


It takes about 2.5 hours before I see any spurious activity and it is still far from the +/- 1.79µV of the 10V range.


2 days does not change the height of the spurious activity. You can see the sinusoidal daily temp variation influence.


I checked the AC line on AC low resolution, zeroed in math mode on trend plot to see if anything crazy was
going on there but nothing that correlates to the spikes on the 10V range over 2 hours.


Thanks for all your help,
Robin

Robin,
Just to keep you up to date, I have forwarded this email to Ray for an answer.  I think his is out of the office until next week, I will let you know when he responds.
Lance

Lance,

No problem, I am not in a hurry.  Just the fact that a question like this is even acknowledged and responded to at all says a lot about Fluke. It is one of the reasons I bought the 8846A. I called for pre sale technical support and got excellent response from people who knew what they were talking about.  I believe I talked to you quite a while ago about switching method for zeroing out thermal emfs on 4 wire connections on the 8846A and got excellent help.
 
Thanks,
Robin

Hello Lance or Ray,
Just checking to see if I am still in the queue?  I know you guys are probably spread too thin so I am not complaining.

Thanks,
Robin

Robin,
See below for Ray’s answer to your inquiry.
 
Hi Robin,

Thanks for being patient, we have been quite busy, and thanks for the plots.

Now that I see some of the plots, I have a better idea of what is likely going on.  The first thing to realize, even though it isn't obvious, is that the numerical resolution shown on the front panel for the plots is not directly related to the internal measurement math.  When the front panel trend plot routine isgiven small numbers, it automatically comes up with 10 nV resolution, simply because of number display autoranging coupled with space limitations.  On the flip side, when the measurements are in volts, the trend plot numbers don't have enough resolution to show the full internal resolution.

When a plot shows no variation on the 10V range for 3 minutes, I'm thinking that the A/D readings are all the same below the internal math resolution, which looks to be around 0.5 uV.  So, when the 1.8 uV spikes appear, it is because a reading has finally come through that is a few internal measurement math counts different.  One spike looks to be +0.5 uV and another -0.9 uV.  0.5 uV isn't as good as 10 nV, but it still is only 0.5 parts out of 10 million, or 0.05 ppm of full scale.

I don't know what causes the decay or tail on the spike, but it may simply be a side effect of the trend plot math plotting routines when the display time resolution is higher than the measurement time interval.

The spikes are probably due to 1/f noise, where a very low frequency noise change results in a different measurement result at the internal math resolution level.  This will be infrequent in time, but will tend to have a consistent appearance over a long period of time, as your plots show.  Although, there may be other more systematic noise causes too. 

As for the more infrequent "spikes" on the 100 mV range, it can be quite difficult to tell the difference between very low frequency 1/f noise, and other causes.  And, because of the different amplifier gains and circuit elements used, each range often has different noise behavior.

It's good that the AC line is not directly affecting the readings.

Hoping this answers your questions,

Ray

Hi Lance/Ray,
 
Thanks for the info. I need to digest that a while (I am not an EE). I appreciate you both taking time to answer questions that are clearly outside the stated specifications. Just a thought on the spikes, could they be popcorn noise related?
 
May I have your permission to post excerpts of this content on a electronics forum?  There is another 8846A owner that is interested in this topic. If not I will respect your wishes.
 
Thanks again for all your patience and help.
Robin

Robin,
It is possible that the spikes are caused by noise of some kind.  I have no problem with you posting this information to assist others.

Lance

[dr.diesel]

That is very interesting!

I wonder if it's younger brother, the 4020/8808A, has a higher resolution via serial?  (time to plug it in and find out)

[quarks]

If you don't mind this looks like a good thread to post some of my correspondence with Fluke tech support.

Not at all. That fits perfect.

I forgot to mention, temp change during the measurement was about 2°C (from 19.5°C to about  21.5°C).

And here is a short update after about another 4 hours (see picture).
Still the same perfect result, only the STATS MAX changed one Digit from 9,999927 to 9,999928 and SD changed. 
I stopped it here and restarted STATS again and also changed 3458A to 8,5 digits.

[quarks]

after the restart it looks again extremly good

[dr.diesel]

I wonder if it's younger brother, the 4020/8808A, has a higher resolution via serial?  (time to plug it in and find out)

It does not, 

But my Hameg HMC8012 does, I'll have to test and see if it's all just noise.

Code: [Select]

4.99981546E+00

[msobreira27]

That's awesome information, thank you! It happens that I have one a tek 4050 and I was characterizing a Precision Designs 2005p and I was puzzled on how to improve the resolution on the analyze functions. I did not think of the zero, now it is much better.

I was researching to understand the level of noise of these multimeters - I'm thinking on participating on the xdevs public experiment on noise figures for multimeters.

https://xdevs.com/article/dmm_noise/

But it is very good to know that that multimeter stands high among other players.