KX Reference

[hwj-d]

If i touch them, it has directly impact to the measurement.

Is the change immediately = capacitive influence
or slowly rising / falling = thermal offset?

I know, that must be kept apart.
If i touch or move them, this becomes visible at the 10 or 20µV curve at the dmm measuring with 100 plc for a time. Is it slow, or immediately?

[Andreas]

Hello,

normally if it is thermal it will change over several 100 NPLC measurements (similar to a e-function) if you put your hand on the wiring and let it there for a while or if you remove it again. Thermal time constants are usually more in a 1 minute (or larger) range.

If it is instaneous (= within one 100 NPLC measurement) you have most likely a capacitive influence = EMI problem.

with best regards

Andreas

[hwj-d]

normally if it is thermal it will change over several 100 NPLC measurements (similar to a e-function) if you put your hand on the wiring and let it there for a while or if you remove it again. Thermal time constants are usually more in a 1 minute (or larger) range.

If it is instaneous (= within one 100 NPLC measurement) you have most likely a capacitive influence = EMI problem.

I know, my measurement is not optimal too. But look at this with 100 NPLC taken:


The impact at the end comes from touching [edit: carefully only DMM] using USB-Stick for screen capture just before. The new measurements after that stuck at this new level. My impression is, that this is caused from capacitive influence, but changes the measuring till other TC influence overlapped the result.

But, i think, the lm399 based 3446xa are VERY TC sensitive as others noticed too. That is not nearly comparable to the stability like a 34470a nor 3458a (... as we know   )

https://www.eevblog.com/forum/metrology/mx-reference/msg1296874/#msg1296874

[hwj-d]

0.4ppm/K tempco is more than 8 times worse than expected (<0.05), so there is definately a system problem.

Yes, but this must match high grade environmental conditions with comparable dmm's like your keithley's or 3458a's. Thats the task, after that, to have one thing in the lab, that provides this requirement, after cal, with yours and others help ...    

[Dr. Frank]

normally if it is thermal it will change over several 100 NPLC measurements (similar to a e-function) if you put your hand on the wiring and let it there for a while or if you remove it again. Thermal time constants are usually more in a 1 minute (or larger) range.

If it is instaneous (= within one 100 NPLC measurement) you have most likely a capacitive influence = EMI problem.

I know, my measurement is not optimal too. But look at this with 100 NPLC taken:


The impact at the end comes from touching [edit: carefully only DMM] using USB-Stick for screen capture just before. The new measurements after that stuck at this new level. My impression is, that this is caused from capacitive influence, but changes the measuring till other TC influence overlapped the result.

But, i think, the lm399 based 3446xa are VERY TC sensitive as others noticed too. That is not nearly comparable to the stability like a 34470a nor 3458a (... as we know   )

https://www.eevblog.com/forum/metrology/mx-reference/msg1296874/#msg1296874

Well, I don't think, that the 3446x instruments are that sensitive.
Never encountered such dips, when touching the case.
I assume, that your LTZ circuit is a bit sensitive.. see current discussion about shielding, in the MX Reference thread.

To have an idea about the stability of the 34465A (LM399) vs. 34470A (LTZ1000A), see my review here:
https://www.eevblog.com/forum/testgear/keysight's-new-34465a-(6-5-digit)-and-34470a-(7-5-digit)-bench-multimeters/msg889217/#msg889217

In your graph, the scale of noise and of the dip is not obvious.
By using the scaling math function and the unit function, you might normalize your graph to read ppm deviation.. then you'd have much more resolution in the statistics also. 

Frank

[d-smes]

I know, my measurement is not optimal too. But look at this with 100 NPLC taken:


The impact at the end comes from touching [edit: carefully only DMM] using USB-Stick for screen capture just before. The new measurements after that stuck at this new level. My impression is, that this is caused from capacitive influence, but changes the measuring till other TC influence overlapped the result.

But, i think, the lm399 based 3446xa are VERY TC sensitive as others noticed too. That is not nearly comparable to the stability like a 34470a nor 3458a (... as we know   )

https://www.eevblog.com/forum/metrology/mx-reference/msg1296874/#msg1296874

I agree '399 based 3446xa are very TC sensitive.  My results at https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg1245750/#msg1245750 Dr. Frank suggested several improvements to my connections, but that didn't alter the high correlation of 34465a readings with temperature.

Regarding the reading shift from touching DMM, I too have noticed this.  I usually change grounding around and/or tie floating DUT to meter case or earth ground.  After a bit of experimenting, I can usually get the touch artifacts to go away. 

[Andreas]

The impact at the end comes from touching [edit: carefully only DMM] using USB-Stick for screen capture just before.

But, i think, the lm399 based 3446xa are VERY TC sensitive as others noticed too. That is not nearly comparable to the stability like a 34470a nor 3458a (... as we know   )

Hello,

a capacitive effect would be reversible. So if it is a one way jump it is a different story.

From my experience the LM399 in HP devices are carefully selected (i.e. unheated tempco near zero).
If you measure the reference output voltage it should be somewhere near 6875 mV. (see LM399 thread).

with best regards

Andreas

[hwj-d]

Well, I don't think, that the 3446x instruments are that sensitive.
Never encountered such dips, when touching the case.
I assume, that your LTZ circuit is a bit sensitive.. see current discussion about shielding, in the MX Reference thread.

To have an idea about the stability of the 34465A (LM399) vs. 34470A (LTZ1000A), see my review here:
https://www.eevblog.com/forum/testgear/keysight's-new-34465a-(6-5-digit)-and-34470a-(7-5-digit)-bench-multimeters/msg889217/#msg889217

In your graph, the scale of noise and of the dip is not obvious.
By using the scaling math function and the unit function, you might normalize your graph to read ppm deviation.. then you'd have much more resolution in the statistics also. 

Thanks Dr. Frank for your answer. I can only learn from that, what happens (t)here.
I know, the scale of noise and of the dip is really not obvious.
I don't want to mix this dip problem with that TC of my meter. That's a 34461a.

Is it right, that this meters have TC of 1ppm/k by itself? So, that we can only measure not better than that TC with it?

Best regards

[TiN]

So, that we can only measure not better than that TC with it?

Not really. If you maintain temperature constant, you have constant tempco as well.
And since you only want to measure tempco drift of DUT, you can ignore any tempco effect on the meter's own reference completely, if it stays unchanged during complete test run. That's why I perform very slow and linear temperature ramp up and down on DUT, to make sure meter's own unstability does not introduce large errors.
If both direction ramps produce matching data, confidence is high. 

[hwj-d]

Hello Andreas,
thanks for your answer too.

If you measure the reference output voltage it should be somewhere near 6875 mV. (see LM399 thread).

I can't really do that. That meter is relative new, one year old. Maybe i put it to an other room with more constant temperature, other shielding methods, and so on.

I assume that this dip is a mistake from my own. But this relatively bad tempco is one of the meters not so good property. d-smes has confirmed this, and Ian Johnston has already dropped such a notice about this meter, as far as i can remember. Certainly it depends on the  environment, this meter must do his job.

[hwj-d]

So, that we can only measure not better than that TC with it?

Not really. If you maintain temperature constant, you have constant tempco as well.
And since you only want to measure tempco drift of DUT, you can ignore any tempco effect on the meter's own reference completely, if it stays unchanged during complete test run. That's why I perform very slow and linear temperature ramp up and down on DUT, to make sure meter's own unstability does not introduce large errors.
If both direction ramps produce matching data, confidence is high. 

That's the problem. If i measure the DUT over the 24h and the temp changes 2-3 °C up and down, i don't know, from what the measured changes comes, maybe in combination of DMM and/or DUT, if i don't have a definitely "comparing-normal" (the objektive meta measurement) to that. So, IF i have a calibrated LTZ reference, and i know THAT TC, the deviation must come fro the dmm. And vice versa. If the LTZ is the DUT, i dont' know anything, if i can't trust the dmm. I always see the TC of the DUT AND the DMM. The LTZ-reference should control that, in the future, if it is calibrated. Maybe i'm wrong? Am I fail to see something?

[TiN]

You are correct in point that data represents total system tempco. Sure, having known "tempco" reference would make things easy.

But even without, you can still variate temperature for parts of the system (DUT only, or meter only) separately, right?
If your oven provides stable chamber inner temperature (<0.05C stability), you can use that easily. You just need to "anchor" one of the parts of the system to establish remaining system tempco. Having LTZ small and low power, I'd expect keeping it at stable temperature is much easier than large DMM.

If you use KX onboard MAX6610 sensors, they will allow you to check board stability well enough for the purpose. Here on page I've logged MAX6610s temperature output during my tempco test. Black ramp is reference box temperature, measured by Honeywell HEL-705 RTD. Using both sensors you can even see when LTZ oven is approaching to thermal headroom limit (closer the temperature directed from less margin from LTZ heater room). You need to stimuli LTZ DUT with much larger temperature delta, to be able even see the tempco effect, doesn't matter what meter you use, be it 34461 or 3458A, just because of the own LTZ's noise. I usually do tempco ramps from 20-25C to 40-45C to get 20C delta. If you get less than 1ppm difference, your reference good to go.

Said above, also means there is little meaning or usefulness in doing box tempco with just few temperature points, instead of the linear ramp, as you don't know tempco behaviour on unknown device/meter. Is it linear? Is it logarithmic? Full world of possibilities... 

P.S. 56K warning, linked pages are realtime JS bodge, may take a minute or two to load...

[Kleinstein]

If one can not make the meters temperature constant, one could try to measure it and than compensate numerical. There is software to get at least linear correlation and this way get estimates for both TC's: the meter and the external reference.

 It can help if one looks at more than one temperature cycle for the external reference. However the speed of the temperature cycle is somewhat limited, as fast temperature changes can have a different effect. Not all temperature effects are instantaneous and fast changes tend to include gradients. Over many cycles variations of the meters temperature can average out, or at least a good decomposition is possible.

[TiN]

However the speed of the temperature cycle is somewhat limited, as fast temperature changes can have a different effect.

Yup, hence all my ramps are graceful and slow (often 40+ hours).

Linear math compensation for meter tempco during 2xLTFLU 13V reference measurement also worked well for me.
Total error was <1ppm, peaks excluded, even with 24C-31C ambient temperature change, but granted I've used bit more stable 2002.
+0.4ppm/K with reference +26.3C temp was used for linear correction.

[hwj-d]

@TiN
Ok, got it.
Then MAX6610 and your measurements is the anchor.

Thanks a lot. 

Edit:

P.S. 56K warning, linked pages are realtime JS bodge, may take a minute or two to load...

Depending your upload, no problem here...

[kj7e]

Maybe the tempco is not as bad as I thought.  Repeated cycling the KX board from 25.0 deg C to 35.0 deg C four times and achieved the exact same results;
@ 25.0 deg C, 7.141510
@ 35.0 deg C, 7.141504
So 6uV over 10 Deg C, or 0.6uV/C

Scaled to 10v, that would a delta of 0.00000084 or 0.084 PPM/C.  Please correct me if I'm wrong here.

[Kleinstein]

0.084 ppm/K is still slightly higher than the data-sheet value of 0.05 ppm/K, but not much. If in an environment with not so much variations, this should no be a real problem. Inside a oven environment it would be not a problem at all.

[Mickle T.]

0.05 ppm/K is a typical value, not maximum.

[Dr. Frank]

Thanks Dr. Frank for your answer. I can only learn from that, what happens (t)here.
I know, the scale of noise and of the dip is really not obvious.
I don't want to mix this dip problem with that TC of my meter. That's a 34461a.

Is it right, that this meters have TC of 1ppm/k by itself? So, that we can only measure not better than that TC with it?

Best regards

Well, at first, the '461A and the '465A probably have the same topology of their reference around the LM399, and also the ADC circuit should be identical.
If you look into the specification, both instruments are nearly identical in DCV, without the ACAL feature of the 465A.

The 465A may have a bit more stable reference , 30ppm/yr. over 35ppm/year, due to a possibly better selected LM399.
The T.C., though, is identical, i.e. 5ppm/K. That is due to the FineLine resistor network, which is used around the reference to generate the diverse ADC reference voltages, and in the ADC itself. See 34411A schematic, which is probably identical.

The LM399 itself is specified 1ppm/k, so the '465A can autocal itself, which gives 2ppm/K. But even that is of limited use, when making T.C. measurements in strongly varying environment.

btw.: d-smes complained about the assumed bad quality and high T.C. of ´his new KS '465A, but he obviously measured 1ppm/K in his experiment, which is very good, compared to the specified 5 ppm/K. I still do not understand his criticism, one can't expect more from that class of instruments, and also I think, that the new DMMs are even better than the old 34401A.

Anyhow, it's very difficult to determine the LTZ1000s T.C. by an instrument, which is specified to have 100 times higher T.C., but it's not impossible.

At first, you may also estimate the T.C. of your '361A by means  of your LTZ1000 circuit, as d-smes has done it. As a proper designed LTZ1000 has at least < 0.1ppm/K, you would have a first guess by variation of room temperature, because the 361A is expected to have a factor of 10..50 higher T.C.
Maybe you also see 1ppm/K only, like d-smes instrument.
Your basement may offer much more stable temperatures, I achieve +/- 0.1 °C over many hours there.
So you could monitor the 361As internal temperature. I don't know if it has that parameter available over the bus. otherwise you'll have to mount a thermometer inside the instrument, and then you might be able to detect changes on the order of 0.1ppm.

If you change the LTZ1000s temperature over +/-5°C quite quickly, that would be 100 times more than the 361A, you might be able to estimate the LTZ1000 T.C.

You will notice some correlations to amount and direction of both temperature changes, so it might be possible to assign these correlations either to the 361A, or to the LTZ1000. If you have a first rough estimate of your LTZ1000 circuit, you may use it as the new baseline for further measurements.
You may also try to improve the LTZ1000s T.C. by trimming the T.C. compensation resistor.

If you are lucky to have a 2nd LTZ1000 circuit, you can bootstrap these T.C. measurements, and T.C. trimming by the same scheme as above, but your 361A would only be used for relative / ratio measurements further on.



I use my 3458A for that purpose, that may seem much easier than using a 361/365A, but it's not.
It has a (measured) T.C. of 0.5ppm/K (as specified), but that's only a factor of 2 (maybe) better than the 6 1/2 digits instrument.

I estimated my LTZ1000 circuit to have a T.C. of about 0.02 ppm/K, as that you cannot distinguish any more between both changes in temperature.
So this trimmed LTZ is at least 10 times more temperature stable than the 3458A, and will serve as the new baseline for the next LTZ circuit, I just have assembled. Goal will be to trim the next reference to ~ 0.01ppm/K, and afterwards to determine the T.C. of the first LTZ with much higher confidence.

Frank

[Andreas]

0.05 ppm/K is a typical value, not maximum.

Not even a typical value.
In older data sheets: "0.05 ppm/K can be reached".
in the newest version: "0.03 ppm/K can be reached"
so it is a best case value.

Without leg length trimming and T.C. trimming I had up to 0.23 ppm/K
and was lucky to trim it down below 0.05ppm/K over 30 deg C for LTZ#3-LTZ#6.

And I fear that there is also some kind of hysteresis which limits T.C. adjustments.

Try to build several LTZ references: in the 100mV range of your DMM you can usually measure
with a factor 10 higher (noise free) resolution than in the 10V range. (as difference between 2 references)

With best regards

Andreas

[kj7e]

I do appreciate and I'm learning from all the experts experience, advice and opinions, that of TiN, Kleinstein, Dr. Frank, Andreas, MisterDiodes and others.  Thanks Guys!

Just because no post is worth reading with out a photo, here are a few from some recent tests;

Cycling the KX board between 25.0 and 35.0 Deg C, shows 6uV swings


10 hour measurement - 100 NPLC + Math 50 reading average, KX Reference on battery power at 35.0 Deg C, 34465a at initial cal temp + or - <1 Deg C.   Less than 2uV deviation.


23 hour measurement, extension of the above 10 hour - at 12 hour mark (mid scale) I switched the KX Ref to external power which introduced a slight amount of noise that corresponds to the room A/C cycling


So as long as the room temp is somewhat stable, so is the 34465a.  It also seems the there is some merit in stabilizing the temperature environment for the LTZ1000A, at least in my case.  Also, battery power is hard to beat when it comes to noise.

[IanJ]

Hi all,

Thought I'd join in the KX build fun as I've had a couple of LTZ1000CH's sitting here since 2015........Time I did something with them!

Will build up two of them over the next few weeks. The plan is to get one of them powered up 24/7 along with one of my PDVS2's and logged via my 3458a.

Ian.

[hwj-d]

Regarding the reading shift from touching DMM, I too have noticed this.  I usually change grounding around and/or tie floating DUT to meter case or earth ground.  After a bit of experimenting, I can usually get the touch artifacts to go away.

In the meantime, i cut pcb grounding to DUT-case. With floating ground, i didn't got this capacitive caused dips anymore. 

[kj7e]

Brand new LTZ1000A showed up from Digikey today, second KX board now running.  Really took my time trying to get every part soldered in place perfectly on this one. 



Decided to place the chip 0.5mm above the board;


Going to let it burn in for a few days before I do any testing.  Running just fine at 7.161634v drawing 27mA, 15v supply.

[TiN]

Well, after all, your KX'es belong to me
Decided to give a helping hand to CalMachine, with his two attempts.



Obviously, my design constrains are bit unfriendly to newcomers, especially tricky SMD film capacitors.
Once I fix these two modules, I'll log them to establish stable voltage levels and send back to CM for his calibration vs their 732B standard.

That will help to also help to keep my lab calibration uncertainty as "reverse-calibration", as my last Volt calibration was done in February 2017, and is overdue.