KX Reference

[Dr. Frank]

BFX
... All measurements done over 3458A in stable room temp (+/-1C), so own DMM tempco nulled out....

Except MX1 ref data, which was tested in similar box, but controlled by ILX 5910B controller instead. Meter is 3458A as well.
MX1 reference definately needs more work, so stay tuned 

Hi Illya,

I went through your different LTZ1000 - T.C. - measurements, but maybe overlooked the decisive hint..

How do you cancel the 3458As own T.C., of about 0.5ppm/K?

If your room temperature is varying +/-1°C, then you will definitely see that variation overwhelmingly in the KX reference measurements.

The TEMP? function also deliver 1/10°C degree of resolution only, so it's quite difficult to numerically equalize temperature changes of the 3458A, when you want to measure DUTs T.C. below 0.05ppm/K.

Frank

[TiN]

Hi Dr.Frank,

Well, hint for that is contained in my old KX datalog. You can see two runs with same config there, one with +24.2 +/-0.1C ambient temp, and second one with temperature raise to +27.2 by 3C. By using LTZ1000 output under test as a reference one can establish DMM's tempco in order of +0.1 ppm/K, so it's better than 0.5 ppm/K you mentioning.

Most of CM's KX run are with ambient temperature varied only by 0.2-0.4 C, not even full 1C. Also since we don't care about comparing absolute value from one run to another, absolute errors can be ignored.
Another confidence check is using both ramp up and ramp down curves from DUT to calculate tempco correlation. If I'd see different magnitude or output voltage not converging in the same temperatures, that would be a red flag. Of course, I would not give a guaranteed certificate on any of above listed tempco measurements, but for relative comparisons and hobby-level use I'm satisfied with numbers.

So in the end, temperature exercise on DUT is at least 70 times larger in magnitude than ambient temperature change, and meter's tempco have non-significant effect on the correlation result.
To further illustrate the point, take a look on this plot from our fellow member's lab 3458A setup. This is good example how ambient variations, like AC cycling affects LTZ DUT measurement (which is in TEC box, controlled by ILX 5910B). Yes, there is significant ~0.8ppm wobble, but you still clearly can plot a linear trend, without too much error from +2C ambient change.

[TiN]

And since you asked, here's today jem.



This is A9 STD reference, unmodified.

And updated summary LTZ tempco's, all together.

[TiN]

Since Andreas asked, here is how it looks like when LTZ reference is out of oven thermal margin:



I'm testing two references with zeners in sockets, same TEC box, two meters.

[Dr. Frank]

And since you asked, here's today jem.



This is A9 STD reference, unmodified. Single digit ppb/K tempco till +47C box temperature !

And updated summary LTZ tempco's, all together.

Sorry Illya,

but I suppose, that you have a fundamental bug in your calculus of the T.C.

You obviously want to calculate the T.C. each 1°C step, by calculating the derivative delta(U)/delta(T), where delta(U) is the change of the reference voltage over 1°C, already expressed in ppm of this reference voltage, as this is already divided by the baseline reference voltage, and delta (T) is the change of the temperature, that is identical to 1°C.

So, the column 'ppm dev' already gives your 'momentary' T.C. value, but in your column 'T.C.' you again (for the second time) divide by the reference voltage 7.184V, which gives insanely small values, because of the quadratic division in Uref.

I also see a numerical problems with your calculus of the T.C., as the measurement of the reference voltage has a typical standard deviation, or reference rms noise of 0.02 .. 0.04ppm already, if you measure at NPLC100.
This noise is not reflected at all in your T.C. curves; they look smooth as a babys butt.. which should you also make you think, that something is fishy here..   

For sure, your measurements of simply Uref over T look exactly like mine, that is with a lot of overlayed noise, and then you would have to use a reasonable smoothing algorithm, or a graphical method, i.e. box method, to get the correct T.C. numbers, or function of T.C. over temperature.

Regarding the T.C. of your 3458A, it would be great if it would really have 0.1ppm/K, but I would also like to see a reliable and solid measurement, assuring that parameter.
 
Frank

[TiN]

Aha, I hear what you saying now. Perhaps I got over-excited there a bit, and got too much zeros.

I acknowledge this calculation topic have icebergs under, perhaps good time to establish common definition to avoid further confusion?

Here's same plot, with actual every data point over the temperature.



Dashed line is step approximation fit, which is not subjected to noise and smooth.
Here also see typical 0.1ppm-ish noise derived from 3458 & DUT.

Is all this look right?

On second part, 3458 tempco, it's great time now, so I'll stabilize temperature of the DUT (2 x LTZ1000) at let's say +24.000 +/-0.005C and variate ambient temperature from current +28 to +22. That should give us 6C of 3458A temperature change to estimate DMM TC?

[Dr. Frank]

Illya,

that all looks much more reasonable and solid.
Anyhow, now one can really see the good quality of your measurements, in relation to the reference noise.


I don't know, where to find an official definition of this 'box method'.
 
As implemented by Vishay for the determination of the T.C. of their resistors, they probably use 3 temperature points only, like -55°C, +25°C and +125°C.
Their German representative once measured my VHP202 at 25, 60, 100 and 125°C only, and calculated the average T.C. between these points.

G.R. measured the new 120 Ohm resistor samples between: 23=> 35 =>45 => 35 => 23°C.
They let the DUTs settle for about 24h at each single point, before they take the measurement.
However, they got totally puzzled values, even with different signs for the T.C., and that was due to this big hysteresis of this special value.


Therefore, our method of continuously measuring the reference voltage en passant, i.e. while the temperature slowly changes, and then to draw a box around the horizontal and vertical limits of the measured graph, gives much better and more information about the DUT.
You only have to take care, that not too big a hysteresis or shift occurs, which often happens, when the temperature is changed too fast.
In a physical sense, the measurements have to be made in or near the 'equilibrium state'.. and it's always a problem to identify this correctly.

This box method is a kind of averaging, and necessary only, when a non-linear or hysteretic behaviour is present.

Yep, if you were able to change the temperature of your 3458A by 5°C, you'll get a real good measurement of its T.C. for DCV, w/o ACAL.
I assume, the root cause of this T.C. is located mainly in the A/D ASIC, i.e. these internal gain resistors, which you were very familiar with, I think 
This famous cal constant 72, related also to the AN18 stability problem, should directly depend on the environmental temperature (which is not considered in the AN18).
What do you think?

I have measured this CAL? 72 regularly, but could not find a real drift over time (fortunately), but a negative dependency over temperature (TEMP? CAL? 175), of about -0.3ppm/°C.
The graph over temperature has a lot of scatter, because the temperature sensor is on the opposite site than the A/D board.
But these -0.3ppm/°C are in good accordance with the DCV T.C. of +0.4ppm/°C, which my wife measured for my 3458A..
The difference in value might to be explained by the T.C. of the internal LTZ1000A  reference.

Frank

[TiN]

Yes, hysteresis is the reason why I use slow ramp up/down for all TC data collection, instead of step gaps.
With Raspberry Pi and GPIB it's all easy to program.

For meter tempco data, be my guest : https://xdevs.com/kx56_test2_tc/
It's current live log. I'll leave it running as low as room AC can go.

Beginning of the graph is with ambient temp +29c (cyan dotted line)
TEMP? readings from both meters are bold blue/violet. One of my meters does not have A1 shield inside, hence RAW TEMP? data is offset by 4.9c. External covers are all on.
DUT reference are two LTZ modules, in own TEC box, controlled at +24C +/-0.02C peak (black is TEC box sensor temp, thermistor YSI 44006).
RAW CSV-datafile is in link under graph as well.

From what I can see - DCV 10V tempco is un-measureable here (<0.1ppm/K). Even if I take worst peak-peak 0.5ppm (dots are individual samples, lines are average) and divide by current TEMP? change (42C -> 38C = 5C) that gives 0.1ppm/K for blue meter. Second "B, green" meter shows only noise, no tempco.

If meter own TC would be 0.1ppm/K, then output reading should be already off by 0.5ppm, which is clearly not the case, if we go by averaged "noise-free" values.

For reference, if that worth anything, "blue" meter has new A3 Keysight ADC made in 2016, bought year ago, "green" meter has A3 Agilent ADC, made in 2004.

[Dr. Frank]

By the way, I think it's instructive to refer to this old NBS Technical Note 1239, about the performance of solid state voltage references:
https://archive.org/details/solidstatevoltag1239fiel

In Design / Performance Goal P4, a T.C. of < 0.01ppm/K is recommended, which obviously none of the commercial references achieves..

A description, how to correctly measure and display that T.C., is also not available.

Frank

[Dr. Frank]

Wow, both your 3458A are really impressive!
Their T.C.s are practically zero.
But it's really hot, where you live.

[TiN]

Outdoor temp today in here north of Taiwan is around +18C. I just had cold last week, so I did not turn on AC last week or so.
I don't have luxury of basement as of yet, it's apartment on 10th floor, with gear running 24/7, so ambient temps easy to climb 28-30c without aircon permanently on.

I think we saw enough on 3458A's tempco question, time to test some other's 
3458A Green = +0.048 ppm/K 10V TC (0.36 ppm with 7.4C drop).
3458A Blue = +0.06 ppm/K 10V TC (0.45 ppm with same drop).

[bopcph]

@CalMachine

If you could get a smaller dome to fit a few mm inside the one you are using now.
You would create a dobb. wall dome decreasing the heat transfer thru the dome(s) and thereby also reducing the risc of turbulence/convection.

Drilling away the free areas around the LTZ, inside the guard-ring reduces the mass of FR4 material, it should make it easier for the LTZ to get the mass in equilibrium - just a guess :-)

[TiN]

Tested good old jumpy ACH. Funny enough, other than jumps it's fine ref.
It's reference of 0.6 ppm jumps. .

[MK]

with the jumpy ACH perhaps you could try a set of different currents, at a bit higher current those jumps may go away?

[beanflying]

Chasing Zero's and accuracy/stability is addictive 

Boards ordered. If anyone in Oz is interested I am going to keep two but one will be available for cost.

Now off to arrange components 

Freaky stable untrimmed Max6350 @ 21 C but I "NEED" more   

[kj7e]

I was not entirely happy with the tempco of my KX reference with the LTZ1000A, it wasn't that bad, but I felt it should have been better.  The best I could test with a 34465A was about 0.08 to 0.1 ppm/K.  Since I had the A version I did as the data sheet said and omitted the 400K tempco resistor.  Just as an experiment I added the tempco resistor starting with 1M Ohm, then 680K, 470K and 300K.  Each time seeing an improvement and finally seeing a nearly flat tempco with 300K.   I even tried 270K and 330K but found the sweet spot at 300K, what a difference! I cant say how pleased I am now.  Since that time, I've added a Keithley DMM7510 and I'm able to further verify my findings.  Here is some data from a 15 Deg C to 35 Deg C temp sweep;

The meter was null'd when the reference temp was 15 Deg C in order to easily see the uV deviation.  The bump about mid scale is where I powered on the oven to continue warming from about 25 Deg C to 35 Deg C;




Zoomed in to show some of the noise and individual data points;


So there is about a 3uV (-1 to +2) change across the sweep, this was repeated about 5 times with the same results. 3uV = 0.42 ppm / 20 Deg sweep = 0.021 ppm/K.

This is on the second KX board I built, the first one (which had a better initial tempco) also improved significantly but sweet spot was 390K.

[TiN]

Alright, both CalMachine's references are now repaired, tested, tweaked and ready to be shipped tomorrow.

I've packaged them into metal enclosures, with Low Thermal binding posts for outputs and regular binding posts for input power.
Final calibration test data and tempco plots for module 2:



 

And time scale plot of the same data set.

Red here is module 1 measured in ambient air, using K2002, powered by VRLA battery.
Green and blue is module 2, on 3458s, measured with 18-50c ramps in TEC box chamber.

[Pipelie]

TiN
Well done on your TC trim. 
I got some data, would like to share.
test gear:
1.HP3458A
2.DIY 16 channel scanner
I always use scanner for TC test, and some times need my DIY thermostatic chamber. 

[TiN]

It seems like Mr. CalMachine does not want to give me shipping info at ticket 119 in time, so these two little boxes will go out tomorrow to person, who gives shipping info. 

[pitagoras]

PM sent

[Pipelie]

Hello TiN,
Did you test the Voltage Noise of LTZ1000 and 10V ?
I think it would be useful and interesting data  worth of collection.
the TC of LTZ1000 could be trim in most case,but noise performance is hard to improve. 
here are the noise performance of my three 732a:
1.  2uVpp,
2.  1.6uVpp
3.  1uVpp
and the noise of LTZ are various too.

[Dr. Frank]

Hello TiN,
Did you test the Voltage Noise of LTZ1000 and 10V ?
I think it would be useful and interesting data  worth of collection.
the TC of LTZ1000 could be trim in most case,but noise performance is hard to improve. 
here are the noise performance of my three 732a:
1.  2uVpp,
2.  1.6uVpp
3.  1uVpp
and the noise of LTZ are various too.

What is the time constant, over which you determined these noise numbers?
on the order of seconds to minutes, you may get the basic zener noise, when you have 1h, for example, you will also measure some popcorn noise .. Allan deviation is probably the best way.. and I did not see such for 732A / 732B, yet.
Would be quite interesting

Frank

[Pipelie]

Hi Dr. Frank,
I use my DIY 0.1-10Hz   low frequency noise meter to test the noise of voltage reference.
I‘av build more than 35 unit for a group-buy, here is what it looks like.
https://www.eevblog.com/forum/metrology/diy-low-frenquency-noise-meter/50/     Reply #68 by zlymex.

and some test result, hope this data could answer your question.
PS. I don't have the suitable equipment to record the output of noise meter for an hour or so.

[beanflying]

Santa arrived late and he forgot my LTZ's and Vishay's. Maybe I wad a bit naughty 

Hurray up and wait some more .....

[kj7e]

Finally installed my first KX board in a proper enclosure, used a Hammond 1590B with some Pomona 3770 binding posts on the output and 3760's for the input.

Sorry, the photo orientation seems to be amiss.







The board is recessed in the foam, only the edge of the board makes contact with the foam surrounding;


The foam on the lid then compresses to gently sandwich in the board in the recessed area;


This seems to work very well and should not present any stress on the board as well as add some thermal insulation to minimize external temperature effects.