Ultra Precision Reference LTZ1000

[Andreas]

 Therefore all high quality / precision resistors as Thin Film, Wire Wound and Metal Foil types have similar specified ageing rates of 20...35 ppm/yr., (typical or maximum).

That is valid for the first year with decreasing ageing rates in the following years.
And ageing can be further stabilized by some burn in. I am currently ageing my Z201 resistors with around 100mW (1/3rd the precision rating) intermittend 1,5 hours on 0,5 hours off at room temperature to stabilize the resistors.

With best regards

Andreas

[Dr. Frank]

That is valid for the first year with decreasing ageing rates in the following years.
And ageing can be further stabilized by some burn in. I am currently ageing my Z201 resistors with around 100mW (1/3rd the precision rating) intermittend 1,5 hours on 0,5 hours off at room temperature to stabilize the resistors.

With best regards

Andreas

Hello Andreas,

I do not recommend a burn-In on Z201 resistors, neither does Vishay!

If you have the molded types, ageing is mainly determined by the oxygen and humidity reactions on the resistor active area. That you cannot influence by elevated temperatures.

If you have the hermetically sealed type, you will create a big ageing rate, as specified by Vishay by the parameter: "Load Life stability", that's about 50ppm @ full load, 20ppm @ 100mW (Z203).
If you simply leave those components as they are, drift is 2ppm/6yr only!

Compare those parameters: A burn-in makes no sense, obviously, if you use them in shelf life mode afterwards, i.e. with very low power dissipation.


Additionally, there's another serious problem on metal foil resistors:
They show a very pronounced hysteresis effect, if they are exposed asymmetrically to high temperatures and brought back to room temperature only.

I have not yet discussed 'conditioning of the LTZ ref. circuit'  (~ chapter 6),  but the other two VHP202Z resistors were re-measured by Vishay, @ 125°C and came back with a +5ppm shift. One of them creeped back to its initial value, but that took over one year. The other resistor did not creep back by itself, instead  I made a temperature cycling on it, analogous to degaussing, and was able to "reset" the hysteresis to < 0.2ppm of the initial value.


Burn-In normally is used for early failure detection (accelerated life test) on "ordinary" parts and assemblies, but I think, burn-in on  high precision devices is not always the right way.
That's also true for the LTZ1000, as I will demonstrate later.

Frank

[quarks]

Still I believe 10V are very uncommon today, just usable to check your DVM/DMM in one range but nothing more as most circuits require a reference voltage of 5V or less.
Sure, it's all about stability, but wouldn't it be worth having a LTZ1000 based voltage source with all common voltages of todays need? On the other hand, wouldn't it be worth to have a decade voltage output to verify all voltage ranges on your DVM/DMM instead of only the 10V range? Mh...

With high precision gear (6.5 to 8.5 Digit DMM and Calibrators and especially traceable NIST/PTB calibration) all DC measurements are accuracy wise directly related to a 10V Standard Reference. Also with all gear using artefact calibration it is a must have. So at least for me it is the most common/important DC value to have/know.

All other DC values are delivered through a calibrator or as described by Dr.Frank with KVD/Reference Devider/Ratio/transfer measurement.

[branadic]

Wow, a whole bunch of german guys in here

With high precision gear (6.5 to 8.5 Digit DMM and Calibrators and especially traceable NIST/PTB calibration) all DC measurements are accuracy wise directly related to a 10V Standard Reference. Also with all gear using artefact calibration it is a must have. So at least for me it is the most common/important DC value to have/know.

All other DC values are delivered through a calibrator or as described by Dr.Frank with KVD/Reference Devider/Ratio/transfer measurement.

This is like stating "We've done it always this way". As you know the Josehpson standard is everything but 10V, the value is reached by connecting serveral Josephson contacts in series. The definition of a 10V standard therefore is confusing itself. So the question might be is 10V still up to date?
Nothing we can really answer as there are enough people out there in the wrong positions stating the same as you do (We've done it always this way.), but please allow for this question.

BTW: SI unit uses 1 volt per definition, not 10V

[Dr. Frank]

Wow, a whole bunch of german guys in here



This is like stating "We've done it always this way". As you know the Josehpson standard is everything but 10V, the value is reached by paralleling serveral Josephson contacts. The definition of a 10V standard therefore is confusing itself. So the question might be is 10V still up to date?
Nothing we can really answer as there are enough people out there in the wrong positions stating the same as you do (We've done it always this way.), but please allow for this question.

BTW: SI unit uses 1 volt per definition, not 10V

Well, Germans (e.g. Wernher-von-Braun) have put the first man on the moon, or not? 

The SI definition currently is indirect, i.e. the VOLT is not one of the basic units.
The SI Volt uncertainty currently is not better than 2e-7 (e.g. volt balance), i.e. an HP3458A and a well designed LTZ 1000 reference are totally sufficient.

I'm waiting desperately for the redefinition of the kg, by a Si sphere, and/or the Watt balance.. then, the Josephson Volt and the von-Klitzing-Ohm will define those units directly and in accordance with SI with much smaller uncertainty.. up to  1e-16, or so.
Then it'll be the time to scrap the analogue DMMs..especially the 3458A. 

Frank

[branadic]

So, to answer your question, "10V" is the more modern cardinal point for this purpose.

Sorry, but this answer doesn't make me happy at all. A 10V based definition of a SI unit that is made by series connection of smaller value voltages can't be accepted by any physicist. It's the same stupid thing as nV/sqrt(Hz), this also makes physically no sense. It would be worth talking in a more physical language, not matematical.
The primary kilogram is what it is 1kg, not 10kg made by 33 parts à 303g. Got me?

[babysitter]

Thank you Dr. Frank for some real live experience and interesting explanations.

Whatever the output voltage is used for, in my eyes this thread is for discussion about care & feeding of LTZs, and together with some reading of the volt-nuts archives and discussion with Dr. Frank gave me an idea how to care and feed mine. Intentionally not going to the end in some directions, like selecting-from-batches the resistors and the LTZ itself due to self-inflicted money limits I didn't even try the high-end resistors but settled in the sub-10-eur-each league.

My reference was made for travelling (its predecessor wasn't stable when doing a round-trip from work, where the calibrated instruments are, and home.) And it successfully made it to Frank who inspected it further and has shown a quite good behaviour. For long-term experience it is just too young, but will go back to Frank for that. Also it will go to some other (hello quarks, your time will come soon! Lets have a Cali-barbeque with this weather when I bring the LTZ! )  people just to collect experience and comparisons, and help giving a idea about the SI volt from some of them who are PTB-traceable to others who aren't (yet?). At home it will take some beating, being a transfer standard and part of a calibrator but also supplying sensors and some fooling around with oscillators and how they perform with a good voltage source, but also spending time off-line. So far I am happy with the results.

But sometimes I see men riding on black horses burning down metrology labs when I look at this thread and it goes into a certain direction...  even metrologists know the art of multiplication and statistics, so why not 10V out of 30K Josephson elements? They even are happy with cesium clocks that probe more than a single but a bunch of cesium ions. And they swallow it that this bunch of atoms isn't even standing still but wobbling around in space and temperature range, giving some nasty effects.

I intend to send my reference to somebody off-forum today, then again to Frank and from there it will go to quarks.
Gotta go, our trainee will etch and populate his second PC today, and test... a LM399 Source as described a few posts above !

[branadic]

Anyway, it doesn't matter why 10V is the "de facto" standard cardinal calibration point, it just is, and we will all have to live with it.

Please allow for other opinions instead of yours. If you can accept this concept, fine.

Oh-- and I still don't see what this has to do with the original subject of this thread.

In this case I quote you: "... it just is, and we will all have to live with it..."

I intend to send my reference to somebody off-forum today, then again to Frank and from there it will go to quarks.

Can offer you to visit Stuttgart, several calibrated test gear is available Keithley 2002, Prema 5017, 34401A and my private Prema 5000.

[quantumvolt]

So, to answer your question, "10V" is the more modern cardinal point for this purpose.

Sorry, but this answer doesn't make me happy at all. A 10V based definition of a SI unit that is made by series connection of smaller value voltages can't be accepted by any physicist. It's the same stupid thing as nV/sqrt(Hz), this also makes physically no sense. It would be worth talking in a more physical language, not matematical.
The primary kilogram is what it is 1kg, not 10kg made by 33 parts à 303g. Got me?

I don't accept the kilogram because "kilo" is Greek for 1000 and  don't ing want one thousand small  grams. Give me ONE gram 

[Jay_Diddy_B]

Hi,
I believe that 10V was chosen as the standard for voltage calibration to minimize the contribution made by thermal emfs.

This is quite different than the SI unit system where the unit of measurement is the volt.

In 1990 the volt was redefined, using Josephson junctions. Prior to this there was about 1.2 ppm difference between the North American Volt and the European Volt.

I have some old HP3455 meters with a green sticker with 1990 inside a diamond. This means the meter was calibrated with the 'new' volt.


Jay_Diddy_B

[nukie]

10V is easy to calculate on the resistor divider such as the Fluke 720A

[Dr. Frank]

Dear fellow-nuts,

please don't struggle about the absolute value of volt standards:

there's always an abstract definition of direct and derived SI units, and in most cases a totally different way to practically realize the unit. (called 'mise en pratique')

For the volt, the official SI-definition from 1948 is:

"The (SI) volt is the potential difference between two points on a conductor which carries a constant current of 1 Ampère, if the dissipated power between those two points equals 1 Watt"

You see, that this academical definition really is  "1 Volt" (based on kg, m,s).

(The kilo-gram is the only SI unit which is based on 1000 units of a gram, for practical reasons of its realization also, i.e. the kilogram artifact in Sèvres, France)
 
The realization of the SI Volt is done either by an Hg electrometer, by Clothier at al (1989) basically a balance, measuring the electrical force, uncertainty 0.27ppm, or a design by Funck et. al. (1991), measuring the force on a capacitor plate (0.31ppm uncertainty).
Both experiments are very complicated and clumsy, and might have been performed only once.

The output voltages of those primary standards (SI) can be different from 1V, given by the practical setup.


Then, there is a different representation of the Volt, that is a more easy-to-manage-way to deliver the volt.
Currently this is the Josephson Volt, a few mV in 1972 were amplified by a cryogenic divider to ~ 1V, 1V in 1985 by a Josephson Junction array, and 10V in 1987 by a longer array. (based instead on 2e/h)

Remark: The definition of the Representation of the Volt has been redefined in 1990, but not the definition of the Volt in the SI. The definition from 1948 is still valid!

The uncertainty between two different Josephson experiments can be as low as dV/V = 3e-19 for single JJs and  1.2e-17 for 0.6V arrays if you compare directly on the cryogenic (i.e. quantum) level.

Remember: The uncertainty between the Josephson Volt and the SI is still 0.4ppm!

10V for secondary reference standards has been chosen only for practical reasons in the analogous world (in contrast to the cryogenic world),  reasons are: the typical offset  of several µV, the limitation to measure volt differences to a few nV only, and because the typical cardinal points 1kV, 100V, 1V and 100mV are symmetrically situated around 10V, so that a 100:1 and a 10:1 divider is sufficient to transfer 10V to all of them.

Anyhow, if one selects a different standard value, e.g. 5V or 7,147V, it's also ok, as there  will be always the necessity to make a transfer to other needed calibration points (by KV divider, or by 3458A).

Therefore, the foregoing  discussion, which calibration point is the 'correct' value is simply a lack of knowledge of the concept of SI - 'Definition', - ' Realization' and - 'Representation'.

I recommend the lectures 'école de physique, Les Houches, “Quantum Metrology and Fundamental Constants” ', Blaise Jeanneret, "Volt metrology: The Josephson effect and SIS junction arrays":

http://www.metas.ch/LesHouches/downloads/talks/15_Jeanneret.pdf.

Frank

[Andreas]

And finally ....

10V is the voltage which can be handled by +/-15V supplied OP-Amps without input voltage dividers.
Thus giving a nearly infinite input impedance instead of the 10 Meg in the other ranges.
10V is the normalization voltage of analog calculators.

With best regards

Andreas

[SeanB]

I don't accept the kilogram because "kilo" is Greek for 1000 and  don't ing want one thousand small  grams. Give me ONE gram 

I do have a calibrated masspiece traceable to a national standard ( it is only 2 levels away from the standard SA kilogram, which is replica 56 of the International standard kilogram. It is 1.000g, with an error of 0.2 mg on that mass. Has held that over a few calibration cycles, as it is rarely used, only being used on a strain wire massmeter that has a resolution down to 0.1mg. It also has a 24 hour warm up time to be stable to that though, and you need to close the doors and not breathe, as it can detect that, and looking in through the doors will show up from the IR radiation you emit. I use it to check linearity across the measuring range, along with the 20,10 and 50g masspieces in the set. I actually had them all calibrated, even the 0.5mg masspieces.

[Andreas]

Hello,

just to bring up some discussion points to the original theme, I want to say some words to my design.
(see attached cirquit diagram below).

The main idea was to have a portable "transfer standard" which can be shipped hot.
To avoid mains disturbances it is also battery powered during operation.
So main concern is minimum current consumption.

This leads to a design with a LTZ1000A with around 50 degrees temperature setpoint.
Thus setpoint divider has 12.5K to 1K.

As resistors I choosed precision wirewound of type UPW50. Unfortunately the 12.5K and 70K values
where not available and had to be replaced by 10K+2K+0.5K and 50K+20k.

Further restriction: the whole cirquit should fit into a Euro-Card aluminium case.
For the first step it should be a unbuffered 7V output reference.

Cirquit description:

Power supply consists of 12 AA NiMh-cells.
The raw voltage of 17.5V (battery full) downto around 14V (empty) is stabilized with a low drop (0.17V), low noise (20uV), low power consumption (1mA) voltage regulator LT1763 down to 14.0V.

The reference section has some modifications to the datasheet.
R13 enables startup even with negative offset of the LT1013.
T2 removes the reference current from LT1013 lowering the self-heating of the LT1013.
Second effect: more headroom for the current regulation with low battery voltages.
R18 limits inrush current through the zener to about 14mA max during switch on.
R17 enhances stability of the current regulation loop since the load is now missing.

C11 + C12 are adapted from Datron cirquit. The effect is having lower noise on heater +
reference current regulation.
To further calm down heater current noise C13, C14 are added. C14 keeps RF noise away from the
negative input of the heater OP preventing the RF from being demodulated by the input diodes
which would give an offset. But this capacitor leads to loop instabilities which are compensated
by C13.
Similar C15 keeps RF away from current regulator OP. C8 together with the additional resistor

R19 compensate for loop stability.
C9 keeps RF from the output connector away from the reference.
C9 can only be added with R19+C8. The original cirquit from datasheet is not stable with
capacitive loads.

R16 is a NTC near the LTZ. So the temperature within the cirquit can be measured from outside.
The auxiliary connector J6 can be used for several things.
- changing the temperature setpoint (Pickering patent)
- heater monitoring (environment too hot/cold) or influencing
- buffered output (or voltage divider cirquit as calibrator)

J1 is a D-Sub connector which I use as main output. The advantage is that neighboured pins are nearly on the same temperature giving low thermoelectric voltages. The metal shield of the connector further equalizes the temperature of the pins.

J4 + J5 are auxiliary outputs.

Mechanical description will follow...

With best regards

Andreas

[Dr. Frank]

If the LTZ1000 and the other buried zener references SZA263 and LT-FLU potentially are stable to 0.2ppm/yr @ 45°C, including 0.02ppm/yr. by usage of hermetically sealed resistors as in the Datron 4910, why are they all specified no better than 1.5ppm/yr? (See picture 1.)

The main reason is, that 10.000V are generally used as a reference, not the raw 6.6 …7.2V of the reference amplifiers, and that those 10.000V have to be generated by resistive amplification.

The error calculus of such a resistive divider is given in picture 2

=> Output Stability (10.000V) = 0.52 *[(TC(R1)-TC(R2)dt +2*|AC(R)|dT]

A T.C. matching of R1 and R2 is required, so that the T.C. effect is nearly cancelled.

The ageing rate of both resistors cannot be matched, so their individual ageing rates add up instead.

Therefore ultra time stable (statistical resistor network in the 7001) or pre-aged resistors (wire wound, sealed types in 732B) must be used.
They still will contribute several tenths, or up to 1 ppm/yr.
Two hermetically sealed VHP202Z would contribute ~ 0.7ppm/yr.

Therefore it’s clear, that the 10.000V reference output is much less stable than the reference amplifiers itself.

The adjustable decade divider in my design will contribute 0.14ppm uncertainty only, and can be recalibrated at any time; therefore the basic stability of the LTZ circuitry is maintained in the 10.xxx V output.

To get plain 7.000V (and 10.000V) from e.g. 7.176V, an additional attenuation of 0.975 is required (see schematic of my design).

Therefore R2= 52k, R1=1k31=>

Output Stability (7.000V) = 0.0245 *[(TC(R1)-TC(R2)dt +(|AC(R1)| + |AC(R2|)dT]

All instabilities of this divider are attenuated by a factor of 40, i.e. T.C. < 0.05ppm/K without matching and A.C. < 0.02ppm/yr with VHP202Z are possible.


If the ultra linear HP3458A would be used to adjust the 7.147 => 10.000V transfer by a  non-inverting amplifier, an uncertainty of < 0.05ppm of the 10.000V output may be achieved.

(to be continued)

[Dr. Frank]

I *think* the drift specs are *very* conservative figures for the first year-- all of the references listed "calm down" after a year (or a few years), and their real drift rate is far less-- and yes, I agree with you, most of this is due to the surrounding circuitry (especially the resistors) that control and/or condition the Zener device.

There was an interesting article on the Vishay Precision Group website about resistors in a hermetic network-- the gist of it was that the resistors in this case, (if they came from the same production lot, and were trimmed by the same person using the same equipment) will have a tendency to drift together over a long period of time-- for a divider (which we are mainly concerned with the ratio, and not the absolute value), then the long term drift can be quite small-- or at least that's what the article *says*-- you have to consider the source-- they are trying to sell resistors, so of course they will not show any data that make their resistors look bad-- all you see is the good outcomes.  You can find the PDF here:

http://vishaypg.com/doc?63512

I do like the architecture of your design-- I have stolen some of the concepts for my own use-- I hope you don't mind...

Well, thank you for the compliment.

I intended to share for copying - so I don't mind at all ..

But I also wanted to initiate some further reviews about the other "gimmicks", I have summarized in my first post.. for reuse in my own redesign..
(Several LTZ1000 directly from LT arrived just by today.)

The documents of Vishay have to be read very carefully, the "typical" parameters are sometimes very optimistic, but they always hold the maximum boundaries, which are mostly impressing, anyhow.
After all, if you dig in the last few tenths of ppm of your design, uncertainty by design or by specification is rarely possible, instead you always have to monitor your finished reference and select the most stable one as your "golden device".
That's what Fluke et al are doing also.

There was a classical document (1) by Fluke, where they monitored many different 732B over years.. and there was no sign of decrease of the ageing rates. The specifications are also quite realistic..
No, in contrary, for the 7001, even a linear drift prediction was specified, which might improve the uncertainty of this reference.

Only heavily drifting references like the one in the HP3458A may calm down, but only because they are operating far away from an equilibrium state.

Frank

(1) "Predictability of Solid State Zener References", David Deaver, Fluke Corp.

[babysitter]

A important point with those stable references is the enclosure. As shown, mine is sitting inside a tinned steel case (solderable) which is housed in a bigger ABS box. The power supply is simply fed thru a hole in the metal box, the output jacks are going thru their own hole each.

This was my simple, cheap and lazy approach but I know that with a few euros and workshop minutes more might have been possible

If I really want to do it better next time (*) I would think about the following:
Discuss those things with me! I am bored if you dont

Use the same kind of solderable tinned sheet metal box again
Buy a batch of feed-thru capacitors
use only the feedthrus and hermetically solder the case
I would try to solder it in a nitrogen or co2 atmosphere, so the oxygen and humidity dont eat our resistors!

to get really esoteric, one could solder in two thin cooper tubes and use them to fill it with silicon oil (available as gear oil for RC cars), remove the air and have a good thermal conductor inside. pinch and solder the tubes to seal.

Also, did anybody consider sodium silicate as a coating to prevent moisture and oxygen reaching the resistors ?

[Andreas]

Buy a batch of feed-thru capacitors
use only the feedthrus and hermetically solder the case

Also, did anybody consider sodium silicate as a coating to prevent moisture and oxygen reaching the resistors ?

Hello babysitter,

if the feed thru C's are really hermetically tight (sealed with glass) the pins probably will be COVAR pins (the only material with the same expansion coefficient as glass) having around 39uV/K against copper.

So you will need at least additional thermal shielding for the feed thru C's.

For me another question is more important.
What is better: having long legs on the LTZ1000 or keep them as short as possible together with a slotted board.
The pins are also of covar (because of the hermetically tight case).
So we are having 2 large thermoelectric junctions on each pin.
One between the bonding wire and covar pin, the other on the pcb between covar and copper.
The junction from the bonding wire to the aluminium mask are also a junction.

And unfortunately the bonding wires are not equally distributed (see chip photo)
http://www.amobbs.com/thread-3593996-1-1.html
picture ourdev_464495.JPG

The heater pin (Pin 1) has 3 bonding wires (cooling the chip down at this edge) and the next 3 pins 8+7+6 of the temperature sensing transistor are also on the same side of the chip at the next edge.

The idea would be to find a geometry where all thermoelectric junctions (or at least those two of the Zener output) have the same temperature.

My idea would be the legs as short as possible and a slotted board having low thermal mass at the solder junctions. But since I have no thermal camera I cannot prove it.

With best regards

Andreas

[robrenz]

It seems like the participants here may have some applied knowledge of thermal EMF generation.

Below is my current understanding as a starting point for a discussion on something that I have been curious about. I am not assuming what I state here is correct.

The thermal EMF generated in a conductor is caused by a thermal diferential/gradient from on end to the other. the gradient does not need to be uniform.

Two parallel identical conductors joined at one end with a thermal gradient have thermal EMFs but they are equal and cancel.

By joining two condutors at one end that have different Seebeck coeficients you will get the differential of the two different Thermal emfs generated by the temperature gradient, a thermocouple.

If you drop the thermocouple junction 200mm deep into a heated nonconductive liquid bath there will be negligible thermal gradient in the junction or the leads that are in the bath and all the gradient will occur in the conductors from the bath surface back to the cold junction. The actual junction itself is not generating the emf

The thermocouple law of intermediate metals states a third metal can be introduced in the juction between the two main leads of a thermocouple and as long as all three materials at the junction are the same temperature the thermocouple will give the same ouput.

This implies to me that no thermal emf is generated just by the contact of two dissimilar metals. The Thermal emfs occur only in the thermal gradients of the two materials. The contact is only an electrical connection.

Even though the mathematics is over my head currently on the inter relationships of the Thomson, Seebeck and Peltier effects. It seems that the Peltier effect is not purely a junction effect but has gradient aspects like the Seebeck effect and joule heating also.

I bring all this up because I think the frequently shown Thermal emf charts for paired metal combinations are extremely misleading They imply that the mere contact point of two particular metals will generate a certain µV/deg. As an example copper vs lead tin solder is listed at 5µV/degC.  So if I take a 2" long piece of copper wire and solder it to a 2" long piece of lead tin solder and maintain a 1 degC temperature difference between the end of the copper wire and the end of the solder and magicaly measure the temperature across that without introducing additional thermal emf  junctions I would read 5µV.

But if I make a Thermocouple out of two identical copper wires and use lead tin solder to make the junction according to thermocouple law of intermediate metals the solder will have no effect on the thermocouple output and since the two leads of the thermocouple are identical materials the output will be zero, no thermal emfs generated.

Should we discuss this here or should I cut and paste this a new thread?

[Rufus]

So if I take a 2" long piece of copper wire and solder it to a 2" long piece of lead tin solder and maintain a 1 degC temperature difference between the end of the copper wire and the end of the solder and magicaly measure the temperature across that without introducing additional thermal emf  junctions I would read 5µV.

But if I make a Thermocouple out of two identical copper wires and use lead tin solder to make the junction according to thermocouple law of intermediate metals the solder will have no effect on the thermocouple output and since the two leads of the thermocouple are identical materials the output will be zero, no thermal emfs generated.

If there is no thermal gradient across the solder bit of the connection. 

How about you take a loop of copper and put a blob of solder on one side? I reckon that will make a thermocouple as long as the solder blob experiences some of the temperature gradient and you could argue there isn't any junction.

The solder effectively changes the seebeck coefficient of that part of the circuit to something between copper and solder. Probably in proportion to the relative cross sectional area and electrical resistance.

So easy to test I just spent 5 minutes trying it. Attached is photo of 2 22 swg tinned copper loops one with some 60/40 solder blobbed on one side. Playing a flame on the end of the loop produced more than 20uV at the other end of the one with the solder and less that 1uV (possibly nothing) with the other.

[eevblogfan]

hey


Flame ?  isn't that <700C ? 

at <700C that's rughly 29nV !

witch is 0.0029ppm/C  at 10V

and 0.029ppm/C at 1V

is that so bad ?

oh and another question : does High silver load of soldier helps ?

if so . can you check that somehow ?

PS , 20uV is too low to be sure you measured 100% of that affect , can you investigate further more and confirm us that it was 20uV ?  ( I know that even 5% if accuracy is enough for that test bt I am wondering if those 20uV are fully result of that thermo couple junction )

Thank you

[quarks]

just to bring up some discussion points to the original theme, I want to say some words to my design.
(see attached cirquit diagram below).
...
Mechanical description will follow...

Hello Andreas,

thanks a lot.
That is exactly the kind of info/input I hoped to see, when I opened this post. 
I hope this will lead to a well thought through optimized design.
 
bye
quarks

[quarks]

It seems like the participants here may have some applied knowledge of thermal EMF generation.
...
Should we discuss this here or should I cut and paste this a new thread?

I think this discussion fits very well to the subject.
But please feel free to open a new thread.

[quantumvolt]

"The voltage is not generated at the junction of the two metals of the thermocouple but rather along that portion of the length of the two dissimilar metals that is subjected to a temperature gradient. Because both lengths of dissimilar metals experience the same temperature gradient, the end result is a measurement of the difference in temperature between the thermocouple junction and the reference junction."

The "characteristic voltage difference (is) independent of many details (the conductors' size, length do not matter)".

http://en.wikipedia.org/wiki/Thermocouple


You can also choose whatever practically implemented "couple" coupling joint you want as long as it is made of one and the same material and its endpoints are at the same temperature (which means no net temperature gradient).

So 1 inch of thin copper wire soldered to 1 foot of thin iron wire as in the symbol < with temperature t1 to the right (both endpoints at one and the same temperature t1) and temperature t2 at the soldered junction to the left will give the same voltage as 1 yard of copper bar and 1 inch of iron bar interconnected with 2 feet of lead tube provided that the two junctions now created at the left side both are at temperature t2. This holds only in equilibrium, i.e. all connecting points and endpoints have settled.

This is imo implications from the link. Please check for yourself. A search 'thermocouple theory" gives several sources stating similar propositions.