Hello Reginald,

the LTZ100 has been discussed and investigated thoroughly, over on eevblog.

https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/

https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/

There you'll find chip and assembly pictures of the LTZ1000 / A, LTFLU,
and SZA 263 as well.

A lot of speculation and investigation about the time, temperature (T.C.
trimming), and hysteretic drift can also be found there.

It's a very long thread, anyhow.

Spoiler: The LTZ1000A shows more hysteresis and also random dips, than
the LTZ1000.

Then I also can recommend John R. Pickerings famous patent to mitigate
hysteresis effects after temperature excursions.

The timely drift of the LTZ1000 is linearly affected by the oven
temperature in a quite predictable manner, as you already assumed.
This has already been investigated and quantified by P J Spreadbury in
1990 "The Ultra Zener .. is it a portable replacement for the Weston
cell?". This basic article I don't find any more on the net, but I have
it in my archive.

The same has been done for the SZA263 and the LTFLU chips inside the
732A/B references:

PREDICTABILITY OF SOLID STATE ZENER REFERENCES, by David Deaver, Fluke.

Myself, I have collected the relative drift data of 4 references, 2x
LTZ1000, 1x 1000A, 1x double SZA263, over about ten years. The first two
references have been running continuously over that time, the other two,
only when used, because these are sitting inside a 3458A, and a 5442A.
All their individual drift rates seem to confirm the mentioned papers.

Recently, about 2 years ago I started another 5 virgin LTZ1000s, nicely
showing the typical initial timely drift of < 2ppm/year @ 50°C.

If you are interested, let's discuss how to exchange these data.. I'm
very interested in a drift estimation model for these relative
measurements, equivalent to the Three Cornered Hat method of the time-nuts.

Frank

Frank,

Thanks. Unsurprisingly IEEE has rid the web of the Spreadbury paper. I'll add it to my list the next time I make a university library visit.

What's rather different about my intended mathematical modeling is I do not assume a simple exponential. I assume that the correct functional model is not a straight line on log paper but is describable by a physical model which can be developed from basic continuum mechanics principles via the constituitive equations and device geometry.

What I want to do is to develop one or more physical models with unknown parameters and then search model space for the parameters and model which best fit. This has been routine in the development of multiparameter equations of state, but there are more recent methods than existed when I last dealt with that topic. Candes named the technique "The Dantzig Selector" in honor of the developer of operations research and the simplex method.

The mathematics I intend to use was developed in 2004-2006 by Emmanuel Candes and David Donoho. I consider it the greatest advance in applied mathematics since Wiener & Shannon in 1930-50. From an old seismic processor, that is high praise indeed.

The part that is mind boggling is you can solve a problem with 50,000 unknowns and 50 measurements. You have an overwhelming probability of a solution and if you get one, it is provably the optimal solution.

I stumbled across it quite accidentally. When it sank in I was doing something I'd been taught was impossible I was so fascinated I spent 3 years reading the most difficult 3000 pages of mathematics I've ever come across. One of the proofs by Donoho was 15 pages long! I even wound up adding a couple of books on the properties of regular polytopes in N dimensional space to my library! My BA is in English lit, so that is a rather far journey from where I started 40+ years ago.

I raised the subject on EEVblog a long time ago, but was never able to get usable data from anyone. I am merely asking for a chance to fail, but have not been able to get data despite repeated requests. Thus my query here.

As Kirkby pointed out, the metadata is as important as the data. Measurement protocol, lab conditions, instrument history are all significant factors. What makes what I want to try so different from what has been done in the past is described in the introduction to this paper by David Donoho from September 2004.

https://statistics.stanford.edu/research/most-large-underdetermined-systems-linear-equations-minimal-l1-norm-solution-also-sparsest

I've very much appreciate an ASCII text file of your data and relevant metainformation. CSV format is probably the easiest, but I can convert any format you have the data in given a description of the format.

Have Fun!
Reg    On Wednesday, August 28, 2019, 5:54:45 PM CDT, Frank Stellmach frank.stellmach@freenet.de wrote:

Hello Reginald,

the LTZ100 has been discussed and investigated thoroughly, over on eevblog.

https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/

https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/

There you'll find chip and assembly pictures of the LTZ1000 / A, LTFLU,
and SZA 263 as well.

A lot of speculation and investigation about the time, temperature (T.C.
trimming), and hysteretic drift can also be found there.

It's a very long thread, anyhow.

Spoiler: The LTZ1000A shows more hysteresis and also random dips, than
the LTZ1000.

Then I also can recommend John R. Pickerings famous patent to mitigate
hysteresis effects after temperature excursions.

The timely drift of the LTZ1000 is linearly affected by the oven
temperature in a quite predictable manner, as you already assumed.
This has already been investigated and quantified by P J Spreadbury in
1990 "The Ultra Zener .. is it a portable replacement for the Weston
cell?". This basic article I don't find any more on the net, but I have
it in my archive.

The same has been done for the SZA263 and the LTFLU chips inside the
732A/B references:

PREDICTABILITY OF SOLID STATE ZENER REFERENCES, by David Deaver, Fluke.

Myself, I have collected the relative drift data of 4 references, 2x
LTZ1000, 1x 1000A, 1x double SZA263, over about ten years. The first two
references have been running continuously over that time, the other two,
only when used, because these are sitting inside a 3458A, and a 5442A.
All their individual drift rates seem to confirm the mentioned papers.

Recently, about 2 years ago I started another 5 virgin LTZ1000s, nicely
showing the typical initial timely drift of < 2ppm/year @ 50°C.

If you are interested, let's discuss how to exchange these data.. I'm
very interested in a drift estimation model for these relative
measurements, equivalent to the Three Cornered Hat method of the time-nuts.

Frank

volt-nuts mailing list -- volt-nuts@lists.febo.com
To unsubscribe, go to http://lists.febo.com/mailman/listinfo/volt-nuts_lists.febo.com
and follow the instructions there.