A Low Cost OSHW Voltage Calibration Reference Project

[rhb]

@svenskelectronik and I have been discussing design  and production of an LM399 based voltage calibrator derived in part from TiN's KX design. He owns a PCBA facility, so he is in a good position to manufacture and market an OSHW product without depending upon it for his livelihood.

Initial outline is as follows:

provide 5-10 steps per decade over 3 decades  from 10 V max to 10 mV min

use low cost 1% metal film resistors for the voltage dividers and op amp feedback resistors

provide on board voltage regulation and filtering

use an MCU and an LCD display to show the selected output voltage

cycle through the outputs via a pushbutton

display the  output voltage on the LCD

package in a shielded box with temperature management and measurement

store calibration constants in the MCU flash or EEPROM

burnin units for 500-1000 hours before shipment on a calibration system to get aging coefficients

temperature cycle the units during burnin to get temperature coefficients

fit burnin data via a sparse L1 pursuit to determine calibration constants

at each calibration,  recalculate calibration coefficients for best fit to all calibrations

make the calibration process completely automated other than physical connection to calibrator

The basic premise is that one can build a device whose behavior can be accurately predicted  at lower cost than making a device which is physically as accurate.  For a voltage calibrator there is no reason to prefer a 1.000000 V output to a 0.997869 V output if one knows the latter as accurately as one knows the former. The MCU and LCD allow displaying a variable  output voltage adjusted for age and temperature. If humidity and barometric pressure prove significant, one can add sensors for those.  A key goal is to reduce the cost of calibration to very little more than shipping cost by automating the procedure.

Barring  device damage, I think such a design could achieve the rather ambitious goals I mentioned earlier of 5-10 ppm after 5 years.  Doing this is a bit beyond what one person can do.  I know the math for the calibration process.  But there are many other important details about which I am ignorant.  I think that most things of significance that can be done by one person have been done.  What remains are things that require the efforts of several people.  Hence the importance of open source efforts.

This should probably move to the Projects forum, but I thought that it should be discussed here first.  I'd like to hit a BoM of under $50.  That's pretty much entirely dependent on being able to calculate the resistances as a function of the relevant factors as well as op amp gains, offsets, etc along with the aging and temperature dependence of the LM399.

Yes, I am acutely aware that this is *very* ambitious.  But if it were easy what would be the point?  You can easily buy a better reference, but it's not cheap.

Have Fun!
Reg

[MisterDiodes]

Your plan sort of falls apart at:

"use low cost 1% metal film resistors for the voltage dividers and op amp feedback resistors".

Therein lies the problem - those will never come close to being stable enough over time, temperature and humidity.  That's why we use PWW and Bulk Metal Foil resistors that are tens or some hundreds of times more stable.

You might want to do more research - there are already other versions of the cal box you're describing, here on the forum and on the 'Net.  Right down the LCD display, pushbuttons, voltage output etc.  Look around.

[TiN]

Sorry, there is a core conflict between the items

5-10 ppm after 5 years.  Doing this is a bit beyond what one person can do.

That's 1-2ppm/annual, which require careful selection of state of the art parts (LTZ) and permanent fixed deployment.
It's nothing to do one or ten person involvement.

and

the LM399

You will need to buy and sort thru lot of LM399's to even approach 2ppm/annual. That task alone will take you few years.

use low cost 1% metal film resistors for the voltage dividers and op amp feedback resistors

That is the main catch. LTZ references are expensive NOT because of 50$ LTZ chip, but because of resistors. And output stage 7V-10V is WAY MORE difficult and demanding than LTZ/LM399 source itself.

I'd like to hit a BoM of under $50.

Okay, just stable and good performance package in a shielded box with temperature management and measurement will cost more than $50.

You see others expensive and often overspecified discussed designs here on forums not because we want to spend lot of money, but because there are no much other ways.
Design is as stable as weakest component in the circuit.
 

[cellularmitosis]

Good feedback from very experienced members in here, but let's play with the idea just a bit further.

If you were doing down the road of "yes my components will drift, but if the drift is predictable and I can model it, I can extrapolate where my next calibration point will be, to X accuracy", what you'd want is a board where each of the resistors have four test-point pins exposed, so that you periodically power down the board and perform 4-wire measurements on all of the resistors.

[TiN]

drift is predictable and I can model it

, I can extrapolate where my next calibration point will be, to X accuracy

How to prove (every time) reality matching the expectations though? .

So what you saying is : instead of forking 50$ for resistor and forgetting about it, I'll spend XX hours every week/month/year to test unstable resistor and wiggle the calibration math to get the numbers to coverge. Sounds like somebody have too much free time  .

[zhtoor]

hello reg,

thanks for stimulating our thoughts with an interesting idea.

now what i understand from your description is:-

1. you model the reference as a "black-box"
2. this black-box is isolated enough to be called a "black-box"
3. this bb produces a voltage output, which is to be modeled against time (drift), and environmental factors (temp, humidity etc.)
4. you come up with an equation like: Vout = f(time, temp, humidity, pressure) whereas time is accumulated in the black-box (infinite battery?)

how about thinking of a common carbon-zinc or an alkaline cell as such a system (with appropriate sensors etc.)?

regards.

-zia

[TiN]

It would be more like Vout = f(time, temp, humidity, pressure, output loading, EMI/RFI condition, mechanical stress, orientation, airflow, setup thermal constant). We talking 1-2 ppm/year, as outline by first post. 

[hwj-d]

Yeah,

play arround with your thoughts, discuss it here, build some refs, and then, after a couple of time for further maturation, repeat your project yourself again.

Take it easy 

[Kalvin]

Here is a recent, practical oriented article published in EDN "Factors affecting reference long-term drift performance":

https://www.edn.com/design/analog/4460470/Factors-affecting-reference-long-term-drift-performance

[rhb]

Zia is the only person who understood the concept.

I have 30 years of very varied experience with multichannel data analysis and processing.  I was the person people went to when they had unusual problems.  See "Random Data" by Bendat and Piersol for the traditional treatment based on Norbert Wiener's work in the late '30's and early '40's.  I also  recently spent 3-4  years learning about sparse L1 pursuits.  See "A Mathematical Introduction to Compressive Sensing" by Foucart and Rauhut for a treatment of the work of Donoho and Candes who did the primary work on sparse L1 pursuits.  For a less intense summary of sparse L1 pursuits see "A Wavelet Tour of Signal Processing" by Mallat, 3rd ed.

I bought Foucart and Rauhut  in 2013 when I realized I was solving problems I'd been taught in grad school at Austin could not be solved.  That was quite exciting, so I wanted to know how and why this was possible.  It took 3 years and 3000+ pages of text, but I now understand it reasonably well.

A simple example.  A voltage divider composed of two stable metal film resistors.  By stable, I mean that  age dependent effects are negligible.  If those resistors are placed in  an enclosure filled with thermal grease  and a thermistor and a precise voltage fed to the pair. by measuring the drops across the resistors as a function of temperature using a heating chamber I can calculate the voltage drops at any temperature over a wide range at any time in the future.  Moreover, I can do this to the same accuracy or better than a pair of expensive low tempco precision resistors. I am in no way constrained by nonlinear behavior and  can readily accommodate thermocouple effects, thermistor errors etc.

If that is not possible, then using the most expensive precision resistors made is futile also.  Air currents, EMI etc all have to be controlled in the usual fashion.  4 wire resistance measurements are not needed.  Measuring the node voltages will suffice.

The accuracy of estimates of the aging of the LM399, or any other reference, are dependent upon how much history is available.  To make use of that one must recalculate the aging function *every* time the device is calibrated.  I see no evidence that is being done today by anyone.  My assertion is that with 5 years of performance data the error will drop to 2-3 ppm or less.  The initial error will be an order of magnitude larger. 

I spent several days modeling predictions of exponential curves.  What I found was that accurate prediction (i.e. 1% prediction error) was possible out to a time twice the length of the prior history.  I've not yet done this for  a logarithmic function, but I shall. As the exponential and the logarithm are closely  related, I'd expect similar behavior from the logarithm.  A device which is changing by 20 ppm per year which has been properly characterized can be modeled to 0.2 ppm if the prediction error is 1%.

I've read many statements now about the "random" variation of a set of references.  But when I look at the data it looks highly correlated to me.  There is a significant random noise component, but the dominant features are not at all random.

The mathematical side of this requires a lot of data and probably a month or two to analyze.  I've gotten data from TiN but I need more.  I especially need data from Andreas and others with similar data in CSV format.  Most of the work is sorting out how to handle the data and developing filters to throw out bad data. Bad data can only be identified by having lots of data and rejecting things which are 3 standard deviations and more out.

For a super major I extracted from the company database 135 GB of geophysical measurements made in well bores in the GoM in over 19,000 wells.  None of the data had any usable pedigree.  To use it I had to examine how each sample related to nearby samples both in the same and in other wells and then develop criteria for rejecting invalid values.  For the wireline data I had to get time on a seismic processing cluster for  overnight runs.  Voltage references are a cakewalk by comparison.

[chickenHeadKnob]

There is a forum member- IanJ (Ian Johnston) who makes and sells a reference which is remarkably similar to the one you are proposing. see forum thread:
https://www.eevblog.com/forum/metrology/ian-johnston-pdvs2/msg1196493/#msg1196493

or Ian's web store:
http://www.ianjohnston.com/index.php/onlineshop/handheld-precision-digital-voltage-source-v2-detail

Of course it doesn't meet your price target and it deviates in other ways, but it can give you an idea of some of the trade offs he made. He has some background on the design and its iterations on his web site.

[TiN]

rhb
Did you see this paper. datasheet and this?

In the end - if you would set target to 4-8 ppm/year, I would be first person to tell you - go all in, and would donate few references for the project.
1-2 ppm/year? Maybe not. Either way 50$ BOM not going to happen . Theory and math correction are great tools, but they should be in connection with practicality and real-world scenario, which is not as simple as tempco/humidity and pressure correction. I would be happy to see this project move from forum discussion to some hardware implementations, and would support just to get nice box with internal oven, control and monitoring functionality.

In applications like low power reference, precision resistors are used due to single reason - better materials that provide long term stability. To be completely correct, there is NO need for precision resistors in DC reference, what is required - long-term stability (both temporal and environmental). Like you already know, it does not usually matter if it's 7.041843V output or 7.000000V output. If there is 5% stable resistor, it would work just as good. It's just from resistor manufacturer business perspective no sense to make 5% ultra-stable resistor with better materials, than 0.1% ultra-stable resistor. You can ask any resistor manufacturer, and if they kind enough then you'd see price between 1% high-end resistor and 0.005% high-end resistor is actually same in manufacturing, it's just extra testing cost to provide you 0.005% tolerance guardband.  Paying extra for 0.005% resistors in LTZ reference circuit it's just waste of money for nothing.

[zhtoor]

hello,

valhalla 2720GS is probably the only reference unit which does remotely like suggested here,
by taking out mal-performing references out of an averaging circuit.

regards.

-zia

[Magnificent Bastard]

Rather than design and build a calibrator that has limited functionality (e.g. low DC voltages only), why not change the specifications a bit, and design one that handles 0Vdc to 1000Vdc, 0Vac to 1000Vac, 0Adc to 30Adc, 0Aac to 30Aac, and ohms from 1-ohm all the way to 1G-ohm... ??

The resistors can be directly connected to the output for those meters that supply higher voltage, and also run through a resistance synthesis circuit to produce fractional values in between.

The DCV, ACV, DCI, ACI functions do not need to have zillion-digit adjustability-- only on tenths.  In other words, 1.0V to 9.9V (etc.)  These could be generated by a PWM DAC (with high stability, but low cost and it does not need a zillion bits).

Then, because we are limiting the settings to a small set, each voltage on every setting can be calibrated by a freshly calibrated 3458A (automated of course), and this calibrator would then be good enough to calibrate and/or adjust a 6.5-digit meter.  This way, we don't need to worry about linearity (etc.)-- only stability and repeatability.

Some meters have a capacitance function, and so it might be a good idea to add that as well (and same thing-- some synthesized values could be made out of a few capacitors and some additional circuitry).  The capacitance function of even high-end DMMs is really for indication only, and so these capacitors do not need to be calibrated to very low uncertainty-- a calibrated GenRad LCR meter would be just fine (0.02%).

I have no idea if inductance should also be added.  Maybe someone can find justification for that.

Of course, this is not all going to happen for US$50-- it will be much more (realistically), but I bet it can be done for lest than US$500 with the right choices of off-the-shelf parts.

[MisterDiodes]

RHB:

You started out with:

"...But there are many other important details about which I am ignorant."

But then in a later post you seem to have it all figured out.  I'm sorry, and I say this will all due respect - It might be time to put the books down and start learning.

Or as the sign over my office says:  "Where theory ends and reality begins"

Here's an easy conceptual test you can try right now.  Take a couple of your "low cost" resistors you think will be stable enough and put them under load for 1000 hrs.  Maybe measure TC and humidity effects.  And while you're at it - since you want a variable voltage output look at Power Coefficient change (PC) and Voltage Coefficient change (VC) as well - something most people here overlook because they're too enamored with TCR.  For instance, realize the power in any resistor at 10V bias is 100 times that of the same resistor at 1V bias, and remember that your drift rate even on the best resistors is highly affected by some ppm by EVERY mW power dissipation over time.  A lot of that is dictated by the thermal coupling between the resistor and its environment.

So do you "burn in" at full power condition or low power condition?  You could argue either way is valid.  What if someone runs their device all the time at max output voltage?  Compared to someone that tosses the device in a drawer for months at a time.  How do those devices compare to each other after a year, and how do you predict the end usage condition?  How do you predict usage history??  Or maybe one device lives in a warm humid place, and another device lives in a dry, cold climate.  While they operate the same way in a year or two?

And what makes you think that "burning in" a cheap resistor will get it to stabilize?  Have you ever tried doing this and tried to come up with a prediction model?  You might want to investigate that part very carefully before proceeding.

So - I'm suggesting as your first step of discovery is take some cheap resistors, and measure their ratios on an 8.5 digit DMM (backed up by a stable Vref and Resistance standard).  Since you're going for low ppm you'll need a capable measuring setup.  Measure changes over time, temperature, humidity, power and bias points for starters.  Also make sure your test resistors are mounted on a realistic PCB and realistic enclosure for realistic thermal coupling, otherwise all you're doing is measuring the "dangling in free air" condition of the resistors...which is not really the end use condition.

One thing you'll come across - you get that pesky power-dependent drift due to bias/power point, and doggone it now you realize you've got hysteresis too: You go to high power for a while, and then when you come back to lower power...Shoot...The resistance value never came back to the same place from where it started.  So you go back to higher power...and now it's different again.  And so on.

Again:  At some point you realize why we have stable PWW and foil resistors.   

As we tell our clients that want a precision circuit designed:  You can have it "Fast", "Cheap" or "Good".  In the Metrology world, you pick ONE.  Generally if you want any performance at all at ppm, you pick "Good".  "Cheap" is almost always a complete waste of time. "Fast" is usually never as important as "Good".

[zhtoor]

Zia is the only person who understood the concept.

thats probably because i am a total newbie to the subject.

The mathematical side of this requires a lot of data and probably a month or two to analyze.  I've gotten data from TiN but I need more.  I especially need data from Andreas and others with similar data in CSV format.  Most of the work is sorting out how to handle the data and developing filters to throw out bad data. Bad data can only be identified by having lots of data and rejecting things which are 3 standard deviations and more out.

andreas has huge amount of experience with LM399, TiN has *huge* amount of experience in LTZ1000, both of them probably can setup an environmentally perturbed
references (both LM399/LTZ1000) in controlled chambers and stream data to rhb for analysis and building a predictive model based on this streamed data. perturbations to
the environmental chambers can be co-ordinated by rhb/andreas/TiN and then the model / results published here.

maybe alex nikitin can pitch in with a JVR and somebody else does an LT1021BCN8-10 on a standardized eevblog vref streaming server (TiN?) 

best regards.

-zia

[rhb]

Arguments that "it won't work" are pointless.  The only way to find out is to try it. 

They told the Wright brother's "it won't work".  Rather than trust other people's data they built a wind tunnel and collected their own.  We all know the outcome.

I feel quite confident that no one is applying the mathematical techniques I described to the implementation of a reference.  What I'm suggesting is a non-trivial application of state of the art mathematical techniques.  Will it make a difference?  No one knows.  Or at least if they do, they have not published.  Ten years ago I should have said that this is not possible.  Now I know it might be.  So it's time for an experiment.  This is the experiment.

If there are aging data for cheap metal film resistors please direct me to it.  I shall certainly require it. Should someone wish to collect some I should be most grateful.  I've got a lot of work to do just setting up the infrastructure to make the measurements.  And collecting 1000 hrs of data will take 6 weeks.

Hysteresis would require logging the relevant variable.  I presume that all behavior can be described by the device physics. Furthermore, I presume that the device physics can be described mathematically.  If the rules change by divine intervention, then obviously such an approach will not work.  But there are a few hundred years of evidence that God does not intervene in day to day physics.

It really doesn't matter what the resulting accuracy is.  Whatever the accuracy is, it *will* be better for the price than the result if one ignores the math.

I do need all the data I can get.  I can deal with any limitations of a dataset if I have enough data.  I should be able to show quite a bit with existing data.

[cellularmitosis]

So it's time for an experiment.

  

If there are aging data for cheap metal film resistors please direct me to it.  I shall certainly require it. Should someone wish to collect some I should be most grateful.

Not related to this project, but I've been wanting ageing data on cheap metal films for a while now as well.  My HP 3478A is just collecting dust, and 5.5-digits will probably be enough to get useful data on cheap metal films, so I'll commit a scanner card to getting this sort of data for you (and for everyone).

Looks like these things are just controlled via a 30-pin card-edge connector, probably 5V TTL.  Should be easy to just rig up an Arduino to control a single card in isolation.

[TiN]

Arguments that "it won't work" are pointless.  The only way to find out is to try it. 

Yup, as MB explained nicely, time to set keyboards aside and start soldering . Sorry If my posts got you wrong thinking, I'm not saying "it won't work", I'm saying it would be very expensive (not in parts cost), long and painful learning process (which may or may not lead onto outcome you desire).

They told the Wright brother's "it won't work".  Rather than trust other people's data they built a wind tunnel and collected their own.  We all know the outcome.

Perfect example, where outcome (flying planes) came from getting hands dirty and doing stuff. No math in the world would help if we just talk about it.

I feel quite confident that no one is applying the mathematical techniques I described to the implementation of a reference.

That is what whole multi-M$ metrology industry is doing for decades. Predictions and math theory goes hand to hand for improving uncertainties of everyday measurements. If you look into primary standards theory and experiments, you will find lot of amazing papers and projects around those. But all those cost bit more than 50$ to achieve even 5ppm/year .

And collecting 1000 hrs of data will take 6 weeks.

That's just a warm up time . Amount of data to collect to correlate for stability, is the reason why stuff in ppm world usually take months and years.

Whatever the accuracy is, it *will* be better for the price than the result if one ignores the math.

I don't know, but time to prove math works also have the price.

cellularmitosis
Good idea. I was meaning to post my K7168 code (I use RPi to control my card).

[cellularmitosis]

cellularmitosis
Good idea. I was meaning to post my K7168 code (I use RPi to control my card).

Aha!  Perfect!  That would be great 

[MisterDiodes]

I love the references to Wright Brothers.  That also reminds me of Hitchhiker's Guide by Adams: There is an art, it says, or rather, a knack to flying. The knack lies in learning how to throw yourself at the ground and miss. Pick a nice day, [The Hitchhiker's Guide to the Galaxy] suggests, and try it.


So go try it!

The difference here is that there is people have been trying to achieve any sort of stability with cheap resistors for decades (trying to ignore basic Chaos Theory).  The exact end result: That's why we have PWW and Foil resistors, which are already good solutions to the "flying" problem.  It's not like we haven't taken flight before in the realm of resistor stability.  The underlying reasons of how & why cheap resistors are built on a grand scale is also the reason they are more unpredictable,  chaotic and not well suited for critical circuits requiring good stability.  You get what you pay for.  Always.

RHB,  you are correct here:  It IS futile to expect any device or circuit to remain "in calibration" forever.  That will never happen on ANY device, no matter what components are used.  That's why all precision instruments must have periodic calibration performed.

So when you design a precision device with some target "drift rate" that also must include the target calibration interval, and you also include device uncertainty calculations over say 15 minutes, an hour, a day, 30 days, 1 year, etc.  This is the metrology section, so every measure and spec includes some non-zero level of uncertainty.

The main problem you'll have to overcome:  Even if you had a model to predict drift of some crappy resistor, it would be impossible to predict the future use case of A LOT of variables that affect ALL resistors - Temperature, power, bias point, humidity, stress, time, duty cycle, etc. etc. etc..  Good resistors eliminate or restrict those variables down to a -much- more manageable scale.

So go start testing cheap resistors!  Let us know how the prediction model works out for you - and don't forget to test repeatability after the power level changes!

[MisterDiodes]

One last thought for RHB - nowhere in your dream $50 spec sheet do I see a target spec for noise (maybe I missed it?) - and that's dependent to a certain extent by the resistor technology used.  Even though the '399 isn't exactly the quietest Vref on it's own, you always have to look at the contribution of resistor noise to the system.  That total apparent noise will always limit the usable uncertainty of the device - especially at lower output levels.

Once you learn why PWW and BMF resistors are (sometimes much) quieter, then you begin to understand the unavoidable low-level chaos and hysteresis effects inherent to other resistor technologies.

Have fun testing!

[rhb]

Perhaps someone would be good enough to point to literature references on the various effects on resistors and their magnitude.  They are after all, sufficiently well known that I had presumed the need to model them even if I did not state such.  Age and temperature are first order effects.  After all, if soldering a resistor will change the value, it's rather obvious that significant changes in current will also.

I'm sorry that I can't supply a complete and finished specification of every last detail at the start of an experimental project.  In my experience, doing that is a waste of time. You estimate the first order issues and then go where the data takes you.

The rock properties project for the super major I mentioned was said to be impossible by a bunch of the research staff at the company lab in California.  I did  it and it worked very well. So well that they had me do it a second time.  The first version was only 9,000 wells predominately on the shelf for the shelf business unit.  Deepwater was so impressed they wanted it done for them.

[hwj-d]

I'm sorry that I can't supply a complete and finished specification of every last detail at the start of an experimental project.  In my experience, doing that is a waste of time.

Sorry, I see some different things that represent "a waste of time". 

[BNElecEng]

rhb, do you have access to IEEExplore? I ask because not everyone does. I did a quick search and there look to be some good papers there I could send you a link to.

In addition to the usual sources, take a look at Vishay Z-Foil datasheets as well as the papers they have published. They have a reputation as one of the leaders in the field of precision resistors. Here is an example:  https://www.vishay.com/docs/28873/predictable-components.pdf

FWIW, I don't think anyone here is questioning your credentials. You asked the question and received the considered answers of multiple forum members.

[rhb]

No I don't have IEEE access.  I'd be grateful for any papers you can send me.   I'll send a PM.

The Vishay paper was quite interesting, but the analysis depends entirely on the behavior being linear on a log plot.  As noted, that may not be the case.  This is where the sparse L1 pursuit comes into play.  One can reasonably compute 100,000 or more possible curves based on physical models which do not follow the  Arrhenius’  equation and then select the best solution.  The Vishay paper employs the analysis I was taught in grad school 30 years ago.  There's nothing wrong with it.  But Donoho and Candes discovered something better.  Having spent the better part of 3 years studying the mathematics, I'd like to try it out on some practical problems that either have not been tried or have not reached COTS status.

The point of this is to substitute  non-recurring engineering cost for BoM costs.  This is an outgrowth of @svenskelectronik's desire to promote OSHW.  I don't have anything to contribute to classical designs using precision components.  I do have something to contribute to device modeling if that would allow using lower cost parts.

This is just a side venture for me.  My primary effort is a compressive sensing DSO based on COTS Zynq based instruments. I have almost all the infrastructure for that in place.  The big obstacle at the moment is having several days free to devote entirely to  MicroZed tutorials.  Mathematically I know what I need to do, but I have no FPGA experience so I need to learn that first.

I don't mind discussing compressive sensing with someone who is willing to actually read the papers and has a solid grasp of traditional analysis, but I discovered to my dismay with the DSO project that most will just spout dogmatic opinion and no amount of time spent explaining the matter is of any use.

[ArthurDent]

Although I'm fully aware of the 'knack' of learning to fly reference mentioned above, I don't think it applies here.

What I believe is that if you try to build a relatively great car by just using cheap parts you're more likely to end up with a Yugo than with a Rolls.

[mycroft]

Please, post the links. Some members may have access to IEEExplore.

rhb, do you have access to IEEExplore? I ask because not everyone does. I did a quick search and there look to be some good papers there I could send you a link to.

[zhtoor]

Although I'm fully aware of the 'knack' of learning to fly reference mentioned above, I don't think it applies here.

What I believe is that if you try to build a relatively great car by just using cheap parts you're more likely to end up with a Yugo than with a Rolls.

i also agree, the "knack of flying" metaphor probably does not apply here,
more likely would be building a car with electronic ignition vs. mechanical ignition,
and not a Yugo vs. Rolls.

-zia

[ArthurDent]

My reference to the "knack of flying" metaphor has a lot to do with my log-in name and very little to do with its applicability. 

[BNElecEng]

Here are a few links to papers I found interesting about factors that affect resistor stability:
http://ieeexplore.ieee.org/document/6898501/
http://ieeexplore.ieee.org/document/5544765/
http://ieeexplore.ieee.org/document/676956/
http://ieeexplore.ieee.org/document/850998/
http://ieeexplore.ieee.org/document/546537/
http://ieeexplore.ieee.org/document/278563/
http://ieeexplore.ieee.org/document/6250962/
http://ieeexplore.ieee.org/document/6513303/

This rabbit hole runs very deep. If you already have access to IEEExplore, search for the term "standard resistor" and you'll get a few hundred hits.

[ch_scr]

So on one hand you propose to make a reference and calibrator with cheap components, based on the other hand on (nice looking) data you get from references that are built with high stability components?
I believe you need data from references with low stability components and base your assumptions on that.
This just crossed my mind, i do not want to discourage the general idea.

[dl1640]

If a few ppm can be achieved under $50 for a calibrator why would one need Fluke?

I'm not saying it is impossible to do it, but how is reliability?

You'll need even more expensive equipments to test the "cheap" components and designs...

[Kleinstein]

The idea of using mathematical ways to correct errors instead of using expensive parts is a good idea. However with predicting drift this is difficult as it takes a very long time to verify the models or only determine the parameters, especially if many processes (and thus more parameters) are involved. So the math can fix quite some errors (e.g. Temperature effect, maybe more than just linear), but with to many factors to compensate it just gets impractical to find a suitable model, as the time needed goes up faster than linear.

It might be a surprise to someone not used to it, but a simple resistive divider can become quite a delicate part when it comes to high precision. Something like a TC could be compensated, but resistor quality is more than just the TC. However the TC is just the easy part and thus used in data-sheets. The difficult parts are those you can't easily put into simple numbers.

There are alternative ways to divide down / scale a reference voltage:
1) a PWM divider
2) feedback through an ADC  (the gain stability of something like an LTC2400 is not that bad compared to similar priced resistors)
3) a transformer
4) capacitive divider / charge pump

[hwj-d]

So, gradually, the whole thing visibly becomes what it actually is, a joke 

[ArthurDent]

 The idea of modeling different variables affecting overall accuracy to predict future performance isn’t new or confined to volts. HP used this idea in their Z3801 GPS frequency standard using their ‘SmartClock’ technology back around 2000. The idea was to continually monitor the variables used to discipline a 10 Mhz crystal oscillator used in the unit compared against the GPS 10 Mhz reference and use that information to predict how to compensate the local oscillator in the future if the GPS signal was lost.  It took a long time for the unit to learn the ways the different variables affected performance but the longer the unit was on the better the prediction was and if the signal was lost the algorithm did a pretty good job of keeping the 10 Mhz oscillator on frequency for up to 24 hours. When the signal was lost the unit was no longer making continual corrections and had to use its ‘best guess’ of how time, temperature, aging, etc., were affecting the performance at the moment the GPS reference signal was lost and the oscillator went freewheeling. HP realized that as soon as the signal was lost the accuracy was going to get worse and worse and the spec on holdover was for 24 hours and this was with one of the best commercially available oscillators at the time.
 
 If you try to design a voltage standard with elaborate microprocessor control using less accurate parts it will kind of work, and I don’t doubt that, but unless it is continually referenced to a standard except during a short period of time when you are using it as a transfer standard, the uncertainty seems to be a big problem.  The butterfly effect so elegantly demonstrated in Homer Simpson’s toaster time machine episode shows the difficulty in predicting future results.

[zhtoor]

So, gradually, the whole thing visibly becomes what it actually is, a joke 

"What can you do, thought Winston, against the lunatic who is more intelligent than yourself,
  who gives your arguments a fair hearing and then simply persists in his lunacy?"

-George Orwell 1984 part 3 chapter 3

[rhb]

I read a number of papers yesterday and looked at some of TiN's data.   I've had a long think about the system this morning.

Error terms I have identified so far:

temperature
pressure
humidity
age
thermal gradient
current level
surface leakage
resonant modes in the enclosure

The errors introduced by changes in temperature,pressure and humidity have demonstrated hysteresis, particularly affecting the resistors.  As a consequence, to mathematically correct for errors they introduce would require a complete record of the environmental changes.  It would also require developing models which would require an extraordinary amount of data.  The effects are near linear with small hysteresis for small excursions, but quickly become problematic. The most appropriate means of addressing these is to control the environment as tightly as possible.

While not appropriate for the requirements posited, in general a thick wall enclosure machined from a block of 1008 low carbon steel, filled with dry nitrogen and sealed is the most economical and easiest option to fabricate.  The thick walls have the additional benefit of providing good magnetic and electric shielding.  Such an enclosure should be annealed after fabrication which requires a very specific temperature cycle over 1-6 hours.  Unless one happens to have the requisite machine tools and furnace, such effort is only relevant to an ultra precision reference because of the cost

If one is willing to forgo very tight pressure control, a lightweight steel enclosure can serve for controlling temperature and humidity at an economical cost while still providing some damping of pressure changes.  An American pattern .50 BMG ammo  should provide a satisfactory  enclosure, allowing plenty of room for insulation and a large battery for hot shipped travel standard use.

Thermal gradient and current levels are probably first order effects.  Irrespective of the enclosure temperature, device heating leads inevitably to thermal gradients.   Thus it is useful to minimize currents as much as is consistent with electrical noise. In the case of a buried zener reference such as the LM399, minimizing the heater current will serve as a better proxy than using a temperature sensor.  However, such an approach may necessitate logging temperature so as to correct for hysteresis and aging effects upon the resistors.

Current levels are a significant factor both in resistor aging and hysteresis.  The latter can be avoided by avoiding changes in current level.  The effect on aging must be dealt with mathematically.

Surface leakage is not something I have any real sense of and undoubtedly varies widely depending upon board cleaning, materials etc.  That will require some investigation.  It may well be a 3rd order error term that can be neglected.  For a high precision device I should expect that a ceramic substrate would be desirable.

Zener diodes are intrinsically broadband noise sources.  Placed in a metal enclosure will it will develop resonances controlled by the enclosure dimensions. Without adequate filtering this can interact with the sampling to produce errors.  While estimates are readily made of the frequencies, power levels and radiation efficiency have a dominant effect.  This argues for a PCB layout which minimizes radiation.

The level of noise which I have observed in the data from TiN suggests to me that it might be better to forgo the PLC sampling and collect measurements at the maximum rate the DMM can make and then use a Fourier transform to determine the DC value.  If one discards all but a random subset of the data, then the requirements for compressive sensing are met and aliasing does not take place.  The use of the Fourier transform has a rather significant advantage in that it  should permit making readings to 8.5 digits with a 5.5 digit DMM by taking a sufficiently long sample.   There are a number of mathematical constraints which must be satisfied, but this is often the case. 

I think I should note that one can acquire data with arbitrary amplitude resolution using a single bit ADC.  Sam Allen published a good bit of work on the subject in the '70s.  In his case he was acquiring very high fold vibroseis data (1024 channels when others were only collecting 48).  The rapid evolution of seismic recording systems overtook him and the technique is largely forgotten.  It's actually a trivial application of the work of Norbert Wiener.

As regards the choice of resistors I have nothing more to say at this time other than they need to be "good enough".  There is wide variation among manufacturers and processes in the behavior of resistors and their sensitivity to environmental factors.  I shall eventually have three LM399s running in a controlled chamber which use low spec resistors.  One is running the best matched pair which had an initial resistance before soldering within 0.02%.  Temperature coefficients were not measured.

As regards the mathematics.   In pore pressure prediction a common approach is to use a reference function which is the sum of a constant and an exponential with a constant exponent.    As I had a vast amount of data I could see that such a form was not a particularly good fit.  It looks pretty good with a small data sample from a few wells, but not when you have thousands.  Ultimately I found that I could get a much better fit by using a function of the same form, but with a polynomial exponent.  Were I to do the same project today, I should construct a vast array of curves calculated from the physics of fluid flow in porous media and then use a sparse L1 pursuit to find the best solution.  None of the papers I've read to date suggest that any of the authors knew how to do the sort of approximation I was doing over 10 years ago.  There may well be a group which has.  If so I look forward to reading their work.

[hwj-d]

The butterfly effect so elegantly demonstrated in Homer Simpson’s toaster time machine episode shows the difficulty in predicting future results.

So, we present a new nickname to rhb 

[eurofox]

While not appropriate for the requirements posited, in general a thick wall enclosure machined from a block of 1008 low carbon steel, filled with dry nitrogen and sealed is the most economical and easiest option to fabricate.  The thick walls have the additional benefit of providing good magnetic and electric shielding.  Such an enclosure should be annealed after fabrication which requires a very specific temperature cycle over 1-6 hours.  Unless one happens to have the requisite machine tools and furnace, such effort is only relevant to an ultra precision reference because of the cost

I have the intention is to apply a thermal treatment (annealing) to remove the mechanical stress after the machining work, I have a oven and equipment for this as well the knowledge to do it.

With respect to filling with inert gas, it cross my mind but somehow at least in my case I have a buffer to output 10V and need to be tuned in the metal container after aging, this mean there need to be an small hole just above the trimmer and need to be closed hermetical, not easy to do with gaz ....
I will mill under the PCB an area to place 2 packets of silicate gell, they will suck the humidity that could enter te container during the trimming job.

eurofox

[rhb]

I have the intention is to apply a thermal treatment (annealing) to remove the mechanical stress after the machining work, I have a oven and equipment for this as well the knowledge to do it.

With respect to filling with inert gas, it cross my mind but somehow at least in my case I have a buffer to output 10V and need to be tuned in the metal container after aging, this mean there need to be an small hole just above the trimmer and need to be closed hermetical, not easy to do with gaz ....
I will mill under the PCB an area to place 2 packets of silicate gell, they will suck the humidity that could enter te container during the trimming job.

Braze an inner tube or tubeless tire stem to the hole for access to the trimmer.  Then use a metal cap and gasket to seal it after adjustment.  It does require having a tank of dry gas around, but I suspect you've got a bottle of argon or argon/CO2.   Nitrogen is just cheaper.  Assuming everything is dry, the gas just displaces any moist air.  It doesn't need to be under pressure, though it will if the atmospheric pressure drops.  I'm suggesting a tube you can poke into the unit to inject the gas near the bottom.

[eurofox]

I have the intention is to apply a thermal treatment (annealing) to remove the mechanical stress after the machining work, I have a oven and equipment for this as well the knowledge to do it.

With respect to filling with inert gas, it cross my mind but somehow at least in my case I have a buffer to output 10V and need to be tuned in the metal container after aging, this mean there need to be an small hole just above the trimmer and need to be closed hermetical, not easy to do with gaz ....
I will mill under the PCB an area to place 2 packets of silicate gell, they will suck the humidity that could enter te container during the trimming job.

Braze an inner tube or tubeless tire stem to the hole for access to the trimmer.  Then use a metal cap and gasket to seal it after adjustment.  It does require having a tank of dry gas around, but I suspect you've got a bottle of argon or argon/CO2.   Nitrogen is just cheaper.  Assuming everything is dry, the gas just displaces any moist air.  It doesn't need to be under pressure, though it will if the atmospheric pressure drops.  I'm suggesting a tube you can poke into the unit to inject the gas near the bottom.

I do have a TIG welding equipment and of course pure argon is available and needed for the TIG welding, is maybe an option. 

eurofox

[rhb]

hello,

if i may suggest a simpler approach first, that would be:-

1. take a standard LM399AH (or LM329AH) the hermetically sealed versions, as a reference element.
2. do not use the heater in case LM399
3. no resistive dividers, use another reference (auxilliary, maybe another LM399) to provide the zener current or bootstrap, direct 6.95V ref would do
4. use a water (gallium) double point cell to locate the electronics in (possibly coated with conformal coating and submerged in compatible oil)
5. now start measuring the reference after the initial couple of 100 hours of operation
6. fit the initial model and proceed where the data leads you

I have 3 basic-lm399 references I was given by @cellularmitosis, so my intent is to use them.  I have a *lot* of work to do on EMI mitigation.  And even more work to do to log data automatically.  I'm looking for the inflection point on the price performance curve, so I want to work my way along.

I have to have some form of buffer amp,so the OP07s that were provided seem a good starting point. Comparing 3  references with a $12 BoM seems to me a useful data point to collect.  Also he is running a slew of LM399s himself.  I don't know if they are all the same design, but I'm sure at least some are.

The LM399s he gave me are used, so I don't think I can determine the  shape of the aging function from them.  For my purposes the first few hundred hours are probably quite important.  Once the rate of change starts to flatten out it becomes difficult to identify the correct function and coefficients.  Correctly finding parameters from measurements near the asymptote is nearly impossible if there is any noise.  The result is that  the error becomes larger over time than it would be if one had the earlier data.

@eurofox  LoL  I was pretty sure I had it right.

[Jay_Diddy_B]

Hi,

What are you going to measure the LM399 reference with during the characterisation phase? you need an LTZ1000 class instrument. You also need low thermal leads etc.

The Solartron 7081 uses this technique in the reference. There are two reference diodes in series in the DMM. A pile of reference diodes were characterised, for several months, separately then pairs were selected, so that the combination has some desirable traits. The meter then compensated for ambient temperature using software. I have one of these meters and it is horrible. It may be o.k. in a 23C calibration lab. It takes a long to stabilize in a normal environment, compared to a heated reference which is stable in a minute or two.

Regards,

Jay_Diddy_B

[MisterDiodes]

RHB:

Not trying to discourage - but more areas you might want to have a think about:

What are your plans and equipment list for a robust measuring system for checking Vref/ resistor drift with low uncertainty??  Groups of LM399's still drift, and if you've ever tried this in the real world any group over 9 begins to offer very minimal returns on noise reduction - and don't forget every resistor in the averager contributes noise as well.

Preferred method of measuring a resistance most accurately is via a resistance bridge, manual or automatic, using a calibrated standard resistor as its reference.

More typically for measuring down to low PPM you'll need something like an 8.5 digit meter, and because those are not long term transfer standards you'll need that backed up by typically 3ea 732a/b Vrefs (at least one in current calibration) and/or SR-104 for uncertainties down to around 2ppm...IF you're interested in calibrated & traceable absolute value measures.  For sure you'll need a system of that caliber to verify long term drift of a Vref accurately if you're claiming 4ppm /yr drift rate.

Always keep an eye on the daily drift & uncertainty of your measuring equipment while you're measuring your Vrefs & resistors as well - if you're trying for  Vref of 4ppm / year on your calibrator box, you'll need sub-ppm drift per 24hrs, and that's something to consider even on a 3458a (check 24hr uncertainties):

http://literature.cdn.keysight.com/litweb/pdf/5965-4971E.pdf

Plus you'll want a low thermal EMF scanner system to test LOTS of resistors of different values, power levels, environmental effects and across several manufacturing batches...And so on.  This could take years.

It still sounds a lot easier to me to buy a few good resistors to put in your Cal box and achieve a more robust, lower noise and more stable end Vref result - but to each his own.

[zhtoor]

hello,

could you please elaborate on this?

The use of the Fourier transform has a rather significant advantage in that it  should permit making readings to 8.5 digits with a 5.5 digit DMM by taking a sufficiently long sample.   There are a number of mathematical constraints which must be satisfied, but this is often the case.

best regards.

-zia

[rhb]

There are a couple of considerations;

If we take the data into the frequency domain then we can characterize the 1/f noise at moderate frequencies to estimate the noise at lower frequencies. That permits setting bounds on error which are tighter for any given measurement period.  There are other methods such as plotting Allan deviation that serve a similar function.

The main reason is it  avoids the truncation and aliasing of signals which are not related to the power line frequency by integrating the measurements.  Instead one takes a long sample, applies a Gaussian taper at the edges and Fourier transforms the series.  The DC value of the transform is the least squares (L2) estimate of the true value.

I would attribute that not being done in the 3458A simply to historical practice in precision voltage measurements.  It's quite easy to build an analog gate and counter  that triggers on the mains zero crossing.  Almost limitless processing power came along, but seems to  not have  permeated precision voltage measurements yet.  Customers are expecting PLC gating, so Keysight provides it.

Going beyond classical Fourier analysis an interesting application of sparse L1 pursuits would be to discard about 80% of the samples collected at regular intervals.  That completely eliminates the need for an anti-alias filter ahead of the ADC thus eliminating the errors associated with those components.

At present it appears that I can characterize a voltage reference and buffer amp with a first order  linear approximation:

expected_value = vref_initial_value + ref_aging + vref_1/f_noise + vref_temp + amp_initial_offset + amp_aging + amp_temp + r1_initial + r1_aging + r1_temp + r2_initial + r2_aging + r2_temp + r3_initial + r3_aging +r3_temp

All of the terms in the linear approximation are non-linear.  However, a sparse L1 pursuit will allow selecting the best choice for each from a very large number of possible non-linear approximations for each term.  That's the reason that I have a rather more optimistic expectation than others.  Aside from being a good bit of work to setup the equations the first time, I see no obstacle to having 10,000 or more approximations of each term above from which to choose.  The actual computation associated with a calibration would take at most a few minutes using a sparse L1 solver on a PC.  Prior to the work of Candes and Donoho, the only known solution was an L0 exhaustive search which is NP-hard.  The universe would end before you found a solution. For the case stated above there are 10,000**16 total choices for the solution.  That is a *very* large number.

Hopefully this will make the project motivation a bit more clear. My primary goal is to improve precision voltage reference performance.  The low cost, OSHW design constraints are to make a better standard available to hobbyists.  The mathematical solution and software applies equally to both a national lab class instrument and the hobby instrument.  I don't need a fancy set of references.  If I can calculate the expected value of the differences among 3 references which is constant with the bounds set by the 1/f noise and tighter than current estimation techniques my work is done.  I *may* build a higher precision reference, but I don't need it to develop the mathematics.

[TiN]

So what you need to get started?

[hwj-d]

So what you need to get started?

A big kick, to come away from mirror fighting   

[rhb]

Mostly free time.  Long term data such as Andreas has presented, but in ASCII.  I can  do quite a lot with the data you have collected already. I've only spent a few hours with the data.  It will need weeks to fully exploit.  This is looking for shadows within shadows.  Or peeling layers from an onion to use Bob Pease's metaphor.  Identify and correct the biggest factor, then the next biggest and so forth.  Bob wrote about it in the context of measurement probing noise, but it applies here as well.

For each of the terms which are a function of another variable I need to get a curve shape which is a close approximation to the device behavior.  The high precision resistors are carefully engineered to have nearly linear tempcos.  While very useful in practice, it's very hard to accurately find the underlying curve form.  Some of the better low end Vishay metal films  with low sensitivity to humidity and mid-range tempcos are probably a good choice for establishing the functional form.  Looser tempco precision metal foil also.  Some precision foils under varying pressure.  In particular, run some metal foils from high atmospheric pressure down to 40,000 ft pressure.  That would be very useful for applying this to a high precision traveling reference.  There is a known and documented hysteresis effect.  But I think it's just documented for increases in pressure, not decreases.  It would be very useful to know if this is a one time  change or if the change is cumulative with the number of pressure cycles.

I can't see any reason we can't have sub-ppm traveling standards.  I can think of *lots* of reasons we don't.  To date the efforts have been focused on making better parts.  I think we are pretty close to the limit of what can be accomplished doing that.  So we need to better describe how and why references change as a consequence of traveling.  If we can accurately describe the cumulative effects of temperature and pressure cycles and age we should be able to correct for it.

Success is reducing the residual error after correction to quantum level noise.

The flat aging curves of the used LM399s I have will make finding the functional form from those difficult if not impossible. Once I'm set up and able to collect good data, I'll replace all the LM399s with new devices so I can measure the first 2000-3000 hours.  A sample of 3 will have a large variance, but it's better than what we have now.  A modest amount of cooperation from Keysight, Keithley et al in collecting early aging data on a large number of voltage references would nail down the mathematical form of the aging curve quite tightly.  However, that involves having tens of thousands of aging curves.

Sparse L1 pursuits are not widely known among the professionals I know.  I can only think of a handful of seismologists.  So I see an opportunity to make a significant contribution to metrology.  I think it would be pretty neat for a bunch of amateurs to pull something like that off.

In 1948 Scientific American published a long series of articles about the construction of a new ruling engine under the leadership of John Strong.  In the article it was mentioned that a number of amateurs had destroyed their health and wealth attempting to build a ruling engine.  In 1958 an Amateur Scientist column described a ruling engine built as a hobby project by a gentleman in England.  He was a professional laboratory instrument technician and builder, so "amateur" may have been a bit of a stretch.  Strong's team followed traditional practice,  A screw made to an accuracy of the order of a millionth of an inch, a vast room size vat of oil held at constant temperature. A human in the room for too long would ruin a grating by perturbing the oil bath temperature.

The amateur project used a cast iron box in which he placed a cadmium lamp interferometer, the best screw he could readily obtain and the ruling carriage, etc.  He counted fringes to position the carriage.  Stick slip made the errors random, so the periodic errors that produce Rowland's ghosts were absent.  Despite being in a sealed cast iron box, he was limited to certain times of year when he could make gratings and the machine had to be in a room on the north side of the house.  But he made gratings reported to be as good as any and better than most.

[rhb]

I'd like to suggest people read the introduction to this  paper:

https://statweb.stanford.edu/~donoho/Reports/2004/l1l0EquivCorrected.pdf

I consider it the most important piece of applied mathematics since Norbert Wiener's "Extrapolation, Interpolation and Smoothing of Time series".  There is no need to read the mathematics unless you like such things.  Donoho states the important aspects of the paper in the abstract and introduction in very clear language.

The single pixel camera is a different case. It exploits the mathematics to convert multiple time samples into multiple pixels.  You can find various versions of the paper online.  It appeared in IEEE Spectrum as I recall.  It's really quite cool as they have some pictures.

In the context of a voltage reference one is solving an inverse problem which is an important application of sparse L1 pursuits and what I was doing when I discovered their properties.  However, Foucart and Rauhut only give glancing attention to that.  Candes wrote a paper on Dantzig selector which is essentially the equivalent of what I'm proposing.  One is then using the inverse function as a predictor of future values.

[ArthurDent]

 You can also save a great deal of money on the measurement side of your project. A lot of us can’t afford a bank of HP 3458A meters to fully characterize a $30 hobbyist voltage standard. Other threads and videos have mentioned a feature of the moderately priced HP3457A that can be used effectively. The HP 3457A is sold as a 6.5 digit meter but a lot of people know that the output on the rear panel will give you one more digit (https://www.eevblog.com/forum/testgear/the-mysterious-%277th-digit%27-(hp-3457a-dmm)/ ).

 What I found through lots of experimentation was if you loop this output to the front input of the HP 3457A and run it through the meter one more time you will gain yet another digit. Great way to save money on a high resolution meter and it fits in well with this voltage reference project.

[rhb]

My hope is to design a multirange  calibrator which travels well and can be calibrated inexpensively which is good enough for a 5.5 digit DMM .  The idea being something that can be sold calibrated to a 3458A automatically and which will improve in accuracy over time as additional calibrations are performed.  I have in mind a retail of $100-150.  The DMMCheck Plus I ordered is essentially what I have in mind but more economical to build and maintain as well as better long term accuracy.

Underlying all this is notion of modeling device behavior as a means of improving accuracy using sparse L1 pursuits.  This thread is really about the  engineering tradeoffs for a mid-range calibrator.  While the techniques are the same, a high precision reference requires a more refined device model..  I think it might be useful to discuss applying sparse L1 pursuits as a separate topic in the context of high precision references

[Theboel]

Dear All,
the discussion is really interesting I am not skeptical nor optimistic about price barrier but I think I can learn something if the project can be done in real world.
for me use the best under sun resistor married with the best reference You can buy is easier than made it to be best reference box some one can do.
but may I suggest to use more "acceptable" for some people than plain XYZ resistor what about use PWW from edwin G pettis the cost is not extra ordinary and the performance is not the worst thing we have seen. Yes some people has "bad experience" about edwin resistor but I do not think thats end of the world I am more optimistic if LTZ1000 with Edwin res will more acceptable, "easier" to deal with, more data already floating around even if You need a special format data some one can and will help.
just my 2 cents opinion.
Anton

[eurofox]

I have the intention is to apply a thermal treatment (annealing) to remove the mechanical stress after the machining work, I have a oven and equipment for this as well the knowledge to do it.

With respect to filling with inert gas, it cross my mind but somehow at least in my case I have a buffer to output 10V and need to be tuned in the metal container after aging, this mean there need to be an small hole just above the trimmer and need to be closed hermetical, not easy to do with gaz ....
I will mill under the PCB an area to place 2 packets of silicate gell, they will suck the humidity that could enter te container during the trimming job.

Braze an inner tube or tubeless tire stem to the hole for access to the trimmer.  Then use a metal cap and gasket to seal it after adjustment.  It does require having a tank of dry gas around, but I suspect you've got a bottle of argon or argon/CO2.   Nitrogen is just cheaper.  Assuming everything is dry, the gas just displaces any moist air.  It doesn't need to be under pressure, though it will if the atmospheric pressure drops.  I'm suggesting a tube you can poke into the unit to inject the gas near the bottom.

In fact it is very easy to fill with argon, argon is more than 1.3 times heavy than air this way it make it easy to fill the "container".
I will move the voltage trimmer outside of the container, wires wil pass outside on copper screws molded in epoxy.

I got the PCB in the mail today, got almost al components including resistors from Edwin Pettis, just need to order the LTZ1000.

eurofox

[rhb]

I think keeping the trimmer inside is the way to go.  Topping it up with a shot of argon after adjustment is easy and it avoids noise pickup.  Also the argon will prevent oxidation of the trim pot.  So the temporal stability should be better.

[rhb]

A very interesting design.  It won't change the thermal noise, but it would reduce the current noise and it would result in more uniform temperatures for the resistors and generally smaller temperature gradients across the board.  It would also reduce the shift from turn on to stable operation caused by resistor heating.  Under assumptions of a Gaussian distribution the resistance tolerances are reduced by almost 1/3.

A very useful experiment would be to set up a voltage reference feeding a divider as presented here and a 3 resistor divider with all the resistors thermally coupled using silicone grease in a small thermally insulated enclosure. and collect data along with resistor ensemble temperature for a few days at the maximum sample rate available i.e. PLC = 0.  That would allow comparing a wide range of noise suppression techniques.

[rhb]

The sort of thing that @zhtoor suggested is exactly what I'm looking for from others in this effort.  This is cost -performance engineering.  The sparse L1 pursuits which I hope to employ to substantially improve performance are better treated separately.

EMI suppression suggests that the reference should be completely  isolated from the UI MCU.  That makes the reference a pure analog design project along traditional lines. This makes it very much easier as the important factors are well understood by most people in this group.

Revised outline:

precision reference
buffer amp
biasing  resistors
3-4 bit precision DACs in a Kelvin -Varley configuration so that each range has an approximately uniform set of steps over the full range.
serial to parallel shift register

single input supply line
single voltage output line
single input control line for selecting the output voltage step
single output line for temperature

careful EMI filtering on all enclosure penetrations
4 layer board to reduce noise propagation
smallest possible shielded enclosure filled with thermal grease

I have a rough idea what the issues are in constructing a reference, but I do not know all the details and I  have no knowledge of the various circuit designs which have been used in high end instruments.

So  can we design  to meet the above criteria for a $30 BoM?  That leaves $20 for UI, etc.  Conrad Hoffman's mini-metrology lab is the best analog to what I'd like to do with this project.  It's almost 40 years later which is a generation or two in technology. Can a team of smart people design a better version using more modern devices?  Can we make getting periodic NIST traceable calibration affordable?

Setting up a sparse L1 pursuit to predict and correct errors due to aging, temperature, etc is a major project in itself.  It might not work.  Based on my experience with sparse L1 pursuits that led to my spending 3 years slogging through a lot of math I happen to think it will make an order of magnitude or better improvement.  But it will take months to develop and several years to test.  I'd like to keep that discussion in the "Applications of sparse L1 pursuits"  thread.

Have Fun!
Reg

[thermistor-guy]

hello guys,

how about starting with this lm399 based scheme (initial proposal):-
...
2. the divider resistor r1 is 10x 10k resistors in parallel. https://www.digikey.com/product-detail/en/yageo/MFP-25BRD52-10K/10KADCT-ND/2059114
3. the divider resistor r2 is 10x 100k resistors in parallel. https://www.digikey.com/product-detail/en/yageo/MFP-25BRD52-100K/100KADCT-ND/2059112
4. the zener bootstrap resistor r3 is 5x 100R resistors in series. https://www.digikey.com/product-detail/en/yageo/MFP-25BRD52-100R/100ADCT-ND/2059113
...

On the selection of resistor values: two non-ideal aspects of resistors are leakage current through the body/coating, and contact resistance.

Points 2,3,4 above are low-cost axial metal film resistors. So the contact resistance from the end caps to the metal film will change with age, temperature cycling, etc. I imagine leakage resistance will also change with age, temperature cycling, applied voltage, humidity history etc.

If, for example, the leakage resistance is around 1E10 ohms, and the contact resistance is around 1E-2 ohms, then the geometric mean of 10 kohms minimizes the impact of both factors (1 ppm each).

Contact and leakage resistance may change in ways that are hard to predict. These also may be, themselves, sources of excess noise. Is it worth choosing resistor values to minimize the impact of these two factors?

[rhb]

The datasheet for the suggested resistors shows that they have rather good specs for stability.  The use of multiple resistors in parallel reduces the effect of self heating which is a significant factor in aging behavior.  And the central limit theorem predicts a reduction of  the tolerance from 0.1% to 0.031% and the tempco  from 25 ppm/C to 8 ppm/C.  Not bad for $0.80 + $0.10 of PCB.

The goal here is the best possible voltage calibrator  which can be manufactured in low volumes and sold for around $100.  That requires astute engineering choices. Ultimately, suggested solutions will have to be tested before being adopted.

[cellularmitosis]

When the drift is predicted, will the output be corrected by adjusting the circuitry, or will it print out a correction value which the user applies on paper?

If you are going to adjust the circuit, you could use jfets as variable trimmers.  On my long-term to-do list is a Harmon divider with an “auto cal” button which uses these jfets to balance itself to an exact 10:1 ratio.

[svenskelektronik]

Hi guys.

I've been following this thread and discussing the matter by mail with rhb. As a digital guy I'm enjoying this project as it's brings something new to the table for me. My motivation to participate and follow this project is also due to my interest in OSHW where I want to test some development techniques widely used in FOSS with the hope to make OSHW more widely adopted.

I've made a schematic drawing draft in KiCAD based on the KX LM399 which has been uploded to https://github.com/svenskelektronik/lm399_precision_voltage. Please consider to fork it and submitt your changes by a pull request. The design uses the advatage of parallell resistor banks as suggested by Zia, please feel free to express your opinions on how many you think is sufficient to use. Another thing on the to do list is to determine what resistor to use, MFP Series as suggested by Zia seems like a good candidate. We may start just determine the footprint and then later descide the resistors. Me and rhb suggests the removal of the second LM399 in the design.

[TiN]

And not a single link to a source? That's kinda sad move.

I had some comments on my schematics (and why it's not very useful in perspective of this thread) but I'll hold my thoughts thanks to this. 

[rhb]

I had in mind simply displaying the "correct" value of the output rather than disturbing the operation of the circuit.  To my mind the reference should be  as simple as possible.    I've set a very difficult BoM cost target.  Autocorrection would require a JFET and DAC.  That costs more money and complicates the system error model.

I'm quite impressed by the performance of the basic-lm399 references I was given.  I don't yet know about long term accuracy, but they certainly seem adequate for calibrating a 3478A with a modest amount of signal processing.

The multiple resistor improvement in accuracy and tempco should be 1/sqrt(N) for random variations.  The reduction in current noise for parallel resistors should also follow 1/sqrt(N).  We should calculate the cost to reach a particular level of precision using various parts accounting for PCB space and placement labor.  Reducing the current in each resistor also improves the aging behavior.

I'm unclear what the best approach to implementing the output voltage divider is. I have in mind approximately equal steps across each range from 10 V to 100 mV max.  It would be nice to have a 10 mV fullscale, but I rather fear that would be too noisy to be of value.

If the device were mounted on a dual banana plug in a well shielded enclosure on battery power so that in use it was mounted on the front panel inputs of a DMM would a 10 mV fullscale range be possible with acceptable performance?  Use an external PS for warm up and switch to the battery for the actual calibration.

[TiN]

Usually DMMs don't have 10mV fullscale, that is realm of nanovoltmeters. nV-meters also rarely banana plugs, as these are not optimal for low-level VDC measurements.

[rhb]

I'd like to suggest that the serious thought be given to placing the critical resistors on a daughter board if that is practical.  That way we can experiment with different resistor options, e.g. PWW, etc, without having to rework the board layout and without using excess board space if parallel resistors are not used.  Perhaps use the footprint of the largest precision resistor anyone would want to use as the daughter board footprint and mount the board at 90 degrees to the main board.

[Andreas]

Hello,

the absolute maximum ratings allow only 0.1V in forward direction for the substrate diode.

Note 2: The substrate is electrically connected to the negative terminal of the temperature stabilizer. The voltage that can be applied to either terminal of the reference is 40V more positive or 0.1V more negative than the substrate.

So perhaps another dead LM399.

with best regards

Andreas

[Andreas]

I understand the datasheet in this way:
no pin of the reference is allowed to be more negative than Pin 4 (substrate) by more than 0.1V.

[cellularmitosis]

Very interesting idea zhtoor!  I am curious to see if the 399 survived.

I was looking at the Spreadbury paper again recently and was reminded that these refs don’t seem to drift much while turned off (and I seem to remember reading in another thread that Dobkin said this is also true of the 399).  Perhaps the poor man’s route is to have an additional 399 which is only turned on one day per year?

[Andreas]

Very interesting idea zhtoor!  I am curious to see if the 399 survived.

The question is:
is there enough "before" data (zener voltage, noise, ageing rate ...) to judge any abnormal behaviour.

[cellularmitosis]

Very interesting idea zhtoor!  I am curious to see if the 399 survived.

The question is:
is there enough "before" data (zener voltage, noise, ageing rate ...) to judge any abnormal behaviour.

I would be happy to offer up a few silicon sacrifices to find out if a low-temperature 399 is possible

I just need to find a way to retire early so that I can devote more time to volt nutting 

[rhb]

I'd like to repeat my plea for data in ASCII format, especially data from from start up of a new device.  To date I've received very little data.  I'm going to digitize the plots from the HPJ paper on the 3458A, but that is just because finding the data 30 years later is unlikely.

I feel pretty good about the results I got for the thermal noise spectrum,  They seem reasonably plausible.  So I'm on to the 1/f noise model.  But without documented aging data, I can go no further.

[rhb]

It seems to me that the heater should be attached to an aluminum enclosure say 6 mm thick with 50 mm of insulation.   Placing the heater on the LM399 makes a large thermal gradient.  Self heating makes  gradients unavoidable, but they can be minimized.

I'd been working on this off and on all day.  I'll just include it as is.

Some random comments.

The resistors are as much of a problem as the zener.  The critical parts of the reference should be in an enclosure made from 5-6 mm aluminum.  Two U shaped pieces with external heaters attached to two faces and embedded in insulation should be inexpensive and provide a uniform temperature.  Place some thermistors in suitable spots to check for gradients, at least for the prototype.

One might let the user input the expected maximum ambient temperature and set the chamber temperature based on that.  But that would conflict with fitting an aging curve calculated from the first 500-1000 hours of operation.  Aside from temperature, that seems to me to be the largest error source.  There is the question of whether it is better to burn in at elevated temperature or at operating temperature.  I'm skeptical of having a stable aging curve with high temperature burn in, but it might not matter.

Interestingly, a 40 x 40 mm Peltier device is under $3 each on Amazon for 10.  For a shallow enclosure, two might allow setting the operating temperature independently of the ambient temperature.  It's a big power drain, but it might be tolerable for a lab reference if the operating temperature were average ambient.  Quite a bit more complicated to design.

Such an enclosure is easily made using a 20 ton hydraulic press and a very simple die set.  Anneal  before and after forming and then mill the mating surfaces, drill and tap.  Seal the enclosure with butyl sealant to eliminate humidity effects.

In reading back through the thread I realized I'd never properly answered the question about how you get 6.5 digits from a 5.5 digit meter.   You deliberately add random Gaussian noise which exceeds the resolution of the 5th digit.  You take two long measurements and crosscorrelate them.  The noise does not correlate so the cross correlation only sees the DC value.  It's apparently called "dithering" in metrology.  The limitation is the length of time it takes to make a measurement.  TANSTASFL.

The early part of the aging curve has the most curvature.  It might be that lowering the operating temperature would extend the duration of the large curvature region.

My original premise was that if you built a device of the best, low cost, parts and then measured the behavior during a 500-1000 hour burn in, that you could describe future performance sufficiently well to pick up an order of magnitude improvement.  Such a burn in would include temperature excursions to quantify resistor TC, tolerances and aging. in addition to the reference and buffer amp behavior.

Fluke believes that the observed hysteresis effects are due to stress on the die and have patented temperature annealing the die on cold startup on the 7001.  I'm not convinced that is sufficiently innovative to warrant a patent, but patents are really just a license to sue others.

[cellularmitosis]

i would think to maybe boil a bunch of lm399's or give them some kind of an un-powered cyclic thermal pre-treatment which takes care of
this factor before going in with electrical burn-in / characterisation. (anybody here with die packaging/attach experience? MisterDiodes maybe?)

This approach seems to have worked in 1960 for regular zener diodes.

[montemcguire]

That basic process could be called 'annealing', and as you state, it can be done by thermally cycling the assembled circuitry, powered or unpowered. Aside from stresses inside of the actual semiconductor devices, there are also sources of mechanical stress caused by soldering during assembly. The thermal coefficient of expansion of copper foil is generally different than that of the PCB substrate and any of the devices soldered to the PCB, so if a PCB is being reflow assembled, the solder joints will harden first, and then all of the components will cool with usually different thermal expansion coefficients, introducing mechanical stresses to the assembly. By temperature cycling the assemblies close to the plastic temperature of the PCB substrate, a portion of these stresses can be relieved, probably minimizing long term drift caused by mechanical stress on the components.

Another approach would be to use conducive epoxy to assemble a circuit, so that the problem of the solder solidus temperature being so far above the working temperature is avoided. Epoxy can bond near the thermal set point temperature, so there's no reason why the assembly process needs to 'build in' mechanical stresses to all of the components.

[z01z]

The paper and the link mentioned here might be of interest.

[rhb]

The kitchen oven and freezer would be more convenient than water. There is also the question of reducing the excursions with time as shown in a Fluke paper about the 7001 in the reference teardown thread.  If you have hysteresis, you need to do that.  Consider demagnetization.

Once the curve flattens out, it becomes difficult to determine the deviation from a straight line.  The computation becomes very sensitive to numerical precision.

A physical model for the drift and the hysteresis would be very useful.  Variations due to stress imply piezoelectric effects so far as I can see.  Is there literature on that in solder joints for example?

A man was walking down the street when he saw another man run out, stomp his foot and then go back to leaning against a building.  As he approached, the man did this several times.  When he reached the man, curiousity got the better of him. 

"What are you doing?" 

"Keeping away tigers." 

"But there are no tigers around  here." 

"Works very well, doesn't it?"

[rhb]

One might consider building a temperature chamber with a pair of Peltier devices so one could raise or lower the temperature as desired.  That would make acquiring data on hysteresis  fairly simple.  Run a group of LM399s with the heaters off and measure differential voltages among the set for a variety of thermal cycles.  Put CPU heatsinks and fans on the Peltier devices to improve heat transfer.

That was an interesting paper, thanks.  There ought to be more data available.  If you get any of the references from the thesis I'd like to have them.

Reg

[cellularmitosis]

hello,

update:

the lm399 has "survived" a week of operation and the measurements are substantially the same. ie;

Tamb = 29 degC
Vref = 7.00390V (main reference)
Vhd = 0.560V (heater in temp. sensor diode mode)

so it does look like a valid mode of operation.

has anybody duplicated the experiment?
has anyone seen an LM399 being used like this?

best regards.

-zia

I definitely plan on duplicating locally!  As soon as get around to it    

[Andreas]

Hello,

you cannot calculate 1/f noise (uVpp below 10 Hz) from noise density data (uVeff above 10 Hz).
Below the 1/f corner frequency the noise density increases (by 1/f).

The stray on LM399 noise is very large.
Good devices have down to 4uVpp 1/f noise (0.1 .. 10Hz)
Not so good devices are above 10uVpp.

Below 0.1 Hz (there where it hurts most) the noise is not specified.
This is the area of popcorn noise with jumps with ppm-Level for some minutes or hours.

best regards

Andreas

[Kleinstein]

To get 4 good LM399 one would need to buy considerably more than 4 units. So there would not be that much cost advantage compared to the LTZ1000 (if not using super high grade resistors).

The LTZ1000 1/f noise corner can be well below 10 Hz. This means the 0.01 Hz to 1 Hz noise can be considerably better than 0.1 -10 Hz noise. For the LM399 the 1/f corner is higher and thus 0.01 Hz-1 Hz noise is expected to be similar to 0.1-10 Hz noise. So even if one averages enough LM399s to get the same 0.1-10 Hz noise level, the LTZ would be still better at 0.1 Hz. Better performance of the LM399s at 10 Hz is not that relevant.

It would be rather difficult to get the LM399 to a stable 25 C. To get good stability without the internal heat, if would likely need a two layer temperature regulation. With just heaters this would be more like 50 C as a minimum. Going  below environment is tricky due to possible condensation.

[Andreas]

Of course I tested that with a very good device.

now tell me which of the randomly choosen measurements have the heater on and which have the heater off 

All measurements with 0.5uV / div and 1 sec / div.

[Andreas]

that was not the question.
This is a comparison between heater on and heater off.
So some measurements are with heater on others with heater off.

[Andreas]

where is your factor 2 in difference for the heater on or off?

this may be valid for the noise density of the wideband noise.
The difference in 1/f noise is much smaller.
But you cannot decide it by a single measurement since the stray from 10 sec to 10 sec window is also very large.
Perhaps you get 20% increase by "heater on" in average. but the stray between measurements can be a factor of 2.

A not so good example (LM399#2 with 1uV/Div) (that does not mean that this is the worst case)
and some typical examples one with one edge of popcorn noise (which is filtered mostly away by the high pass filter).

most of my LM399 are good old National Semiconductor.

Also measured a newer LTC LM399 (last picture also with 1uV/Div).

By the way: #3 and #4 are with heater off.

with best regards

Andreas

[cellularmitosis]

Hmm, so the lower heater temperature benefit is now only a question of possibly slower ageing rate?

[rhb]

Collecting data with heater on and off should allow determining the coefficient of the thermal noise for the devices as well as the coefficients of the 1/f noise.  The different slopes of the amplitude spectra make them nearly orthogonal and thus easily separable.

If one used an oven, one could derive a more refined model for the thermal noise by making measurements at multiple temperatures.

I've not yet considered the problem of popcorn noise.  My 8648C arrived yesterday and my 5386A on Monday, so I'm comparing them at the moment while I reorganize my lab/closet.  Next up is to reference each in turn to my GPSDO and trim them and the 8560A .  After that I need to build a distribution system for the GPSDO reference signal.  But shielded and filtered AC power comes first.  I may well go over the top and encase everything in steel from the IEC socket on the instruments all the way to the entrance panel.

I ordered 3 AD584 "precision" voltage references from eBay for $3 each.  These are the J parts with 3000 ppm initial, 30 ppm/C and 25 ppm/yr or thereabouts.  I'll set them up with the well aged LM399s and collect differential readings to see how well I can predict the output voltages at 2000 hours from the readings over 1000 hours and the output at 4000 hours from the readings over 2000 hours.

[rhb]

Not actually what I've been talking about, but very similar and  dead simple to do.  The noise is a boxcar in time so it is a sin(f)/f in frequency. That is very different from the 1/f decline and the growing amplitude of the aliased thermal noise.  I'll try to get a simulation put together in the next few days of what the aliasing looks like to measurements with an integrating ADC with PLC=10.  Thermal and 1/f can be done with L2, but popcorn noise requires a sparse L1 pursuit.  You build a large dictionary of possible popcorn events of various lengths and amplitudes and then find the set which best fits the data.

Unlike thermal and 1/f  noise, it might be possible to remove popcorn noise, at least under certain circumstances.