USA CAl Club: Round 3 suggestions

[rhb]

I'm all too acutely aware of the propensity for engineers to use Excel.  I've also had to sit on the tarmac while they rebooted an Airbus plane.

[SilverSolder]

The software revolution is definitely both good and bad...   (mostly good, I would add!)

[tomato]

H.-J. Sun, Kaoru Fukuda and B. D. McCullough
"Inaccurate Regression Coefficients in Microsoft Excel 2003: An Investigation of Volpi's Zero Bug"
Computational Statistics 32(4), 1411-1421, 2017

Who publishes a paper about software that's 14 years (and 4 generations) out of date? 

[bitseeker]

Academic papers, journal articles, etc., are about digging into subjects of interest. Time is not necessarily a factor.

From a practical standpoint, though, it would be nice to know how many of these issues still exist. Having used Excel since its very first version, I've seen bugs linger for decades, so anything's possible.

[tomato]

Academic papers, journal articles, etc., are about digging into subjects of interest. Time is not necessarily a factor.

Time may not be a factor, but academic papers should be timely. 

From a practical standpoint, though, it would be nice to know how many of these issues still exist.

Yes, that really is the point.   A paper published in 2017 should have looked at Excel 2013 or 2016, not Excel 2003.

[CatalinaWOW]

Private individuals and universities with limited budgets may well still be using Excel 2003.  Not wanting to pay for the upgrades.  Or not wanting to learn a new user interface.

You only have to read this forum to find numerous individuals using software dating back to the start of this century for a variety of reasons.

[rhb]

It might be appropriate to read the paper and find out why it was written before making statements based only upon the title and when it was published.

I'm not aware of a statute of limitations that says that an error in a paper published over x years ago  is no longer an error.  Authors have been known to publish erroneous conclusions.  The reason it is wrong might not be found until many years after publication.

It's quite common for grad students to find such errors during the work for their dissertation and be obligated to correct the prior author in the process of supporting their work.  So they have to write a minor paper to document the issue to satisfy their committee.

Everyone who gets a PhD is very familiar with the routine.  It is standard practice.  If you obtain results which contradict prior work, you have to prove the prior work was wrong and explain why.

[tomato]

... universities with limited budgets may well still be using Excel 2003.

Doubtful.  Regardless, analyzing software that is 4 generations old has limited utility.

[SilverSolder]

Would be good to have some simple,  easily verifiable worked examples that we could try in Excel (using whatever version we have) to check whether there are significant problems.

[rhb]

Would be good to have some simple,  easily verifiable worked examples that we could try in Excel (using whatever version we have) to check whether there are significant problems.

That's what Bruce did.  There are a number of canonical datasets in statistics which have been exhaustively analyzed.  He used some to test Excel.  I've asked if he has an Excel file I can post.

Here's a link to Bruce's homepage:

https://www.lebow.drexel.edu/people/brucemccullough

You can read the abstract of the paper that my personal troll, @tomato, complained about.  If you check his posting history you will see that his sole interest on the forum is harassing me.  Which must have been very traumatic when I went on vacation for a couple of weeks at the end of June and did not post anything.

The problem of irreproducible results is a very serious issue in the research community.  There are a lot of people clamoring for archiving all the software and data used in a paper.  It is depressingly common that  others are unable to replicate published work.  Jon Claerbout of Stanford was one of the first when he found that his grad students could not reproduce the work done by a previous student of his.

[tomato]

You can read the abstract of the paper that my personal troll, @tomato, complained about.  If you check his posting history you will see that his sole interest on the forum is harassing me.  Which must have been very traumatic when I went on vacation for a couple of weeks at the end of June and did not post anything.

My post had absolutely nothing to do with you.  Get over yourself.

[rhb]

A reply from Bruce.

The data are all available at https://itl.nist.gov/div898/strd/

To answer Tomato's questions:

(1) the bug (Excel giving inaccurate solutions to regression problems in certain situations) was known very shortly after Excel 2003 was released, but nobody knew why it happened

(2) the paper was started when Excel 2003 was in service

(3) when the paper was published, Excel 2003 was still in wide use     

(4) the interest??? This paper documents that Microsoft programmed Excel to swap out correctly calculated regression coefficients for inaccurate regression coefficients.?? This is not a case of the software incorrectly computing the regression coefficients. Excel calculated the correct coefficients internally and then returned inaccurate coefficients to the user.

Sure, @tomato.  What ever you say.    However, anyone can look at your posting history just by clicking on your user name and the "show posts" tag.

[tomato]

Sure, @tomato.  What ever you say.    However, anyone can look at your posting history just by clicking on your user name and the "show posts" tag.

Seriously man, get over yourself.  You've made 600 posts in 2019.  I've made 50 posts in 2019, and the majority have nothing to do with you. 

[Tj138waterboy]

I'd like to pose two questions:

Best way to extend the references to other ranges including things like the 3478A with its "3" leading digit scales?  Simple divider or divider and op amp?

Best way to produce a DC current source? Preferably multi-range, but anything is better than nothing.

It seems to me that those should be addressed before dealing with AC V & A.

Resistors seem fairly simple, more of them in a more robust package separate from the other references.

Would there be horrible consequences in adding a ina106 on the output as a switchable 11-1 divider as per datasheet example or would it make more sense just to roll your own?

[rhb]

After much thought on my part and little comment from TiN and Andreas, to whom the question was largely directed, I concluded that a 150K-100K-10K-1K divider made from good grade metal film or wire wound resistors with banana jacks at the nodes in the divider  is the most economical option if using a 7 or 10 V reference.  All additional electronics adds noise.

Metal foil would be nice, but aren't really needed.  What matters is stability, not precision values.  By adding small wound copper resistors in series the TC can be nulled.

At the moment a reliable RTC-T-H-GPIB interface is at the top of the list.  I discovered another Tempduino issue while working on my PX/FX ratio data yesterday. 

I've got all the parts I need for the logger now, but first I have to fix some bugs in a 15 year old research code I wrote for a major project for big oil.

[Tj138waterboy]

You seem to be working on 2 seperate dividers. One for px and one for fx. I was generalizing my question too broadly. From the spec sheet pg5 Referring to Figure 1, the CMR depends upon the match of
the internal R4/R3 ratio to the R1/R2 ratio. A CMR of 106dB
requires resistor matching of 0.005%. To maintain high
CMR over temperature, the resistor TCR tracking must be
better than 2ppm/°C. These accuracies are difficult and
expensive to reliably achieve with discrete components.

Am I correct to assume this is stating that the resistors inside of ina106 are matched to 0.005% and tcr tracking better than 2ppm? If so I would think that the tradeoff of an ots componet like this would at worst be a temporary solution in lieu of trying to find a good match through the hoffman mini lab method.

[rhb]

OK complete misunderstanding.  I was talking about an external divider to drop  a 7 reference to under 3 V which is the full scale for the 3478A and to scale voltages to lower scales.  This is part of doing a separate PX only track for lower end DMMs so as to reduce the length of time for a cal round.  I'm hoping this becomes an annual ritual.

I'm unclear what "ina106" is.  I'm also unclear which datasheet you are referring to.

[tomato]

I'm unclear what "ina106" is.  I'm also unclear which datasheet you are referring to.

An INA106 is a differential amplifier made by Burr-Brown/TI.  Google will give you a link to the data sheet.

[hwj-d]

I'm unclear what "ina106" is.  I'm also unclear which datasheet you are referring to.

An INA106 is a differential amplifier made by Burr-Brown/TI.  Google will give you a link to the data sheet.

Yes, and what a coincidence, it is (t)here   
https://www.digikey.de/product-detail/de/texas-instruments/INA106KP/INA106KP-ND/251075

[TiN]

I have a few suggestions for Round 3 :

* Allocate time-slot for every participating member, so there are somewhat scheduled in-outs. This can help to align calibrations if somebody has specific times when gear calibrated annually, so USAC travel refs can benefit from recent calibrations.

* Mandatory listing of equipment that each member have available for measurements. Also, calibration history/traceability attached to each equipment would be great to know. If the member does not have recently calibrated gear, that's ok, can list the latest known calibration instead.

Ideally, this information can be useful to estimate the uncertainty of each member over the round, and gain better confidence in travel transfers.

Another suggestion is to suspend shipments during the cold/hot season (e.g. ambient <+10°C or over +40°C), to avoid thermal/humidity stress to the references unless shipping out and shipping in members have high-end calibrated standards in-house to verify the shifts accurately.

[rhb]

I very much agree with TiN's comments.  It requires some prior planning, but nothing very difficult to accomplish.

Suspending all shipments during hot & cold periods will severely limit the number of cal runs.    However,  if we have a southern and northern tier hub we could extend the travel period by doing the northern tier in the summer and the southern tier in the winter with due attention to the weather forecast so that things are not shipped when the forecast is unfavorable.  I  think that there should be tighter restrictions on when the FX & PX reference kit is shipped.  If we have the three PX references I proposed that would not impede the schedule excessively.  USPS Priority Mail seems to be pretty reliable and NWS forecasts for the next 3-4 days also pretty reliable.

I was a bit disturbed to find myself receiving the  references in July  The inside of a truck can reach ~60 C in the summer in the southern US even when the outdoor ambient is below 40 C.  We had an office move when I lived in Dallas and all our Sun workstations spent the weekend in a moving van.  Not a single system would boot on Monday morning until I went around and reseated all the internal cabling.  The truck temperature is a wild guess, but when Motorola made the first transistor car radios, they were quite surprised at the number of failures.  Until someone went out to the parking lot,  put a thermometer under the dashboard and discovered it was  over 135 F (57 C)

I am working on a T&H logger.  It's been several years since I last messed around with MCUs so I've got a lot of paging to do.  But I have some FRAM based MSP430 dev boards which should be good candidates.  However, at the moment I don't even know what I have, much less the capabilities of the boards.  To add to the fun, the system disk I was using for MCU work died  and I did not have that work backed up.  So I'm starting from scratch.  But I'm hunting down RTC and T&H sensor documentation for the units I have on hand.

[CatalinaWOW]

I very much agree with TiN's comments.  It requires some prior planning, but nothing very difficult to accomplish.

Suspending all shipments during hot & cold periods will severely limit the number of cal runs.    However,  if we have a southern and northern tier hub we could extend the travel period by doing the northern tier in the summer and the southern tier in the winter with due attention to the weather forecast so that things are not shipped when the forecast is unfavorable.  I  think that there should be tighter restrictions on when the FX & PX reference kit is shipped.  If we have the three PX references I proposed that would not impede the schedule excessively.  USPS Priority Mail seems to be pretty reliable and NWS forecasts for the next 3-4 days also pretty reliable.

I was a bit disturbed to find myself receiving the  references in July  The inside of a truck can reach ~60 C in the summer in the southern US even when the outdoor ambient is below 40 C.  We had an office move when I lived in Dallas and all our Sun workstations spent the weekend in a moving van.  Not a single system would boot on Monday morning until I went around and reseated all the internal cabling.  The truck temperature is a wild guess, but when Motorola made the first transistor car radios, they were quite surprised at the number of failures.  Until someone went out to the parking lot,  put a thermometer under the dashboard and discovered it was  over 135 F (57 C)

I am working on a T&H logger.  It's been several years since I last messed around with MCUs so I've got a lot of paging to do.  But I have some FRAM based MSP430 dev boards which should be good candidates.  However, at the moment I don't even know what I have, much less the capabilities of the boards.  To add to the fun, the system disk I was using for MCU work died  and I did not have that work backed up.  So I'm starting from scratch.  But I'm hunting down RTC and T&H sensor documentation for the units I have on hand.

Any attempt to split into northern and southern tier clubs would have to carefully account for the shippers movement strategy.  As everyone knows all US FedEx packages go to a single point for sorting and distribution.  All of the other shippers have their own somewhat similar strategy.  The long distances in the US mean that cargo in air freight can to very low temperatures at any time of the year.   And finally, truck cargo compartments can reach truly amazing temperatures in the summer in all of the continental US.  Dry, clear air leads to really impressive thermal loads. 

[SilverSolder]

Is the temperature in the cargo hold of a plane as tightly controlled as the passenger compartment?  I have heard stories of items freezing inside checked luggage so perhaps large temperature swings are a normal "fact of life" of air cargo transportation?

Then there are the pressure differences experienced during flight - what is the influence of that?   Humidity changes? 

Mollycoddling a set of references could get complicated...

[RandallMcRee]

I think it was earlier in this thread that the point was made that things should be kept simple.

Its pretty much impossible to guarantee any sort of thermal control over packages short of herculean efforts.

In my view, we should just use this as an opportunity to investigate the possible effect that thermal changes might make on the PX and FX references. I sort of thought that is exactly why we return them to the gatekeeper each time.

One fairly simple thing we could do is a battery-operated temperature logger in the cal club package. Btw, the package itself is already well insulated. So there is that.

[rhb]

How often do we want to log and for how long?  An MSP430FR2433 with an RTC running draws 750 nA typical. So we should be able to log for the cell shelf life using a single CR2032 coin cell.

It's got around 15 KB of FRAM so we could comfortably store a 4 byte time, 2 byte temperature and 2 byte humidity once a minute to FRAM and then once a day copy the FRAM contents to an external flash memory device.  A 32 MB device would store almost 8 years of data.

The TI dev board is ~ 6 x 7.5 cm, however, a custom board could be made the size of a CR2032 holder.

This solution would have the advantage of providing a lifetime reference history rather than just transit logging.  By using a larger flash device we could store voltage reference data on it also at each measurement location.