[volt-nuts] Back to voltage was Re: Precision current source

Marv Gozum @ JHN marvin.gozum at jefferson.edu
Mon Aug 23 19:00:01 UTC 2010

Hello Bill,

Thank you so much for these pearls and the details!  I love the way 
you reference links.  A good idea!

I think the gist of your post is making a reference is going to be 
futile, because we, as individuals, do not have the resources to 
build a 732 like device, nor access to calibrate it against a 
Josephson Junction, JJ.

I think a reasonable facsimile or approximation can be done, its why 
I've joined this forum, to find others who have.  Read on.

But a short answer to the futility question is simple, buy a used 
732, put in back into working order, and find a lab willing to 
calibrate it against their NIST traceable references, and now, 
maintain your 732.  Now you have a reference well characterized by 
Fluke on the assumption, the rebuild has put it back into like 
original factory condition.

On your anecdote, while temperature drift is a concern, its minimized 
by ovens in the 732 to keep a stable temperature for the reference, 
regardless of ambient.  I do not know if more than one voltage 
reference chip is used inside each Fluke 732 or any equivalent solid 
state reference or others matching - or + tempco references, but if 
you can find out, I'd love to read of it.  Its common to use 
resistors with such characteristics, where they are applied.

What is typically done to most solid state references to improve 
their performance is apply a correction factor to the output every so 
many months to fix drift, and this can be done on any device based on 
a single reference once its characterized over time.  Here is such a 
process.  Its another freely available paper that was published in an 
IEEE proceeds but copied here, but there are others like it.


Historical data from Fluke show aged mature 732 model A trends +/- 1 
1ppm over a decade, so net over time its stable. Some of this data is 
abstracted in the 732 series user and maintenance manual available 
from Fluke's website, and it gives the correction equation.  There 
may be more recent ones, but this one is free, for those interested:


Regardless, a lot of circuit manipulation and other controls has not 
made solid state references as good as a JJ, over 40+ years of using 
zener based references.  This conclusion led to the development of 
the portable JJ now being studied as improved transfer reference.



A question for the user is really, how accurate is your need: e.g. 
300, 30, 3, 0.3 ppm?  For most professional production MetCal uses, 
the Fluke 732s [ or equivalent] suffices, and for others not.  For 
most users, a NIST traceable cal is acceptable, for others one needs 
a _direct_ cal from NIST itself.


Thus, if a less stringent 30 ppm or more suffices for one's need, why 
won't something like Geller's board work, and thus save the owner a 
lot of money?  However, it must be characterized just like Fluke does 
for the 732.  Empirically, such a home brew could rival a more 
expensive standard up to a ppm per time period, limited circumstances 
that a Fluke 732 cannot contend with given its intended use thus 
requiring the extra electronics, housings, redundancy and overall, 
cost.  Such circumstances may not even be related to stabilizing the 
volt, such as passing UL, CSA, IEC safety requirements, which add to 
the cost of the device.

Thus, what is missing in this noble and learned group's archived 
discussions [ and brought up in later post by Greg ] is collecting 
performance measurements of any device designed and used over time to 
ascertain stability, regardless of how the standard is constructed, 
be it as robust as the Fluke 732 or as spartan as a Geller 
SVR.  Thus, one needs to consider basic statistics on the output of 
these devices if one is to maintain them as a local reference.

I think electronic engineering uses the metric "reliability" to 
reflect stability, which is the combination of a devices' accuracy 
and precision _over_ time.

Looking forward to a stimulating and enlightened discussion!  Onward, 
to a home based volt reference on the cheap!

Some minor comments below.

At 07:26 PM 8/22/2010, WB6BNQ wrote:
>    Hi Marvin,
>    I have a friend, Art Rizzi, who was responsible for the Navys DC
>    voltage reference.  He worked at the Navys version of NBS at the North
>    Island Naval Air Station, San Diego, CA.  His lab was also responsible
>    for approval and acceptance of lesser standards used though out the
>    Navy.
>    In 1970, his lab received around 100 Fluke 731 voltage standards.
>    Though out 1970 and 1971 he spent considerable time characterizing
>    these devices.  One aspect, performance over temperature, was quite
>    important because but for his environmentally controlled laboratory,
>    the rest of the Navy saw quite a variation in temperature and humidity.
>    Art was the original developer of the idea of paralleling multiple
>    Fluke 731 voltage standards to account for temperature coefficients.
>    The basic idea is to take multiple units that had the correct mixture
>    of temperature coefficients such that they reduce the effects of
>    temperature variation as much as possible (within limits of course).
>    Arts efforts, along with other improvements he suggested, were the
>    basis of, and incorporated into, the development of Flukes 732 voltage
>    standard.
>    Clearly, each of the voltage standards needs an output stage that can
>    sink and source a given amount of current (a few milliamps) without
>    affecting its internal reference or the temperature of an individual
>    standards internal environment.  Ground paths within this ensemble are
>    very important.

Yes, but generally when testing a voltage reference or standard, its 
vital the DUT draw near zero amps.  A null voltmeter is preferred, 
but a very high impedance voltmeter can suffice.  So such capacities 
to sink/source are not usually necessary of a reference, long term 
stability is.

>    Creating such an ensemble for an experimenter is not an easy task to
>    accomplish.  Not withstanding needing expensive test equipment like
>    precise temperature and a very stable known reference to compare to, a
>    highly controlled lab environment and an environmental chamber are also
>    needed.  Then you need a few hundred voltage references to, hopefully,
>    find the ones that will accomplish the right mixture.  Finally you will
>    need 6 months to a year or more to do the work !  Did I mention money ?

You certainly need time to gather data, but it needn't be expensive, 
or as involved as you suggest.  For over 20 years, the references 
used inside high end DVM like HP 3458A or 3456A are typically just an 
LM399H or an LTZ1000 in a separate board very much like Geller's but, 
like wine, chosen, aged and characterized, will work without special 
environmental controls.  It will work better with environmental 
controls, but its not necessary given such errors are already 
accounted for when calculating the frequency of a device's calibration cycle.

>    However, to answer your question, NONE of the above does anything for
>    stability.  All it does is help to correct for temperature variation,
>    within limits of course.

Reliability, as a synonym for stability, is subject to many 
variables, temp is just one of them.  Humidity, and pressure also 
affect solid state references and have been studied too.

Thus, the reliability calculation and variability takes all of these 
variables into account as a phenomenon, assuming that time exposes 
the DUT to all variations the variables can throw at it in the 
measurement cycle, say 90 days, to 1 year.

The same can be done for something as cheap, and as exposed as a 
Geller reference.

>    Stability is an entirely different animal !  Many factors control
>    stability, some you can deal with and others you have no control over,
>    like the manufacturing processes.  So, you buy the best diodes or
>    reference devices you can afford and hope for the best.  You can take
>    it to the bank that Fluke and HP spend much time and money in producing
>    their top of the line products.
>    Bill....WB6BNQ

What I see others perceive is that looking for a single absolute 
reference, that is solid as rock, 1.0180000000000 or 10.0000000000 
etc., with ~ zero variability over time.  Even the JJ is not 
that.  JJ is simply better than zener, but it is subject to 
uncertainty too, now ~  1 pp billion, versus a good zener at 0.1 ppm.

So, why not 1 ppm or 10 ppm, or 30 ppm, as you funds or need allow.

Best Wishes,

Marv Gozum

More information about the volt-nuts mailing list