ahhh typos : "Early in the development of this board, we used AD587JN
or LN reference"
should read AD587 JN (20 ppm / c) or KN (10 ppm / c). -Joe, KO2Y

wrote:
Some time back, there was a discussion of voltage references. The post
is now in the public domain forever, so I feel a need to reply.
Unfortunately I only saw it recently so I wasn't able to comment at the
time. There is an incorrect statement regarding one of our products.

Early in the development of our most inexpensive transfer standard we
did loan a board to an amateur. He compared it to his Fluke 731B and
observed a change in reference voltage with temperature. He now offers
his "spread sheet" in public asserting in this forum that our reference
boards do not perform as advertised: "If you can duplicate his
temperature you will do pretty good but probably not quite as good as
he advertises."

Our inexpensive "SVR" boards are essentially high quality support
boards for the AD587 reference chips. The specs. (including
temperature coefficient) are largely determined by the specs. of the
reference chip itself. Then it is important not to further degrade
that tempco when adding trim resistors. Also, for the purposes of a
trimmed transfer standard there should be a sufficient trim vernier to
trim to a desired resolution without losing precision.

Early in the development of this board, we used AD587JN or LN reference
chips with a temperature coefficient of 20 ppm / c (JN) or 10 ppm / c
(LN). For a 10 Volt chip, 1 part per million (1 ppm) is one millionth
of 10 V or 10 uV. Therefore an AD587 KN based reference can move by up
to 100 uV per degree Centigrade and still be perfectly "within
specification" (10 ppm over 1 degree c is 10 * 10 uV or 100 uV).

We now use the AD587 LN grade reference chip rated at 5 ppm / c. We
have also improved the tempcos of the trim resistors, now using
precision IRC 0.1% metal film resistors (for the reduced tempco, not
the precision and note that the tempco of the trim resistors is not 1:1
because of the OpAmp circuit in the reference chip).

It should also be noted that most reference chips change reference
voltage with time; usually this appears in a specification sheet as ppm
/ 1,000 hours. For example the AD587 series is spec'd at 15 ppm in the
first 1,000 hours of operation. This is one of the reasons that we
advertise this product as a transfer reference.

Our inexpensive "SVR" boards http://www.gellerlabs.com/SVR%20M.html are
rated at 5 ppm / c and do that well or better. The previous post
suggesting that our products do not meet advertised specification was
an incorrect statement. We set the reference on our calibration bench
to a resolution of 10 uV and record the temperature. We conservatively
advertise a lower "transfer accuracy: Calibration data will be reported
to 10 uV resolution (.0001%). The last digit of resolution will only
be useful in the short term at the reported temperature and power
supply voltage (15V). Under these conditions, it is possible to obtain
short term transfer accuracy of better than +/- .0005% absolute. The
AD587 can move 15 ppm during the first 1,000 hours. The temperature
coefficient is 5 ppm / C or better as limited by the AD587LN."

The previous poster's assertion that this specification is not
obtainable is without merit and incorrect. We have never compared our
SVR boards to a Fluke 731 that initially cost $5,000 and has now
enhanced stability with typically 10 to 30 years of run time. Our
boards can however be very useful to check the absolute voltage as a
transfer standard with minimal attention paid to ambient temperature
and good test practices.

Joe Geller, K02Y
you can contact me by email at joegeller at gellerlabs.com