[volt-nuts] Best reference after LTZ1000

[WarrenS]

Marv Gozum

You don't need anything very fancy to take measurements like that I did, 
Just lots of time and CARE and more than one 10 volt reference.
What I did was compare TWO 10 Volt references at a time, showing the 
difference between them using a good null-meter and
recording its output with an analog type of strip chart recorder  (using a 
PC or rustrak)

One of the units (the reference unit) is held in a constant environment and 
the environment of the Device Under Test is allowed or force to change as 
desired.
Repeatable is the important thing. If you can not repeat the results, then 
they are not accurate enough to be useful.

And beware,  Burn-in is not hours, sometimes it is more like months or even 
years.
Whatever it takes to be sure it has fully stabilized and settled down to do 
whatever it is going to do.
I generally do not even start looking at them until I've tortured them 
(carefully) for a month or so.

WarrenS

********************
********************

From: "Marvin E. Gozum"
Subject: Re: [volt-nuts] Best reference after LTZ1000

This is what's needed.
... snip

WarrenS has just posted a dataset similar to what I am still
collecting.  It has been done!  But his reflects special selection of
the AD587 chip and a modification to it he describes in an old post
at time-nuts.

http://www.mail-archive.com/time-nuts@febo.com/msg16518.html

FWIW, I am collecting preliminary data before connecting my PC and
doing something similar to WarrenS, so I can get a feel for the
response.  I make several thousand automated measurements daily,
covering 75-80F, 35-50% Relative humidity, over the past
week.  However, I am using a 3456A and have no GPIB adapter so I
cannot download the data, so I am doing it the hard way, allowing the
HP to do statistical analysis, and recording the variance for fixed
time intervals, and recording temp and humidity manually.

My curve is similar to WarrenS but I can only resolve 10uV.  Just
quickly, my stats as I type, my standard deviation is 15uV from a
mean of 10.000 05 Vdc.  Thus, my 95% confidence interval is 10.000 00
to 10.000 10, and this includes the internal error of the HP 3456a
too.  This is a fairly consistent response over 1 week except the
mean will travel as the temp deviates from the 73F when Geller
calibrated the SVR.

It is fairly crude compared to WarrenS's results, but it has a
similar response. I cannot be sure which contributes to drifting
more, the HP 3456a, the Geller, or both!

But overall, as you see, the reference is not shabby, and now 1 week
since purchase is still reading true.


At 06:43 PM 8/24/2010, Andrea Baldoni wrote:
> > measure it as soon as possible, lest it drift!  Sure enough, under the
> > same temperature conditions, its accurate to my calibrated 8456A,
> > reading 10.000 00 Vdc.  I am now gathering stats on its performance.
>
>You are doing an interesting work. Interesting also would be
>to buy a barebone chip and measure it untrimmed.

As WarrenS comments, a reason for getting the assembled board is a
need to burn in the assembly and the chip for hours, I forget if
Geller does 10 or 200 hours before calibration, then calibrate
it.  The SVR calibration includes both trimmed and untrimmed
calibration values, so if you think the pot has gone bad, you can
remove a jumper to detach it from the circuit, and measure the native
reference output directly.


Sincerely,
Marv Gozum
Philadelphia, PA