[volt-nuts] "averaging reference" / "Perfect Volt"

[Dr. David Kirkby (Kirkby Microwave Ltd)]

On 22 Dec 2014 08:51, "Joel Setton" <setton at free.fr> wrote:
>
> Frank (and all!),
>
> These are all very valid questions.
>
> The context is a hobby activity, and the purpose of the reference voltage
is to have something which is good enough to calibrate, for example, a
30-year old DVM after restoring or repairing it.

What class of DVM? Are you going to be restoring 30 year old 3.5, 4.5, 5.5,
6.5, 7.8 or 8.5 digit DVMs?

> With currently available reference chips, 3 1/2 digits is relatively
easy, 4 1/2 digits is much more difficult, and 5 1/2 digits is almost
completely out of reach.

So you obviously realize that the sort of instrument you restore would
dictate the stability of the reference you need.

> And as with many physical measurements, you may have access to an
external standard at a given point in time (for example with help from a
fellow volt-nuts list member), but after this initial calibration the
stability of your reference over time becomes important since you can't
send it out for recalibration every year.

I don't suppose it would cost too much in postage to do it, and I don't
suppose that you would find it too hard to find willing people.

>From what I have read, the
LM399 &  LTZ1000 are the best devices. So how about this for a suggestion:

1) Build one reference based on the LM399 in one box.

2) Build a second based on an  LTZ1000 based in a second box, using any
critical components, such as resistors from different manufactures.

3) Did *not* perform any sort of averaging in electronic hardware.

4) Wait a few months, to reduce, but not eliminate the "shock" to the
devices of being built.

5) Got the two units measured by volt-nuts.

At this point you should have two pretty darn good references,  with I
believe the LTZ1000 being the better of the two. So you are a man with two
clocks, but both your clocks are good, and you know one is better than the
other. You can also compare them, and would be aware of any significant
differences which indicate a problem with one of them.

At this point you can do any averaging, with greater weight given to the
source expected to be better? Use different calculators to do the averaging
just in case you find a new variation of the "Pentium bug"

The drift, temperature sensitively of each can calibrated. Any drift which
does occur is unlikely to be correlated as the two "house standards" are
built from different components.

Would the weighted average of these two house-standards be good enough to
calibrate the DVMs you want to restore? It really depends on what you want
to restore - 3.5, 4.5 ...8.5 digits.

The fact you have two devices, if the difference between them exceeds some
threshold, you must conclude one of them has drifted too much.

Personally I would have thought that it would offer something useful, but
obviously if you are going to be restoring 3458As, it is not going to be.

A sample of two is not going to be the ultimate in metrology, but it may be
good enough.

You could do the maths by looking at the prices of components, but I would
think that building just two independent references with decent components
is going to work out much cheaper than building a lot from poorer
components and averaging all those.

Someone else can probably think of something better, but it is all going to
come down to cost. But I doubt building a LM399 and  a LTZ1000 reference is
going to break the bank.

Dave