On 20/02/2017 18:46, Russ Ramirez wrote:

Hi Russ,

some voltmeters will supply extra digits in the output they provide, but whether they are of any use depends upon how
stable the reference is in that particular 6.5 digit dvm, unfortunately it needs a dvm with better capability to prove
if those digits are believable.

My question is partly based on a real situation where I have 3 DMMs of
decent quality Keithley 2000, 34970A (34401A), and 3457A that all agree
with a good quality voltage reference that is exceeding its ppm/degree C
spec as far as I am able to tell. I felt that this was telling me that I
would need a 3458A to know how much closer to 10V the reference was at this
point. A quick calculation of 20 bits of ADC for example told me that I was
looking at 9.5 uV of resolution, or 0.6 uV at 24 bits noise free, etc. so
it did not seem to be entirely unreasonable to see the LSD of a 6 1/2 digit
DMM be higher than 1 or 2 counts.

Russ

On Mon, Feb 20, 2017 at 12:57 PM, m k m1k3k1@hotmail.com wrote:

In message CAJgz9AYgQQjOmOK57yBiohF6Qh0DX6sEH-M_RxaBWQC54uDt7Q@mail.gmail.com, Russ Ramirez writes:

Somewhere in your instrument there is an A/D converter of some kind.

Even if it is perfect, you can never know if the last bit from the A/D
represents a voltage which is slightly higher or slightly lower than
the number it reads.

This is why specifications for digital instruments always includes "± 1 digit"

In theory, if you have truly perfect electronics, it could be "± 0.5 digit",
because in that case you might as well use the same ±1 as everybody else
and double the size of the range.

The A/D in a 6½ digit instrument by definition must have at least
22 bits (2^21 = 2097152 + sign bit) and if we scale things so the
least significant bit is 5 microvolts, we have a full scale voltage
of 2^21 * 5e-6V = 10.48576V.

QED:  Uncertainty on your 6½ digit meter on a 10V signal will always
be at least ± 5 microvolts.

If you do the same math for a true 8½ digit DVM, you need a 29 bit
A/D converter (2^28 = 268435456 + sign) which means you can use 40
nV as stepsize yielding 10.73V range, and thus in teory get ± 40nV
measurements.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

I believe you can get 7-1/2 digits out of the 3457A via the HPIB port.

Regards,
Tom

----- Original Message -----
From: "Russ Ramirez" russ.ramirez@gmail.com
To: "Discussion of precise voltage measurement" volt-nuts@febo.com
Sent: Monday, February 20, 2017 2:26 PM
Subject: Re: [volt-nuts] Basic question concerning voltage references

Right, and also the 34970 using BenchView, though I could not find where
they claimed that. In the 34970/401 case I was simply seeing a lot of
variation in the LSD, and immediately discounted it as useful in any way. 6
1/2 digits had a close correlation (0.893 I think the number was) to
temperature, but 7 1/2 was all over the place and a relatively low value
for sigma.

Now you know how I spent part of last week and this weekend!

Russ

On Mon, Feb 20, 2017 at 1:33 PM, Tom Miller tmiller11147@verizon.net
wrote:

You can get 7.5 digits using the HIRES register over GPIB or by using the stat function from the front panel.

Thanks Jerry, I'll give that a try later just to see what it does :-)

Russ

On Mon, Feb 20, 2017 at 2:53 PM, Jerry Hancock jerry@hanler.com wrote:

On Mon, 20 Feb 2017 19:29:55 +0000, you wrote:

I just have a minor quibble with the above which does not alter your
point.

There is no requirement for an ADC output to be a power of 2.  Slope
integration and voltage-to-frequency designs commonly use or used BCD
counters to produce a power of 10 output.

I remember when supressing negative zero was a new feature.