[volt-nuts] HP 3456A Setup

[WarrenS]

Part of the Fun and challenge for some volt nuts is learning new ways to get 
the most from their old equipment.

An accuracy test I did on my HP3456A, 6 digit DMM, was to connect it to my 
Fluke 720A type 7 digit Kelvin Varley voltage divider
and by using the 3456A's null function to correct for zero offset, and 
adjusting the value of the 10 Volts going to
the resistive divider to compensate for any Full Scale gain errors on each 
of the DMM ranges,
All those "glowing red digits" on the HP3456A read the same, with-in 1 
count, as the knob numbers on the divider for every position of the dials.
The results where the same on the 100 mv, 1V and 10 V ranges, in both 
polarities doing this short term test.

True this is Not absolute accuracy, for that one would need a voltage 
standard known to better than 1 PPM,
but it is relative accuracy, aka linearity, and shows what kind of 
resolution, linearity and accuracy is possible short term.
Easy enough to make repeatable, meaningful, measurements that are within one 
count on the HP3456A .
So the stated 56 count error can be lowered to be one count max with a 
little knowledge and effort.
And the HP3456A is certainly capable of measuring voltage down to 0.1 uV 
accuracy, if one has the skill and knowledge to use it (and is not in a 
hurry).

OR better!
Another test I did was to measure where the HP3456A's LS digit readout would 
switch between any two adjacent numbers, from say a solid 1004 to a solid 
1005.
The measured width of the toggle noise at the lower end of the 10 volt range 
was under 0.1 count.
In other words the 6 digit DMM could be used to get 7 digit resolution by 
adjusting the input voltage to toggle between values.

All test where done with AZ on, Filter off, 100 cyc int selected, and 6 
digits selected.
This gives about one new reading every 3 seconds and averages for 100 line 
cycles.

If you really want to go ALL out nuts, you can average any number of 
readings using the built in "mean" average of n samples.
But because the noise is so low, you still only end up with 6 digits of 
resolution unless you add greater than 1 count of non noise dither.
I'll have to try that some day and see how good of a 7 digit meter it makes.

Got to love those HP volt meters.

Having lots of nut fun
ws

*******************
>
> Charles P. Steinmetz charles_steinmetz at lavabit.com wrote:
>
>>As far as zeroing the meter with its leads shorted, using "Math
>>Null" (aka Tare) is the only way to zero the 3456A's small offset
>>that is left over after "auto zero".
>
>>When using the Null command with the 3456A's  slow update rate, high
>>resolution, and filter off, you can measure voltages down to 0.1 uv
>>if you're real careful.
>
> My point was that doing this most likely does nothing to reduce
> systematic errors -- rather, it just adjusts readings by the
> particular random noise sample it takes when you activate the Null 
> function.
>
> Certainly, you can READ a 3456A to 0.1 uV on its lowest range, but so
> what?  There is a big difference between resolution and
> accuracy.  You have to be very careful that those glowing red digits
> don't lull you into thinking you have measured something with an
> accuracy of 0.1 uV just because the last digit reads tenths of a
> uV.  In fact, the ACCURACY (uncertainty) of the reading is much, much
> poorer than 0.1 uV -- in this case, well over +/- 2 uV.
>
> If you look at the 3456A specs, you will see that the best specified
> accuracy on the lowest range is +/- 22 ppm +/- 24 counts -- and
> that's only for 24 hours.  (At 90 days it is +/- 34 ppm +/- 24
> counts.)  On the lowest range, 24 counts is 2.4 uV.  So even without
> any scale errors, you can only expect useful accuracy of +/- 2.4
> uV.  The scale error can add another 2 or 3 uV to this, depending on
> the voltage being measured.  So, you can read a 3456A down to 0.1 uV,
> but you can't say that you MEASURED a voltage down to 0.1 uV with it.
>
> We need to be very careful not to get sucked into the resolution
> trap.  It is accuracy that volt nuts are interested in.
>
> Best regards,
>
> Charles
>