All DMM's that I know of (that use an integrating ADC) require that the ~7V reference be converted to +/-10V for use by the converter. This would require 2 op-amps and 2 resistor dividers, all of which will contribute to ADC drift-- possibly more than the reference itself. For example, in the Agilent/HP 3458A 8.5 digit DMM, they use a (drifty) dual LT1013 op-amp, a cheap (and drifty) resistor network to go from +7V to +10V, and a couple of cheap (and drifty) metal-film resistors to go from +10V to -10V. This causes horrible temp-co drift (below the 6th digit), which forces the user to auto-cal with a temperature change of more than 1-deg-C-- (which is *very* annoying if you don't have a temperature controlled lab)-- in fact, it's so bad that if you don't control the room temperature at 23C +/-0.1C, then the last 2 digits are almost useless (except perhaps for ratio measurements). By contrast, the Fluke 8805A 8.5 digit DMM uses chopper amps and ultra high stability metal-foil resistors for this function, and as a result there is no noticeable drift +/-5 degrees-C from Tcal (which is usually 23C). So, while the reference is important, other components and design trade-offs can and do affect that ultimate accuracy of any DMM.
With the exception of the Fluke 8805A (which is around US$12K), I have yet to see any manufacturer design a DMM for accuracy and stability as the #1 priority-- they all seem to be more focused on "whiz-bang" features and pretty user interfaces (which is nice), but I would love to see just one manufacturer produce a 6.5 digit DMM that had honest to goodness *real* accuracy to the last digit (+/- one count on the last digit) for basic DC, and with best-in-class accuracy on all of the other functions-- and also would maintain that level of accuracy for at least one year between calibrations (and at +/-5C)-- and would also be able to be calibrated with just +10V and 10K references (ie. "artifact calibration"). In other words, something in between the realm of 6.5 digit DMM's and the Agilent 3458A. Myself, I would be willing to pay US$2K to US$5K for that. You can get that today by purchasing an old HP 3458A, and then set the digits displayed to 6.5-- which will remain stable at this number of digits for a long time if you "auto-cal" once per day-- but the 3458A is a rather large and heavy beast, with a very noisy fan. I would love to have what I want without all of the "fluff", and without having to over-pay for a higher level DMM...
That said, at the US$1K price point, this 34461A is a great DMM-- and I would buy this over a less expensive Rigol 6.5 DMM just because it is designed and supported by Agilent.
Sorry D.M.,
what you assume about the 3458A is absolutely incorrect!
NIST, PTB and all other metrological institutions wouldn't use such a scrap instrument, like you describe the 3458A.. but you can see it in every other photograph around JJ standards there...
Simply study its specifications, i.e. 0.05 + 0.05 ppm transfer accuracy for 10min. and +/- 0.5°C, and you'll directly recognize, that your theoretical assumptions are false.
Obviously, you compare the 3458A with the Fluke 8508A, (a 8805A does not exist!), but latter stability specs are even slightly worse, (0.12 + 0.1 ppm for 20min. / +/- 1°C)
And also, its T.C. is 0.3ppm/K, compared to 0.15ppm/K for the 3458A.
From practise, i.e. from my own measurements, I can tell you, that the 3458A really is able to deliver 8 stable digits on a short time scale, and to < +/-0.1ppm over > 24h, even if the internal temperature changes for a few tenths of °C.
Reason for your wrong assumptions: You have lost the overview in the 3458A CLIP, very obviously:
The +/-10V reference you mentioned: U400 = LT1013 with RP400, R402, R403, is used in the OHM circuitry only, and its stability is not important for that function at all.
The A/D in the 3458A runs instead on +/-12V references, generated on the A/D board by U160, U165 = LT1001 and 4 resistors on the A/D hybrid.
For sure, those resistors are T.C. matched, and they are isothermal, so that their influence on stability is very,very low (as the spec proves right).
Those OpAmps contribute very little to T.C., but do not add noise in the A/D circuitry, as ChopAmps (as in the 8508A???) would do in that place.
I can absolutely recommend to you the 3458A, if you go for best DCV stability and linearity.. although there are some disadvantages in that old design.
One is the mistreated LTZ1000 reference, at 95°C , which nearly gives no better 1yr. stability than a LM399. (Can easily be pimped to much better stability).
And the 1000V range is not compensated for self heating effects, as is the 8508A.
The new 34461A is really nice, because it's a modern and cheaper version of the old 34401A. And the scrappy current ranges are improved a lot.
Once again, an improved 1yr. stability for the 34461A would require a well designed LTZ1000 only ( agilent would suck several 100 bucks more for that).
You can age the LM399 as long as you like, but you will NEVER get a stable reference out of it..
As there exists no spec and no method for reliable
ageing for a smaller 1yr. stability, it will drift forever, and everything else is simply wishful and very theoretical thinking.
You only have the chance to
monitor running references and to sort out the most stable - and that's exactly, what they all (Fluke, agilent,..) are doing.
Not to forget: If you need for some reasons 6,5 digits of stability and uncertainty in your
measurement, you'll have to use a 7,5 digit
instrument, due to the digital, i.e. 10:1 ranging.
A 6,5 digit instrument 'precise' to the last digit won't make sense, therefore.
Therefore, a 3458A would be the right choice for that requirement.
Frank
Home
Help
Search
Login
Register
