10:1 Reference Divider a.k.a. "752A Junior"
(work in progress, be sure to hit refresh to pick up our latest changes and entries)
The well known Fluke 752A 10:1 Voltage reference divider2 allows for the transfer of the absolute accuracy (within some tolerance) of a 10 V transfer standard (today, typically a 732B 10 V transfer standard) to 1 V and 0.1 V outputs.
With the advent of ultra high accuracy digital multimeters, such as the Agilent 3458A, such devices have become far less important. However, with an older, typically used 3458A or a somewhat less accurate hp 3456A, there is still a need to derive a 1 V reference point from a known 10 V reference value.
The Fluke 752A is "self-calibrating" using two basic methods. First, it has an internal divide by two with a trim to cause both Rs to be identical. Second, it uses a technique of three matched resistors3 which can be switch configured in series or parallel. The three resistor "module" is placed in series with a lower 10:1 divider resistor in the traditional way. The idea of the three resistor module is that when the three Rs are parallel they can be set equal in value to the lower R of the 10:1 divider, yielding a 2:1 divider for calibration purposes. The 2:1 configuration allows for the simple Wheatstone bridge comparison to the internal divide by two. But, how do you know how good the divide by two is? Using a 752A polarity switch, Fluke has you rotate the divide by two (top and bottom resistors in the divide by two string). When the two values match, the divide by two is exactly balanced! Then a center null meter allows the module of three Rs to be exactly matched to the lower R of the 10:1 divider. Whala! A perfect 10:1 divider (assuming a HiZ meter input, e.g. 1 Gig, such as a 3456A, or Agilent 34410A in the "HiZ" mode).
I was inspired1 to try to assemble a "752A Junior" JPG from "precision" wire wound Rs JPG to make a spot check of my present 1V scale calibration of our hp 3456A as well as our Agilent 34410A DMM. Using the non-buffered 1.018 V output of our Fluke 732B, we had already calibrated our hp 3456A 1 V scale to match the relatively new 34410A reading. However, our $600+ annual calibration at Fluke is only for the 10 V output. So, while we know the 3456A matches the 34410A reading, we lacked a way to verify the 1 V scale absolute readings.
From a large box of relatively ancient wirewound resistors (an eBay buy from years past), I recovered a number of ~27k and 3k wirewound Rs. The manufacturer "DAVEN" took the trouble to give each one a serial number, so at least there was a superficial chance they are low temperature coefficient Rs. We mounted them on a fiber board with HH Smith (Abbatron) banana jacks JPG, not quite Pomona 3770 series low thermal EMF jacks, but good enough for this exercise.
During initial testing, two things become apparent. First, our divide by two was insufficient to accurately trim the 3R module, and second, our trim circuit was deteriorating our otherwise short term stable 10:1 divider. Here is how we ended up using the "752A junior" without the divide by two divider:
1) configure the 3R module to all parallel (using two Pomona 2948-24 gold banana leads) by "shorting" each of two of the yellow banana jacks in the vertical direction. Configure the Agilent 34410A to four wire ohms mode, and measure the 3R parallel combination to 6 digits (100 plc integration, sense leads closest to the jacks). Then move the 34410A four wires over to the lower R and adjust a Shallcross 827 decade box across one of three series 3 kohm DAVEN Rs (the lower leg of the 10:1 divider ~9k R) to give exactly the same R, again to 6 digits. Repeat the process until the numbers repeat. This setting is only good for the present room temperature and short term.
2) remove the two jumpers that made the 3R module into three parallel Rs, thus returning it to three series Rs. The newly series configured 3R module in series with the trimmed lower R, now yields a nearly perfect (6 digits) 10:1 divider!
Apply the Fluke 10 V output to the 752A Junior 10:1 input. Connect the 752A Junior 10:1 output to one or more HiZ (e.g. 1Gig!) DMM inputs. Measure 1 V! This is a "rough" instrument at best, however it is always good to have two or more ways to check a calibration point. JPG
okay, what were the odds of both meters reading exactly 1.000 000 V? First, I just adjusted the 1V trimmer on the 3456A to match the 34410A reading of our Fluke 732B 1.018 V output a few days ago (the 3456A having a very HiZ input and the 34410A set to its "HiZ mode", since the 732B 1V output is a simple internal (non-buffered divider). Just after trimming the lower R with the Shallcross box, if anything, the 34410A occasionally was flickering to 1.000 001 V and the 3456A occasionally to 0.999 999V.
okay, but how close could you have possibly got the three Rs of the 3R module (the "top" R of the divider) so that the 3R parallel series "trick" works to six digits!? Well, it turns out those serialized DAVEN Rs are pretty darn close:
26,995.44, 26,993.19, 26,991.09
The average of the three is 29,993.24, which means the net deviation from average is as follows:
+0.0082%, -0.0002%, -0.008%
So, we can build a spread sheet to see the best that can be done in terms of the final 10:1 ratio caused by differences in the individual Rs of the 3R module. Assuming a perfect match to the parallel 3R, for the measured variation in our 3R module, there is no perceptable error in the sixth digit of the ratio! If there were no other errors, the actual ratio would be:
0.0999999996 that's right, the error is in the -Tenth!- digit of the ratio. XLSX, PDF Now obviously there are other errors, such as including the resolution of the Shallcross trim on the one R of the lower series combinations of three DAVEN 3k Rs, any contact R with the Pomona gold jumper leads, any thermal EMFs anywhere in the circuit, actual lead Rs, etc! However, it was also not out of the question to hit 1.000 000 V!
So, how far off (how much variation) can there be in each R of the 3R network before we see a change in the sixth or lower digit of the ratio? The second set of numbers on the same spread sheet XLSX, PDF show hypothetical values of each R of the three R module. One is about +.19% from the average of the three. Another is about -.17%. The ratio is now in error by 2 parts in the sixth digit. All other errors insignificant at the sixth digit (a premise, not a statement), a 10.000 00 V input would yield an output of 0.999 998 V, or 2 uV low.
We had the good fortune to have almost perfectly matched Rs by manufacture. In the 752A (the real one!), Fluke matches longer strings to very high precision. In fact, I should think each R in a 752A costs more than our surplus contraption!
Someday we hope to buy another new Agilent 3458A (or, its replacement) and/or a FLUKE 752A. However, as we note here, a lot of low cost "metrology" can be done with high quality surplus parts at low cost.
1 Private communications with Lars Walenius, many thanks for our continuing discussion of voltage references and related metrology issues.
2 Fluke 752A product page with manual http://us.flukecal.com/products/electrical-calibration/electrical-standards/752a-reference-divider?geoip=1 ; 752A schematic diagrams at the ko4bb manual site. See also: Ratio calibration—what’s the big deal? By Chuck Newcombe, February 2012.
3 An Easy to Build 0.1X and 0.01X Resistive Divider, B.V. Hamon's Clever Trick , see also: Journal of Scientific Instruments Volume 31 Number 12 Create an alert RSS this journal, B V Hamon 1954 J. Sci. Instrum. 31 450 doi:10.1088/0950-7671/31/12/307, A 1-100 Ω build-up resistor for the calibration of standard resistors
Comments! Typos! and/or Errors? Please send an email to me at joegeller -at- gellerlabs.com
COPYRIGHT © 2012 JOSEPH M. GELLER