6.5 Digital Multimeter Temperature Coefficients
A tech note in progress ... revised 8/18/2013, use refresh on your browser to be sure to see the latest changes
A common question related to digital multimeter (DMM) calibration is, "How accurate" does a voltage reference need to be to properly test or calibrate a 6 1/2 digit or even higher resolution multimeter. "Accurate" in this case often means how much does the output voltage of the voltage reference change with ambient temperature? with time? with power supply voltage? Is it sensitive to humidity?
However, many users of high precision DMMs do not consider (or do not think) that their DMM calibration moves with temperature, time, or other factors.
Recently, with improved automatic test routines that we developed for use with out micro-environmental chamber, it occurred to me that similar routines could be used to make a first pass measurement of the temperature coeffiicent of some of our DMMs. While some DMMs, such as the new 34461A have low resolution (coarse valued) internal temperature sensors and digitization, most others do not. However, many high quality precision DMMs, such as the Agilent 34401A, 34410A, and 34461A use relatively thick copper terminals which are soldered to the main circuit board. Also, we use 100 ohm aluminum cylinder RTDs for some our measurements, which by chance, fit perfectly into the rear panel banana jacks.
Especially during the nice weather of late summer and fall, we often are able to leave the windows open and allow the lab temperature to vary. Usually with doors shut overnight, the temperature rises from running equipment dumping heat into the room. By recording the voltage from our Fluke 732B (tempco of < 0.04 ppm / c) and the RTD temperature from an unused rear panel heavy copper terminal, we have been able to start to make first pass estimates of the temperature coefficients of our DMMs. Pomona low thermal EMF leads were used for this testing.
It is very important to emphasize that measurements of one or two units of any particular type say very little about typical numbers for any brand or model number. However, we hope these measurements will give some pause to those believe that their 6.5 digit+ meters have a nearly zero temperature coefficient.
Why does it matter? Take one of our original 10 V SVR boards rated at 5 ppm / c or better. Suppose the board was calibrated to 10.000 00 V at 24 c (for the moment we do not consider our absolute accuracy specification of +/- 5 ppm, and assume that the value is indeed 10.000 00 V exactly at 24c). Also, lets give the SVR board a hypothetical temperature coefficient of 2.0 ppm / c (for a 10 V reference, 1 ppm = 10 uV), so that is 20 uV / c. A user views the SVR board with a perfectly calibrated DMM at 24c and reads 10.000 00V! Now, the user says, let's measure the temperature coefficent of the SVR board over some hours. With a zero tempco DMM, at 26c, the user would record (for a +2c change) 10.000 04 V (a change of 2 x 20 uV = +40 uV). Now, if the tempco of the DMM is actually 1 ppm / c itself, that same user will see +40 uV from the change in the voltage reference, plus an additional 20 uV from DMM drift, or 10.000 06 V. Two counts more error at 6 digits, maybe not so bad?
However, now take our SVR-T board with a tempco of <1 ppm / c. As above, the SVR-T board has a output value of 10.000 00 V at 24c. It has a temperature coefficient of +0.2 ppm C. For a +2 c change in temperature, the output voltage rises by +4 uV for a perfect (zero tempco) DMM reading of 10.000 00 V. However, if the DMM has a 1 ppm / c tempco itself, now the DMM displays the new reference voltage 10.000 004 plus another 20 uV caused by the DMM itself. Now, the DMM "sees" 10.000 024 V and if it is a 6.5 digit unit, the display reads 10.000 02 V for a 2 c change. Many high precision DMM owners would then conclude that the SVR-T board has a temperature coefficient of about 1 ppm / c and in their view might be close to or out of specification.
The phenomena is not just a matter of whether the "measured" temperature coefficient of SVR-T board is accurate. Many such experimenter are amateur scientists or amateur metrologists developing their own high precision references or standards. DMM users need to be aware that even a high cost, high quality, high resolution DMM might have a temperature coefficient of at least up to about 1 ppm / c.
As this note progresses, we will publish the measured data for our DMMs. These single or two unit measurements are not meant in any way to categorize the tempco of any given model or brand, but, simply make experimenters aware of DMM temperature coefficients.
Note that the specs for the 3458A DMM temperature coefficient without use of ACAL for every 1c change, are pretty comparable to modern high quality 6.5 digit DMMs. The 3458A page has a link to the data sheet PDF. For example, on page 10 of the data sheet, the temperature coefficient for the 10 V scale without use of ACAL for a >1 c change is (0.5 ppm of the reading plus 0.01 ppm of the range) per degree c. With ACAL, the temperature coefficient for the 10 V scale is (0.15 ppm of the reading plus 0.01 ppm of the range) per degree c. Some 3458A might perform much better than spec.
While we do not have raw day yet suitable for publication, here are some estimates our own DMM tempcos:
Agilent 34401A - TBD < 1 ppm / c for an intermal temperature of between about 34c and 35c. Example of one tempco test in progress (10V scale, 6 digit "slow"): LV front panel PDF, Raw Data TXT, Excel graph PDF (This initial data might have included some warm up time.) Later data suggests a somewhat lower tempco. This is not a trivial measurement: LV front panel PDF, Raw Data TXT, Excel graph PDF (the linear fit is no longer reliable, the slope is higher than the average). Some later testing has indicated closer to 0.5 ppm / c.
Agilent 34410A (from 2006) - was very near to zero ppm / c. Ironically during this testing, it became clear that the internal LM399 was from time to time causing a 1 ppm pk to pk "hash". The failing LM399 PDF (or simply intermittently noisy part) was replaced with a LM299AH-20 part. The "hash" noise is gone, however now the DMM has a tempco on the order of 0.25 ppm / c.
Agilent 34410A (from 2009) - not enough data to report yet (TBD).
Agilent 34461A (from 2013) - needs more study, however appears be about 1 ppm / c. We hope to also be able to characterize LM399s in our micro-environmental chamber. A quick first test indicated that the 1 ppm / c DMM tempco was not simply a matter of LM399 tempco, this needs more study.
These values may or may not be linear over wide temperature swings of more than about +/- 2c or 3c.
The bottom line is: 6.5 DMMs may have a temperature coefficient that is not zero. A few experimenters are studying many references and are able to periodically check one or more references with a calibration laboratory. For others, short of a study at a calibration laboratory or by use of a Fluke 732B type reference or equivalent known to be in good working order, most DMM users may have difficulty determining (or be unaware of) the tempco of their DMM.
Also, many types of references and/or reference parts (or even DMM products for that matter) might statistically be more likely to trend one way or another. For example if a population of reference parts trends towards a statistical peak at +0.5 ppm / c, then many of the parts more likely than not might look in agreement and therefore "stable" even though the mean of the group is moving at +0.5 ppm. Many experimenters can detect and correct for such trends. To others, a few prototype references, all moving at -0.7 ppm / c (hypothetical case) might all look the "same" at several temperatures, and therefore be deemed to be "ultra stable".
Unfortunately, the same principle applies to DMMs. Say two DMMs are moving with similar tempcos, one at +0.45 c / ppm, and a second at +0.57 ppm / c. To an unaware observer of a voltage reference with a 0 ppm / c tempco, the reference looks to have a measurable temperature coefficient. Why? Because both high precision DMMs say so!
(Also, as an aside, while many lead connections are less of a problem at 1 to 10 ppm, thermocouple junctions caused by various test connections can be another contributing error in measuring tempcos relying on data measured to better than 1 ppm.)
In a related matter, there are starting to be a large number of demonstrations and blog posts on the web showing very high resolution (e.g. 10 to 50 uV) 34461A histograms of various voltage reference chips under test. Many (not all) writers conclude that the shape of the histogram is absolutely determinative of the performance of the voltage reference chip under test. Some have commented on odd twin peaks. It could be the reference moving between two voltages as over small changes in room temperature, however at the ppm level more likely, the changes with temperature of both the 34461A and the reference under test for room temperature swings of a couple to few degress F (typical of some HVAC system performance as measured at the bench, sometimes worse for hot air systems where hot air might be blowing in the immediate vicinity of the test bench). This is not to criticize the 34461A, it is a remarkable instrument. However, all errors, including DMM errors must be considered for a data set to be most meaningful.
COPYRIGHT © 2013 JOSEPH M. GELLER