Stats - HP3456A

[Krytron]

Its been a long time since I have played with volt-nuts and DC calibration.  I have just recently moved and unpacking my electronics shop.  I have three of the HP3456A DVMs which can do stats with the measurements.  I have a fellow ham radio operator who brought over a $20 Chinese voltage reference from ebay thats based on the AD584L.  He was happy to just measure the output at 6 digits, but I believe we also need to know abouts its drift over time and temperature.

As a review, the 3456A can do stats, and then you can retrieve the mean and variance and number of measurements performed.  If i do a square root on the variance and divide the result by the mean, I should derive a result that I can convert to ppm.  The plan is to measure the reference at freezer temp, room temp, and then low oven temp to derive the temperature related specs.

Does the procedure look correct ?  All comments welcomed.  Thanks.

[MosherIV]

One of the problems with the AD584 (all variants) is that it is too noisey when measuring with a 6.5 digit meter.

The minium noise is 50uV, right in the last digit of 6.5 digit meters.

[Andreas]

Hello,

the AD584 also has a relative large hysteresis.
The hysteresis is dependant on (IMHO square of) the temperature excursion.

So it is wise to note the value at room temperature before and after the test.

Attached one measurement (temperature cycling) of a AD584JH. (metal can case)
There is obviously large ageing drift and hysteresis during the measurement.

with best regards

Andreas

[guenthert]

As a review, the 3456A can do stats, and then you can retrieve the mean and variance and number of measurements performed.  If i do a square root on the variance and divide the result by the mean, I should derive a result that I can convert to ppm. 

You can convert anything you like to ppm, but what are you trying to accomplish?  If you are trying to estimate the deviation of the voltage reference from a true value, you either need a second source with a known value and uncertainty or a calibrated meter.  Now while the 3456a are amazing instruments (I'm a proud owner myself), they are long time obsolete and I doubt there are many around which are calibrated.  Strictly speaking, having multiple uncalibrated meters doesn't get you any further in determining the true value of a source.

If you just want to determine the temperature coefficient of the reference then you don't need accurate measurements, just sufficiently precise ones.  The TC can be stated as ppm/K (typically some average or maximum around a given ambient temperature), no need to bother with the variance.  The variance is a indication of the quality of your measurement rather.  If the temperature is constant, then a high variance would be the result of noise (in the meter itself or from the source).

BTW: I got a cheap ad584jq module and put it into a small self-regulating oven using a TL431 and a NTC much like http://www.techlib.com/electronics/ovenckts.htm  -- the temperature remains at ~34degree C (a tad low perhaps, but so far has been sufficient) +/- <0.5K and the TC (to ambient) of the output voltage is about 2ppm/K.  This was just a quick hack, I'm not particularly good with the soldering iron and could have put more effort into thermal insulation, but I'm quite happy with the result already.  In fact, the TC (to ambient) above is the combination of the TC of the ovenized reference and of my 3456a (and the little peltier elements along the path between them).

[Galaxyrise]

I have been led astray just trying to take a few discrete measurements and draw conclusions from them.  It's much better to chart over time, like Andreas did.  Perhaps look in to acquiring a GPIB interface for your meters.  (I don't have any recommendations, I'm afraid.  My oldest stuff supports rs232.)