Please do not beat me too hard for my first topic.
I have a few 6.5 digit meters (3456A and Keithley 196) and have been long pondering a long/medium-term monitoring set up and finally managed to get it done. I would like to share the results and ask for feedback. My main goal was to check the stability of a Keithley 196, which looked bad to me.
I am measuring the output of a Vref board donated by another (dead) 3456A. In the original schematics it uses a (buffer?) circuit with a transistor and a resistor divider (U500) to produce -12V. I only added a 10k:20k divider made from two 0.01% resistors, which I had at hand, without anything else. The output voltage is now about 10.5V so the LM399 zener current is lower than in the 3456A, about 1mA now. I also have a 10k:14k divider, which would give me 12V, but that causes a 3456A to go one range up with 10x less resolution. I don't care much about absolute values and more about stability.
I measure 10.5V with one 3456A and one K196, both uncalibrated. The meters are controlled by an ATMega328 board over GPIB and the readings are saved by another ATMega to an SD card (Open Logger). I used a battery to power both of them so the control and logging setup is really small and low power. The biggest downside of this, I think, is lack of precise (or any, so to say) timestamps on readings. The Open Logger board has an oscillator so that more-or-less stable time can be derived from it, but I did not feel like hacking its firmware this time. Both meters are triggered alternatively by firmware in one ATMega so samples always go in pairs at the rate of about 2.15 pairs/s and the total duration of this experiment was about 5 days (terminated due to battery loss). I did not notice that, but the polarity was flipped on the 3456A so the numbers are of opposite signs. I did not monitor the temperature. I attach the logs and a few plots.
There are a number of sharp excursions on the graphs, which I attribute to EMI. Most of them are only one sample wide, but some are longer. Sometimes they occurs in both meters in the same sample, but they are much more often in 3456a. This could be partially because its integrating time is longer and the meters sample alternatively. However, the Keithley seems to average the measurements over a window of about 10 points, so it might just filter short ones out. I included the biggest “glitch” in one of the graphs and the averaging is evident at its edges. I have the “FILTER” setting off, by the way.
My biggest concern though is that the Keithley is indeed quite unstable. It wanders around by about 10 ppm, in an uncorrelated way from the other meter, which excludes the reference as a culprit. It also swings at startup quite a lot, taking several hours to stabilize. Is it likely due to a fault in the meter or it is just too sensitive to some environmental factors that I cannot account for?
I guess I need to think about reducing the EMI exposure. I could shield the reference board and use shielded test leads, but it would be more difficult to stop people from using the hairdryer
I will appreciate any comments especially re. K196 stability and about my monitoring setup overall.