Okay, so I finally burned an entire day recalibrating my 8506A.
The output of the Active Filter was about 10uV higher than the output of the DC Signal Conditioner with the meter set to 10V range and input shorted with some copper wire. Ten microvolts is equivalent to 1 count at the ADC and is well within spec (i.e., ±2 counts or ±20uV). But being a perfectionist, I adjusted its output to match the DC Signal Conditioner's output (after hardware zeroing the 100mV range) to within 0.5uV (best I could do with the 1-turn trimmer and checking that the DMM4050 was zero corrected). Step 3) in the troubleshooting section for adjusting the Active Filter zero is pointless, as it does not account for offsets introduced by the ADC, which should be adjusted by R8 instead. So it's better to match the DC Signal Conditioner than zero the display. In this way, offsets introduced in the path from the DC Signal Conditioner to the output of the Active Filter are effectively eliminated. As a result, the ADC or Active Filter are not being used to remove offsets from other modules.
To speed up nulling Ib, I connected a Keithley 617 electrometer to the input and adjusted the bias pot for a reading of <1pA, shooting for ±0.5pA. I then connected the specified 1MΩ//0.22uF (with leads soldered into some banana plugs) to the input and placed a small concave metal sheet connected to chassis ground around the input, set sampling to 10 and checked for a reading of less than 10 counts on the display, which it was. Pleased with the result, I called it good. The manual asks for <30 counts (3pA), so I was well within adjustment.
Next I zeroed the ADC and performed the optional 10V ladder adjustment and ran through the normal 7-step ADC ladder adjustment three times before I was able to match my recently calibrated DMM4050 to within 2ppm (with one 4ppm outlier) on the table 4-5 linearity check. This is where much of my time was spent. A bit like the blind leading the blind, but I would rather the meters match measurements, even if their absolute values might not be correct.
Then it was back to the DC Signal conditioner for setting the 100mV, 1V and 100V full scale hardware adjustments (clearing the software gain corrections first, of course) with a round of software corrections for zero offset and gain for +/- values on each range. The separate software corrections for each polarity are very nice and help dial in the 1000V range since there is no hardware adjustment for it (though it barely needed it).
All this took about 5 hours with a separate 4 hour warm up! I would never tell anyone to perform a full hardware cal on one of these unless their linearity is out of spec. Hardware zero and software cal is how these should be calibrated as to do anything else is an entire day spent laboring over ppms while battling the effects of temperature changes. Thankfully I had a small heater I could use to keep the room within <1C of 23C.
I should point out that the manual makes the full DC / ADC calibration routine seem more complicated than it actually is and could be streamlined to:
0) Warm up: Power on UUT and wait a minimum of 4 hours for unit to fully stabilize. Keep room within 1C of 23C at all times.
1) Power supply adjustment: Remove top cover and set cal switch to on. Configure filter and sampling for appropriate values, then perform standard power supply adjustments and optional non-recurring PSU adjustment.
2) Hardware zero corrections: Short input and set range to 100mV. Disable software zero. Adjust zero offset and bias current on DC Signal Conditioner for 100mV range. Set range to 10V. Optionally check the Active Filter and adjust R14 so its output matches the DC Signal Conditioner’s output using a DMM with 0.1uV resolution. Adjust zero pot on the R2 ADC for a display of 0.0000000. Set new software zero for each range using the [Zero Vdc/Ω] key.
3) 10V hardware calibration / R2 ADC calibration: Set range to 10V and clear software gain corrections for 10V range. Perform optional 10V ladder adjustment. Perform a two-part iterative cycle involving a 3-step ladder adj. (match ±10.10000V and remainder) and a 4-step ladder adj. (5V, 2.5V, 1.25V and 0.625V ladder) until linearity verification passes limits in table 4-5.
4) 100mV, 1V and 100V hardware calibration: Clear software gain corrections for 100mV range and store new software zero (if needed) then perfrom FS hardware gain adjustment of 100mV range. Repeat for 1V and 100V ranges.
5) Software calibration: Put top cover back on UUT. Apply software zero (if needed) and gain correction for +/- polarity of each range (10V, 100mV, 1V, 100V, 1kV, in this order for each polarity) at a value between 60% and 190% for 10V ranges and below or a maximum of 128V and 1200V, for the 100V and 1kV ranges. Calibration is now complete. Set cal switch to off.
It's a lot of work, but isn't as daunting as some people have made it out to be. And again, not needed unless linearity is out of tolerance or you have the time, equipment and patience to fiddle with chasing ppms. In general, steps 0-2 and 5 are all that's needed.
NOTES:
Since I often use the average mode during normal use, I performed the entire calibration with filter F enabled to account for any offsets it might add. Sampling was done between S9 and S11, depending on the step for dialing in sub ppm on the 10V range. The added delays between readings also helps to let the meter settle, which is needed for some of the ladder adjustments and obtaining a reliable software zero on the 100mV range.
With the electrometer already setup, I decided to check my other DMMs for comparison. The slightly outdated Tektronix DMM4050 measured about 16pA and the HP 3456A was 4pA with AZ on and about 1.6 with it off, though its value is affected by the meter's offset voltage and its reading quite jumpy when AZ was turned on.
The 8506A does not have an auto zero circuit like the HP 3456A, and is thus sensitive to temperature variations. Without auto zero on, the 3456A appears to perform worse than the 8506A during my fiddling around. So Fluke did make a valiant effort with an outdated approach to DMM design. It also has lower bias current and while there is some noise to it, it pales in comparison to the noise generated by the 3456A's AZ circuit.