FieldFox Drift Compensation

[jjoonathan]

I recently switched from an 8510C to a FieldFox as my main VNA. As you can imagine, size and mobility requirements were the driving factor, but it did come at a cost to stability. A small cost in absolute terms compared to my needs, but large enough in relative terms that I can't help but want to chase it:

https://jjoonathan.github.io/plot_drift.html

the time scale pretty strongly suggests that the effect is thermal in nature, and the piping hot ports on the FieldFox sort of point in that direction as well. There is a temperature reading in System > Service Diagnostics > Internal Temperatures, but I can't figure out for the life of me how to get it out of SCPI. It looks like Keysight's bigger VNAs have :SYSTem:TEMPerature? but that doesn't seem to be present on the FieldFox. So that's my first question, if anyone around here knows how to get at FieldFox service diagnostics through SCPI, or another way of getting at the temperatures. I might like to do some kind of automatic compensation, so taping a thermocouple to it isn't really what I'm going for, but I might get there eventually.

My second question is more fundamental and has to do with the interpretation of the time domain results. Here's the setup:

• Let the FieldFox warm up for several hours and obtain a calibration using 85056D slotless standards applied directly to ports.
• Let the FieldFox sit and cool down for a day.
• Cold start FieldFox, attach 85056D short to port 1 (see picture), apply previously collected calibration.
• Start a script that grabs a trace every minute.
• Let it run for 16 hours.
• Time domain transform all traces using scikit-rf net.impulse_response()
• Plot ir-ir_final (the impulse_response function already stripped phase information)

This results in the "TDR" plot in the link above. At first I was surprised to see the bump at 50ps (7.5mm) in front of the reference plane. Was I looking at a temperature sensitive standard? The more I thought about it the less likely that seemed. It's a short, so it should be all metal and I'd think that most of the expansion/contraction should show up near the reference plane, not 7.5mm into the thing. My current hypothesis is that the first order culprit for the drift is something that applies to the whole trace, like relative receiver sensitivity, and the fact that this "shows up" at 7mm is just because it changes the reflection and that's where the reflection is (see pictured time domain trace of standard).

I don't really _need_ a good answer here, my primary purpose was just to become familiar with the first-order drift behavior of my new VNA so I can recognize it easily and not spend time chasing "DUT behavior" that is actually instrument behavior superimposed on the DUT. A little bit of tail-chasing up front saves a lot of it down the road, or that's the theory anyway.  Still, I'd like to bounce the "receiver temperature sensitivity" hypothesis off the crowd. I was stuck on "but why is the short standard temperature sensitive?" for longer than I should have been and I'd like to be a little more sure that I've really got the correct mechanism in mind this time. Even just writing it down helps collect my thoughts, but I'd certainly love to hear from anyone who has spent time chasing the drift behavior of the FieldFox or any other kind of VNA.

[KE5FX]

Interesting set of observations, and the time slider is very cool. 

I imagine the directional coupler (which I'd assume is a resistive bridge in the FieldFox) is what's doing the drifting.  It is making the short standard's return loss appear to vary as the unit warms up.  You could verify that easily enough by hitting it with some R134a and/or a heat gun while you measure it on either VNA after warmup.

Something to remember about the 85xx VNAs is that they are based on technologies and components that predate digital error correction.  They had to be ridiculously stable by modern standards, so it's no surprise that the FieldFox falls short in a head-to-head comparison. 

Also, calibrate them both now, then come back 6 months later and see how the calibrations are holding up.  I'd expect the 8510C to win that contest for sure, for the same reason, but there's always room for surprise in this business.

[jjoonathan]

Hmm, yes, answering this experimentally is probably quite practical. A short cable for thermal isolation would let me play with the standard and see if it generates similar results, and to test the source and receivers I could bring the FieldFox outside (it's summer and hot) or even put it in the refrigerator. Ha, that might have been tough with the 8510C!

[Joel_Dunsmore]

The fact that the bump in time domain appears pretty suddenly seems and occurs at a slightly positive time, and the phase of the S11 moves in a positive direction, implies the short circuit distance is getting shorter (if you used a calkit short, it is actually an offset short). If you looked at the raw response you might see a big discontinuity there, and it is being corrected for; but with time and temperature, something shifted (10 degrees phase at 50 GHz implies a physical distance in reflection of about 88 microns).  So a slight shift in a discontinuity going through the test port interface to the port can account for the change in TDR response.  You can think of it as subtracting a big error and the position of the big error shifts slightly. 

In Fieldfox, the bridge is on the PCB which is connected to the front panel through a small cable and connector.  In the 8510 coupler, the test port connector is integrated into the bridge and so much less susceptible to any thermal effects.  And as mentioned earlier, the raw performance is much, much better.  So when the math is subtracting the raw performance to make it appear perfect, it is subtracting a much smaller number, so a change in that number has a much smaller effect.

And, for reference, an HP 8510 system was about $200k in 1985; that's $600K in todays dollars.  Fieldfox is a bargain compared to that.

[jjoonathan]

Yep, it's an offset short at 6.75mm. I'd expect thermal expansion in the metal on the order of 10ppm/K * 30K * 10mm, but that's just 3 micron. If the reflectometers are on the PCB (similar to Shahriar's FieldFox video), perhaps that's where most of the expansion is coming from. The intuition that polymers have a CTE around 100ppm/K with a few cm of distance along the PCB would place us right at 90 micron of expansion, which agrees quite nicely with the observation. This is good news, it means that letting the FieldFox warm up really is a pretty good answer here, at least for my purposes. The measurements already showed this, but it's nice to have a mental model with numbers that check out.

It's amazing how long the HP 8510 has held on. I am very glad to live in an era when instrumentation of this quality can drift into my budget range. 40 years and people are still doing real work with them. I had no trouble at all selling it on once I confirmed that the FieldFox was workable. I saw the raw performance -- that's actually why I went to such pains with the confirmation -- but it pulled through. Besides, I need to leave myself some motivation to turn the pile of 1.85mm microwave bits I have accumulated over the years into an actual VNA. Most of the "marine snow" falling off the microwave industry whales seems to be Anritsu kit these days, but there is an HP doubler and some Centellax amplifiers in the pile.

As you might have seen in the notebook I linked, I am working on getting the 2.4mm and 1.85mm HPAK calibration kits into scikit-rf (still working on banded cal). At one point I started writing a parser for the files, but then Keysight pulled the cal kit download page and pointed at a PDF, so now I'm thinking I should just commit the latest version of the hardcoded values and leave it at that. I suppose they probably aren't updated very often, but I don't have good visibility into how not-often that is. If the cal kit constants change less often than the file formats, that's probably the right approach  but in any case is there a mailing list for "ping me if the constants change"?