Mixed Coax/Microstrip calibration on VNAs

[ConKbot]

I'm trying to measure the insertion loss of some custom flex cables for USB3, using a VNA and have been running into problems trying to simplify my calibration solution to calibrate out the adapter fixture (SMA to the FFC connector).  I'm using an E5071C that's fairly well optioned (Enhanced TDR and Ecal module), and ideally I can characterize the fixture, plug the S4P file into the TDR setup, and use a 4-port E-cal to be ready to go.  But the game of getting reference planes where I need them to be has been giving me hell. 

For the adapter board, all 3 pairs are matched length and identical for the transition from 2x 50 ohm microstrips from the SMA down to a 90 ohm differential pair. I also have a calibration board with a matching par, but with SMAs on the far end a much shorter transition, and TRL/TRM calibration standards.

So far I can:
-TRL/TRM calibrate 2 ports, and get reasonable numbers for loss on the Line standards.
-Ecal the two other ports and get reasonable numbers for connector to connector on the Line standard
-Port-extend the TRL calibrated ports a couple mm to move the reference plane for those ports to the very end of the 90 ohm pair.

This gives me reference planes at the SMA and at the end of the 90 ohm pair, but of course there is no calibration between ports 1/2 and 3/4 so this isnt any good. I checked just to be sure, and measuring a Line standard, its giving 0.2 db for using ports 3/4 (the TRL calibrated ports)  0.8 db for ports 1/2, (the Ecal ports)  both reasonable. But 3.x dB for ports 3/4, (TRL calibrated port to Ecal port) So that is definitely no good. 

If I TRL calibrate all 4 ports, (port extending two to get to the end of the 90 ohm pair, and leaving the other two untouched) I can measure from the reference plane, though the long 50-90 ohm transition, and the 90 ohm pair, and get a S4P of that, but it still leaves the SMA-microstrip transition unaccounted for.   The TDR setup will also take an S2P file for each transition, so if I can do that, I can just calibrate with the ecal and everything will be happy. I can also do a 4-port TRL cal every time, as the reference plane will then be on the board where it needs to be, but with two line standards and a match standard, that makes for 28-step process, which I dont mind for a one-time measurement, I dont want to have to repeat every time I'm setting up to measure USB flex cables if possible

Searching online I found this, which on page 27/28 suggests an unknown-thru could be used to connect between the two pairs, but I'm not 100% clear on what the procedure for that would be.  http://anlage.umd.edu/Agilent_Advanced_VNA_calibration.pdf

Page 27 is talking about a SOLT calibration with an unknown thru, which is out, given the difficulty in an open standard on a microstrip.  Just to be sure, I tried.  I modified a microstrip short standard into an open, and measuring it with the VNA after it was TRL/TRM calibrated. It was not something I could fit a calibration polynomial to for the open standard. The Short standard was something that could have a polynomial fitted to it, but without a useable open, its pointless. 
Page 28 is talking about TRL calibrations with an unknown thru. At first it seems to suggest to me that the unknown-thru calibration is a separate process, but I cant seem to find anything about that in my VNA calibration page.   Does this actually mean full 4-port calibration with multiple types of standards? I.e. in my case I would fabricate coaxial standards for Offset Thru and Line, characterize them, and then use a short/load standard from a mechanical cal-kit for Reflect and Match. Then the calibration procedure would be do a full 4-port TRL calibration with:
-PCB standards for ports 3/4,
-Coaxial standards for 1/2.
-Use the PCB Thru standard but marked as an unknown-thru for Ports 1-3, 1-4, 2-3, 2-4
-Skip (?) the Line/Match measurements for Ports 1-3, 1-4, 2-3, 2-4 (A lot of the measurements are marked as optional, but I dont want to skip for no reason, but even duplicating the standard entry in the cal kit table, and adjusting the offset delay because of the moved reference plane wouldn't be good, as the longer standards wouldn't cover my top-end anymore)

Which should give me a reference plane at the SMAs for ports 1/2 and on the PCB at the end of the 90 ohm pair for Ports 3/4, so I can get a proper measurement of the fixture, and I can then use an E-cal from there out and just load the fixture file to de-embed out to the flex connector itself.

"Suck it up and just do the 4-port TRL/M every time" is definitely a valid answer here, but that PDF had me holding onto hope that I'm misunderstanding something.

TLDR: How the heck do I get a VNA calibrated so that one port has a reference plane on a microstrip, another has it at the end of the coax, and the error terms between the two are calibrated also, so I can get a proper S21 measurement of a SMA to microstrip transition. 

[Marsupilami]

Do you mind posting a picture or a drawing? It's a bit hard to follow what exactly do you have / want to have connected and where.

[ConKbot]

Sure.  Image added here. Its all hand-drawn so pardon the janky diff pairs. They are length matched with as gentle of a bend radius as possible where the bends are on the actual PCB.  2X Measurement adapters PCBs going to be in use ultimately.

For characterizing the measurement adapter PCB is the "Adapter Characterization" PCB, which has the traces from the "Measurement adapter" PCB with a short breakout to the SMAs (Ports A/B) and a short that I can use to measure electrical length to use auto port-extension to bump the reference plane out the 1-2mm to the end of the 90 ohm diffpair.

For the Cal PCB, Thru is a 0-length thru, nothing remarkable there. Reflect is a very large diameter via to get as best of an approximation of a vertical wall to the ground plane as I can (60 mil hole, ~6 mil to the ground plane).  Lines are meant to stay within the 20 degree-160 degree phase delay for their intended range, and have been verified as correct,  and matches are 2x 100R 0201's laying on the trace, with another low-inductance ground, >30db return loss from 0-500MHz, calibrated at the SMA (I.e. reflections from the SMA connector, resistors, and ground via all included in that 30db)

Currently for measuring the Adapter Characterization PCB, a 4-port TRL cal can get my reference planes as depicted. Bump A/B out with the port-extension, and I can get a .S4P of the microstrip structures upto the FFC, and feed that into the TDR emulator on the VNA, and do what I need to measure the flex cables. But a 4-port TRL cal is required every time.

Ideally, I'd like to extend the reference planes back to the face of the SMA connectors for connectors C/D so an ECal of the cable-set is all that needs to be done once I get the Adapter characterized and fed into the TDR emulator.

My theoretical procedure referencing Page 28 in the Agilent presentation would be:
-PCB TRL/M standards for ports 3/4, (Intended for connectors A/B)
-Coaxial TRL/M standards for 1/2. (Intended for connectors C/D)
-Use the PCB Thru standard but marked as an unknown-thru for Ports 1-3, 1-4, 2-3, 2-4
-Skip (?) the Line/Match measurements for Ports 1-3, 1-4, 2-3, 2-4 (A lot of the measurements are marked as optional, but I dont want to skip for no reason, but even duplicating the standard entry in the cal kit table, and adjusting the offset delay because of the moved reference plane wouldn't be good, as the longer standards wouldn't cover my top-end anymore)
-Cal is complete
-Port-extend A/B using the connector with a short on it on the Adapter Characterization PCB
-1>C, 2>D, 3>A, 4>B on the Adapter Characterization PCB
-Measure .S4P of card

[Marsupilami]

Ok, this helps.
What I would look into is rather than calibrating with it, to use the TRL cal board to get a good characterization of your SMA to microstrip transition, then use that data to deembedd any subsequent measurement.
It might need some thinking as probably it's not a ready to use option on your analyzer but I'm certain that the math works out.
You calibrate all 4 ports with ECal to the SMAs. Then cal two of those, let's say P1 and P2 on the TRL PCB.
Now for P1 and P2 you have two cal planes, P1_SMA, P1_uStrip... you only need one. Basically you have to subtract the two cal planes from each other to find out the network parameters of the transition.
This can be done multiple ways. One is if you can get 8-term error coefficients from the analyzer for the two cals. These error matrices can be interpreted as T-Parameters which leads to a very easy de-cascade just be a single inverse matrix, then transform the resulting t-parameters to s-parameters and you have your SMA->uStrip stransion s2p.
Another way, would be is to use only the SMA cal plane and simply measure the TRL PCB standards, save s2ps and do the math on them manually to extract a single SMA->uStrip s2p. I have better confidence in this.

https://www.rfmentor.com/sites/default/files/NA_Error_Models_and_Cal_Methods.pdf
Look at the 3-term model for basic, you can skip the 12-term (SOLT), read about the 8-term and then there are examples for each a few, it starts with TRL on page 30. and shows step by step how to extract the error coefficients from measuring the standards. Again the resulting 8 error terms belong to 2 2x2 matrices (1 for each port) and those 2x2 matrices can be interpreted as t-paramters. Convert it to S. Profit.

After you have your SMA->uStrip transition you can make any measurement with the 4 VNA ports calibrated at the SMAs and simply dembedd from between the transions.

[virtualparticles]

I would make two or three lines of different lengths on the board, each with an SMA on each end. If you measure all three you can back out the S-parameters of the microstrip. This leaves the effect of two SMA connectors which you can bisect to get the effect of just one. None of these mathematical tricks are easy though. Bisection is particularly difficult since there is no unique square root for a 2x2 matrix.

Also, beware doing TRL calibration on a PCB. The reflection of the connectors usually ruin the measurement unless they are really good connectors and the transitions to are PCB is perfect. I've commonly seen 15 dB RL at moderate frequencies which renders TRL useless.

You can PM me if you need more info. I work for a well known VNA manufacturer.

Cheers!

Brian

[antenna]

Admittedly, I cannot say I understand everything said above, so this may be a dumb question. If the only thing of interest is the loss through the cable, why not solder a termination resistor (the same impedance of the cable) to the end, drive the other end with an arbitrary amount of power, then, using a good high impedance active probe, get a relative voltage measurement between the cable input and that at the termination and find the dB loss via the cables voltage level drop?

[Joel_Dunsmore]

Not sure if you are in a professional environment, but we do make a software product, PLTS, which is purpose built for dealing with these kinds of problems. It has a software option called Automatic Fixture Removal (AFR) that allows you to essentially remove the effects of fixtures used to connect to your device under test.  Here's a link to it: https://www.keysight.com/us/en/product/N1930B/physical-layer-test-system-plts-2020-software.html  This is commonly used for testing back-plane connectors for high-speed interfaces, differential lines such as USB, HDMI as well as removing fixtures in RF and Microwave applications.  Typically works up to about 40 GHz.  Can be used with S4P files measured from you ENA on bits and pieces.  There is one-port version (either one port single-ended or one-port differential) that only requires a short or open at the end of the fixture to characterize it.  "It's the modern way to remove test fixtures".  Here's link that let's you download the application note (sorry, you probably have to give an email address): https://www.keysight.com/us/en/assets/7018-03037/application-notes/5990-8443.pdf