measuring chip capacitor against frequency using network analyzer.

I have an Agilent PNA network analyzer up to 40GHz.
And I have a bunch of HF HQ capacitors, beginning from 0.1pF in nominal value.

I have a milling machine and connectors to design a microstrip line for measurements.

How can I measure actual capacitance of those capacitors against frequency?

------------------------------
One option I was advised is to make a few 50-ohm microstrip stubs, shorted and open. Calibrate the PNA using them as 1-port loads. Then take an open-ended stub and shunt it with a capacitor at the end. Then, estimate the capacitance from S11=(Zx-Zo)/(Zx+Zo).
This method appears one-legged because the capacitor should be soldered to the stub and then disposed of. Another problem is hand-made micro-strip stubs, it is just not right to calibrate PNA with this stuff. The results are very strange.

I found Agilent mentioning some 2-port measurement setup. But no explanation, how it should be organized and how precise measurements should be performed. If possible, I want to avoid soldering the capacitors. Just pinning them to the circuit with a PFTE tube would be enough, I believe.

Maybe one option would be to measure at lower frequency to reduce the impact of solder and microstrip and extract a model of your capacitor. If you really want to measure up to 40 GHz, I personally think that the usage of microstrip is not a good idea. Furthermore, the impact of the solder joints on measurement results increases. And if you solder by hand, every other cap with same values will produce different measurement results because you can't solder all caps in the same way. But the company should provide some measurement results, or not?

OK, up to 20GHz would be enough.
Yes, that's why I don't like soldering, though I can make it carefully under microscope.
I don't know how to extract the model, those 0.1pF multilayer chip capacitors are completely unknown to me.
Those are some experimental samples we got many years ago. Today I want to set an experiment, and 0.1-02. pF values seem just perfect. I got just few approximate resonance curves for higher values of these series. Not for those caps I am interested in. The best for me would be just measure every cap I use vs. frequency without destroying it.

So, my Idea was to build some kind of microstrip or coplanar line with a gap, which I can fill with a solid metal bridge, or a capacitor of interest. Putting it on the top of the strip is easier than soldering as a via. The technique, I don't know how to organize it in better way. And accurate parameter extraction.... Even 2nd rank matrix equation seems too complicated to trust it. There should be some good approximations.

I think, even 20 GHz is too much, I'm designing homemade pcbs for 10-14 GHz, it's difficult to get what you want but possible if you well understand how and what you're doing. And I also solder sometimes under microscope but nevertheless the results will differ.

Furthermore, you should consider that a use of these capacitors at frequencies > 14-15 GHz is not useful, and it's better to utilize microstrip or cpw structure for capacitors and inductors.

Back to your problem: I suggest, make a pcb with 50 Ohm input and output trace and solder or cap between them. And use a really small amount of solder, just what's necessary to connect well the cap. Then move the calibration planes to the caps by adding a delay in the setup of your vna. If you don't know your delay, make a pcb with an input or output trace and measure first. I think it's the fastest way to measure your caps. (In vna it should be called port extension.)

Thank you, I will consider this way, however I am not sure that a real life microstrip can be approximated just with a delay... but comparing to one-port shunt measuring it looks more solid.

It depends on your vna, some vna also take the loss of the transmisson line into account. If you have some EDA-software like agilent, you can also try to measure the feeding trace alone and without cap and import the results into ads. There you can de-embed the feeding lines, and approximate capacitor's characteristic.

put 3 of them in series along a 50 ohm line, and measure Mag S21. Curve fit to get the capacitance value, and look at the high frequency end to see if there are parasitic problems.

biff44, thank you for trying to help, but i didn't get the idea. maybe that is a lack of education. what do you mean by "along the line"? And the curve fit, what model should I use?

BTW, the VNA is Agilent PNA N5224A. And I have HFSS. I can calibrate anything I want there, but by my experience, deembedding with inverse ABCD matrices in Matlab gives more sensible results. The thing I don't know is a good procedure to avoid parasitic effects and singularities in matrix relations.

Concerning your first idea, I attached you 2 pdfs, maybe you have it already. It explains the usage of test fixtures.
----
Also a paper about s-parameters and passive components and their modeling.