Port extensions on vector network analyzers

Hi - when tuning a PCB mounted RF antenna I typically do something like this:

calibrate to end of VNA.
Solder cable to antenna, and then place a short across the very end (antenna end) of the cable. In my case, the short is typically a blob of solder.
Port extend to there.
Remove short, and tune antenna.

I was at a conference recently and the presenter gave a similar description of setting up for tuning, except that he said it's best to use a capacitor that has a SRF of the frequency of interest.

What do you all think is the better way of doing a port extension?

Assuming you have thin (lossy) cables for antenna connection, calibration to the cable end is a better method, I think, even if the calibration elements aren't perfect.

I am not sure exactly what you mean by "SRF of the frequency of interest". If I was testing a 2.4 GHz antenna, and I was using a capacitor with a SRF of 2.4 ghz as some sort of calibraiton, I would be calibrating with a complete unknown!

I like the first method. One addition, when I have the short circuit in place, I adjust the "electrical delay" to form the smallest circle in S11 on a polar display. I then adjust the "electrical phase" so that the small S11 dot gets place centered at -180 degrees.

Tuning antennas is not a precise task, and therefore does not require the most precise calibration. You just need to know where the center of the smith chart is, and if an added tuning element should be inductive or capacitive. after that you just hack away until it is "tuned" as best as it will get.

I have written a bit about calibration here.
In other pages on that site can you find how to avoid component guesses or why to include a matching network from first antenna design step :shock:

It seems like you also prefer open-short-load calibration with non-perfect elements at the antenna "reference plane" over exact calibration with a calibration kit and port extension to the antenna plane. As you apparently do a lot of work with antenna calibration, did you compare the results of both methods?

FvM said:

Assuming you have thin (lossy) cables for antenna connection, calibration to the cable end is a better method, I think, even if the calibration elements aren't perfect.

Click to expand...

Hi FvM - I calibrate to the end of the cable coming from the VNA with a SMA cal kit. But then I have another cable that is attached to my antenna (as I don't have a connector built onto the PCB, due to size constraints). So I short the end of that antenna and port extend to it. That cable is normally 5cm long or so.

That's probably a good compromise. I was thinking of situations with a thin and lossy antenna cable of e.g. 0.5 to 1 m length, that's used with ferrite beads to minimize the cable effect of the antenna impedance. You most likely don't wnat to have a connector near the antenna.

As you apparently do a lot of work with antenna calibration, did you compare the results of both methods?

Yes, my recommended calibration procedure is not a perfect method, especially the "open" is obvious that it adds a reactive error, but most of that effect is taken care of by calibration and it is so far best found practical method. For checking amount of error do I typically use a resistor 10 kOhm as load and compare what the VNA measure. As stray-capacitance and inductive losses is expected to be same as for the short and 50 Ohm calibration-resistor can I pretty good see if reactive load measure correct enough. As I often design antennas with that kind of a few kOhm as impedance must calculating of a matching network be based on very correct values.
Doing calibration with a calibration kit, and mechanical compensate for thin coaxial cables length do work ok but in GHz-range and with intention to design a good T or PI matching network for a wide band antenna can a few degrees measurement error in smith-chart be difference between success and failure.
Making a impedance correcting PI network a bit beyond "fine-tuning" in GHz-range, requires that calibration must be very good if theoretical calculated correcting network and final measured smith-diagram should have any similarities. I do mostly succeed with that kind of network in first try now, but it did take me long time to learn how.