Strange results from LibreCAL

[iwbnwif]

Please can I check on the correct procedure of calibrating using the LibreCAL.

If I manually calibrate using the buttons on the LibreCAL and the Calibration->Calibration Measurements menu in the LibreVNA GUI, then the calibration seems to be very accurate (e.g., load appears in the middle of the Smith Chart, open / short at the extremes).

However, if I use the Calibration->Electronic Calibration then the calibration seems to be off:

It seems to me that the coefficients file is incorrect.

Do I have to do anything special on first use to set the coefficients file or is the factory provided one supposed to be correct?

The only thing that I can think of is that I did a firmware update when I first connected the unit - is there an incompatibility between firmware and coefficients files?

[jankae]

Please can I check on the correct procedure of calibrating using the LibreCAL.

Sure. Using the LibreCAL should be straight forward. You just connect it, open the electronic calibration dialog and start the calibration. Everything else is taken care of automatically.

First question: what is connected to the LibreVNA in your screenshot? I am going to assume that you have connected the LibreCAL with the ports set to the "LOAD" option in this answer.

Lets start with some VNA calibration basics: Without a calibration, a VNA is almost useless (as I am sure you are aware). The calibration is used to tell the VNA what a certain (uncalibrated) measurement actually means: you connect a SHORT standard and tell the VNA "this is what a SHORT looks like". And if you do this with 3 different standards (typically OPEN, SHORT and LOAD), you can calibrate the response for one port.

The crucial part in the calibration is that whatever you attach to the VNA actually matches what you tell the VNA it is. Just from the perspective of the math in the calibration, there is no need to actually use an OPEN, SHORT and LOAD. The only requirement is that three sufficiently different standards are used. In practice, this is best achieved by using OPEN, SHORT and LOAD but in theory you could also use different standards as long as you can accurately describe how they are behaving.

And this is where calibration kit files come into play. No OPEN, SHORT or LOAD standard is actually perfect. No matter how precise they are machined, there are always small imperfections. So you use the calibration kit files to tell the VNA about these imperfections. Example with the OPEN: "this is what an OPEN standard looks like, but it actually it has this delay and that fringing capacitance". The VNA takes this information into account when calculating the calibration coefficients.

Whatever standard you use during the calibration will show up exactly as described by the calibration kit file after the calibration. This is the logical consequence: you connect something to the VNA and tell it how its response looks like. If you then connect the same thing after the calibration, the VNA should come back with: "hey, this looks exactly as you have described it earlier".

So this

If I manually calibrate using the buttons on the LibreCAL and the Calibration->Calibration Measurements menu in the LibreVNA GUI, then the calibration seems to be very accurate (e.g., load appears in the middle of the Smith Chart, open / short at the extremes).

simply means that you have set the calibration kit file to assume perfect standards (which is the default if you do not modify it). If you then connect the same calibration standards again, they HAVE to show up as perfect dots in the smith chart - because that is what you have told the VNA they look like.
Note that this does NOT mean that your standards are actually perfect. The VNA can not tell you how good your standards are. It is a bit of a chicken-and-egg problem: you need a calibrated VNA to accurately measure standards but you also need accurately defined standards to calibrate a VNA.
Basic example to prove my point: perform a SOL calibration using the mechanical standards but swap the LOAD and the SHORT (e.g. when you tell the VNA that you measure the SHORT, you actually connect the LOAD and vice versa). If you then connect your SHORT standard after the calibration, it will show up in the center of the smith chart - because it has the response which during the calibration you told the VNA was a LOAD. This LOOKS very accurate but the calibration is completely wrong because you have used standards that do not match their description.

The mechanical standards supplied with the LibreVNA are reasonable close to perfect. Don't get me wrong, they are still cheap standards in the overall scheme of calibration kits but for most measurements they are good enough. Crucially, the OPEN and SHORT have almost no delay, their reference plane is almost exactly at the mating surface of the connector. Unless you need very very precise measurements, it is okay to leave the calibration kit coefficients at their defaults (which will get you these nice looking perfect dots on the smith chart after calibration).

On to the LibreCAL: Its standards are far from perfect. There is significant delay (there are traces from the connectors to the internal switches), attenuation in the switches and the LOAD also has significant reflections. Sounds bad? It would be, if you would still assume perfect calibration standards. But remember: the VNA does not care that the standards are a perfect OPEN, SHORT and LOAD. All it requires are three sufficiently different standards that are well known. And that is what the LibreCAL has: the actual S parameters for all standards are measured for every assembled LibreCAL individually and stored on it. When you use it with the LibreVNA, these coefficients are extracted during the calibration. The VNA measures the (imperfect) standards of the LibreCAL and performs its calibration while taking the S parameters of the standards into account. The LibreCAL is designed to keep the standards stable across temperature and it was measured using a good VNA and calibration kit during manufacturing. This means that the coefficients stored on it very closely match the actual behavior of the standards and that means that you can perform good calibrations with it.

But if you afterwards connect the LibreCAL and set it to LOAD (which I assume is what you have done), you do not get a perfect dot in the center of the smith chart, because that is NOT what the LibreCAL LOAD standard actually looks like.

Long story short: VNA calibrations are complicated and it is much more important that your standards match their descriptions rather than how perfect your standards are. I do think that everything looks good on your end (with mechanical standards and with the LibreCAL) from the information I have so far.

[iwbnwif]

Firstly, thank you very much for taking the time to write such a clear and detailed reply.

Yes, your assumption is correct - the screenshot is of the LibreCAL set to Load on both Port 1 and Port 2.

It all makes perfect sense now, so the traces in my screenshot are showing the imperfections in the LibreCAL ports. If, after running Electronic Calibration, I replace the LibreCAL ports with the 50-ohm standards at the end of the SMA cables, then I see the near perfect match as expected. De-embedding allows me to take out the extra length caused by using a female-female adapter.

The fact that the LibreVNA now shows a 'near perfect' match (e.g., 51.12+j0.04) rather than a perfect match is more than likely down to slight imperfections in the 50-ohm standard and variable room temperature. This is just an inquisitive observation and I certainly don't need anything like that degree of accuracy at the moment.

This has been very interesting and I can see that I am going to learn a lot from this new device! Thank you once again.