Kirkby calibration kit alternatives?

[hendorog]

After yesterday measurements, I had asked ChatGPT what RL to expect for a good 50Ω load 0-3.2GHz and the answer was high-quality load (metrology-grade, precision loads) better than -45 dB up to 3 GHz.  Standard lab-grade (high-quality but not metrology grade) better than -35 dB to -40 dB across 3.2 GHz.  General purpose 50 Ω terminations (not for precision calibration) -30 dB or worst at higher frequencies. Based on this I keep my modest aim on getting loads -35 dB or better, if I can help it.

As pointed out by virtualparticles, the return loass of your calibration load is what dictates how accurately you can measure return loss in your DUT.

See image attached showing the error, from this document:
https://markimicrowave.com/assets/e01dd483-b079-4da5-9d03-79b6db1f5969/directivity_and_vswr_measurements.pdf

(In an OSL calibration, when the Load is assumed perfect, the Directivity error is exactly the same as the true Return Loss of your calibration load)

For my calibration/measurements, I had increased the trace point to 2000 for 0-3.2GHz.  Surprised hendorog uses the default 201 points for 0-7.5GHz. On the other screen covering ANNE-50+ and ANNEF-50+, S11 and S22 respectively, to 8.5 GHz I cannot see the number of points.  Should not make a hughe difference but couldn't that miss some peaks or valleys, if present?

Potentially, but would have made that measurement and calibration take significantly longer.
In my first post, which had measurements done on a Siglent SNA, I used 500Hz IFBW, no averaging and 2001 points. Sweep time 3.5 seconds.
In the second post, which had measurements done on a Siglent SSA, I used 100 averaging and 201 points. Sweep time 6.4 seconds (incl averaging). Sweep time would have been 64 seconds with 2001 points and 100 averages. Not worth it for measuring a load which was already known good.
The SSA does not permit IFBW adjustment, much to my dismay.

Couple of further points:
Slowing down the measurements with too much averaging and/or too many points on a low end VNA can be worse because they drift more.
Taking more time to measure means the instrument can drift quite far between completing the Load calibration step, and completing your measurement.
The SNA is very stable in comparison to the SSA.

And of course:
Something which has a sharp enough notch response that you cannot see it in the GHz range with 200 points means it has about 37 MHz bandwidth. Seems a bit unlikely to happen by accident in a load.

Hendorog's female ANNE dropping roughly from -40 dB to -75 dB after removal and reinsertion is crazy, no?  When I measured the Smith charts I first looked at the RL (after power cycle, calibration recall, had even removed N-to-SMA and put back to restore same cal plane, etc) and the RL for all loads looked indistinguishable from the original uploaded measurements (to my uncalibrated eye).

Sorry I was not very clear and so you missed the point I was making with that image.
The -75dB images in both of my posts were of the same Agilent load which I used in the calibration. In the first post, from the SNA, I disconnected and reconnected the same load used for the calibration. In the second post it was just left on the port and measured.
This will always result in an excellent return loss, as the VNA is almost always set to assume the cal load is perfect.
If I had not disconnected and reconnected it, and instead made the load the last measurement in the calibration, then the return loss trace on the SNA can be below -100dB
This number tells us nothing about the calibration load itself. I could have used the ANNE load or an ultra cheap NanoVNA load for the calibration, and I would get the same result.

Hendorog's cal load on SSA3075XR with 100 averages is showing roughly below -60 dB (around -70 dB even), and without averages roughly at -40 dB (similar to my measurement). Question.  Shouldn't we care about the worst case RL and engage max hold instead of any averaging? (I think I'm gong to try that, unless I hear it makes no sense).

In my original measurements for RL, I had 10 averages on, but at around 1-2 sec sweep time (cannot recall) I probably did not even wait all 10 runs before taking screenshots. For Smith I do not think I turned on averaging at all. On hendorog's measurement to 7.5GHz of the calibration load I see the same (lots of) jagged edges (be it noise or uncertainty??), same as the measurement of my ANNE cal load (immediately after calibration without removing it). Curious to know why measuring whatever cal load is used results in a more jagged trace than measuring loads other than the reference cal load, independent of overall badness of RL of each load. (Will try to cal with my Long1 and then measure my ANNE, expect the jaggedness to go away, even if I don't know why).

The spikes are due to the noise of the VNA.
So max hold will measure the peak of the VNA noise as well as what is measured from the DUT. It will also capture the drift of the VNA very nicely.
When you are measuring the same load you used in the calibration, then by definition there is no reflection from the DUT being measured.  Any reflection is calibrated out - because it is used to determine the Directivity error of the VNA.

@joeqsmith I like your "the VNA was the limiting factor, not the cheap standards I use with them" comment and since I have a NanoVNA-F3 will try and experiment to see what I find now that I know a bit more about he quality of my loads (how good or bad the NanoVNA is compared with the Siglent, using same loads).

Try on different VNA's:
OSL calibration and then measure the same load without disconnecting it.
See how low you can get the trace by messing with averaging and IFBW (if available), and how long it stays in one place. That tells you something about the quality of the VNA.

Just learning all this and, overall, I am very pleased nobody mentioned DCR in their responses, a topic that was very popular on older posts in this thread ;-)

This is where the Smith chart is useful. You can turn on a marker and you should see a resistance close to the DCR of the load _when the marker is close to the DC end_ of the frequency range. Then observe how it changes as you increase the frequency.

[vsilves]

> I did provide a link for DCR measurements.

Sorry, I genuinely missed it.  However, my point was that I continue to agree with Dr. Kirkby's statements and DCR should not matter so long as low frequency response is not affected, and for that you only need to look at RL, not DCR.

> How do you know what yours is?  You would need something as a reference to compare it with.

Definitely agree that I don't know without a better reference to compare to. 

In fact, I calibrated SVA with (crappy) China1 load and that made (awesome) ANNE looked pretty crappy (1st pic below).  Shows about -12 dB RL (at roughly 3 GHz).  ANNE min specs are -26 dB (at 4 GHz, so should be way better at 3 GHz) and typical -33.81 dB (at 3 GHz). My conclusion, unless ANNE is broken, which I believe it is not, then the China1 is as crappy as I suspected.  Even if China1 cal makes China1..5 RL appear decent (a bit worse than -30 dB RL --2nd pic below).

I remembered I have an SMA 2W 10 dB attenuator rated to 6GHz (on label) and it may be worth measuring with different reference cals.  The 3rd pic shows the attenuator's insertion loss (flat at roughly -10 dB, with a -10 dB TG input).  Using normalized tracking generator function, so not dependent on any external load.  Looking at S11 of the attenuator with a (presumably good similar to ANNE on previous tests) Long1 load OSL cal, I get a flat response roughly at -20 dB RL (see 4th picture).  Port 2 of attenuator was open.  This all makes sense in my mind: -10 + (-10) = -20.  Given that the first TG measurement did not depend on the load, any load that agrees with TG must be right.

Then, if I measure the same S11 of the attenuator after a China1 OSL calibration, results get clearly off from what the TG said (picture 5 below). Wildly swinging from -22 dB to  -7 dB.

I feel pretty good about this.  Like I just proved a theorem or something.  But not as good as a known good cal reference.  There might be fallacies in my logic, and, as I hobby-study more, I hope to find out one way or the other.  I will probably not find any better cal kit, and not worth the $$$ for my hobby purposes.

Just saw hendorog posted replies (still have to read and study to see if I can comprehend).  Again, thanks for all the hints. I learned a lot reading this blog, and Google, over the years. Finally posting some. Honored to get dialog with the big guns, haha.

[joeqsmith]

....There might be fallacies in my logic, and, as I hobby-study more, I hope to find out one way or the other.  I will probably not find any better cal kit, and not worth the $$$ for my hobby purposes.   ....

Consider that when I received that first NanoVNA in 2019, I used the supplied standards to create a demonstration video of it.   I also used the ideal model (told the VNA the standards were perfect).   I made several measurements of various passive networks and everything measured as it should.   Often I won't calibrate the VNA, or it is the last step.  Over the last four years,  I've ran a lot of tests and shown a lot of data using these cheap VNAs and poor quality standards.   I'm the last person that would suggest you need to spend a lot of $$$ for a hobby and learning how to use a VNA.   That's the whole point of these cheap VNAs.

As long as you understand the limitations of your particular VNA (for example not being able to make narrow band measurements with the V2Plus/LiteVNA, drift, effects of the squarewave drive on your circuit) you shouldn't have any problem.    The most common problem I see in the forums is the hams connecting them to their antennas and taking them out with ESD.  Even then, they damage a $200 at most.  Good learning experience and may save a costly repair bill in the future. 

[vsilves]

When I said this in my previous post, -20 made sense for TG insertion loss test only (shown as -10, instead of -20, since it was normalized with -10 dBm TG output). For the SVA S11 measurement I had no idea why it should, or should not, be -20, but it looked like a very round number (coming out of my fourth point of contact).

> This all makes sense in my mind: -10 + (-10) = -20.

Chit chatting with ChatGPT afterwards about what to expect for the S11 on an ideal terminator of known value with no parasitics and open port 2, it spit out some formulas to calculate the reflection coefficient as Γeff = Γopen  x 10 ^ (-2 A ÷ 20), where Γopen = 1 (full reflection since port 2 is open), factor A is the attenuation, and factor 2 accounts for the attenuation happening twice on the way to port 2 and on the way back from port 2, which for a 10 dB attenuator would be Γeff = 1 x 10 ^ (-2 x 10 ÷ 20) = 10^-1.  Then from that it calculated the S11 as S11 ‎ =  20 log10 |Γeff|  or  S11 ‎ =  20 log10 (10^-1) = 20 x (-1) = -20 dB.  Basically, this calculation will always be twice whatever the value of the attenuator used.

Looking at the SVA1032X Tech Sheet specifications, I see the dynamic range for the directivity of the calibration S11, Log mag, Average=50, >50MHz, > 40 dB.  Which I think I should interpret as: there is a -40 dB floor of how low I can see S11 reflections, (I assume) relative to the output power of the TG (port 1 of SVA). Not sure (until I look) if the amplitude button/menu shows or controls output power during S11 measurements.

Earlier, I had measured the S11 (pic 1) and S22 (pic2) of a cheap SMA 10W 40dB attenuator (rated to 3 GHz by vendor). Based on the dynamic range limitations of the SVA, I think I cannot see or do anything with respect to the S11 of this 40 dB attenuator (i.e., cannot expect to see -80 dB).

My arsenal only has two 3 dB attenuators, and two 30 dB attenuators left.  A single 30 dB may also lay beyond the S11 dynamic range. My plan for tomorrow is to measure the S11 of one 3 dB attenuator alone (should read -6 dB), two 3 dB attenuators in series (should read -12 dB), and the 10 dB attenuator in series with a single 3 dB attenuator (should read -26 db).  I think it will be interesting to see how the S11 RL trace correlates with the S22 insertion loss of the attenuators.

[vsilves]

Specs in Sysjoint NanoVNA-F V3 manual show better S11 dynamic range (50 dB) than Siglent SVA1032X (40 dB), if true.

> NanoVNA-F V3 is a Portable Vector Network Analyzer (VNA) with frequency range of 1MH
z ~ 6GHz. It can be used for S11 and S21 measurements. The S21 dynamic range of NanoVNA-
F V3 is 65 dB, while the S11 dynamic range is 50 dB.

Further down a table in the manual states:

> S11 dynamic range:
> 50dB for < 3GHz
> 40dB  for > 3GHz

Will try and repeat whatever I get with SVA1032X later with the NanoVNA-F V3.  My goal is to see how the load in the OSL calibration affects results for S11 measurements of port 2 open attenuators, in order to judge the quality of the 50Ω loads themselves.

[hendorog]

In fact, I calibrated SVA with (crappy) China1 load and that made (awesome) ANNE looked pretty crappy (1st pic below).  Shows about -12 dB RL (at roughly 3 GHz).  ANNE min specs are -26 dB (at 4 GHz, so should be way better at 3 GHz) and typical -33.81 dB (at 3 GHz). My conclusion, unless ANNE is broken, which I believe it is not, then the China1 is as crappy as I suspected.  Even if China1 cal makes China1..5 RL appear decent (a bit worse than -30 dB RL --2nd pic below).

Something that I discovered a while ago.

If you have two loads and you:

Preset the VNA
Calibrate with Load A and then measure Load B and record the trace.
Preset the VNA
Calibrate with Load B and then measure Load A and record the trace.

Both traces are the same.

If I understand correctly, then you did this. But the traces are not the same.

As another sanity check, even the cheap loads generally have decent Return Loss at very low frequencies (your sweep starts at 100kHz)

Your cheap load was around 50.9 ohm - say 51 ohms. That should get you -40dB return loss.
The ANNE load should be below -40dB return loss there.
So the sanity check suggests your recent screenshot should be better than what it is. (The first screenshot in post 226)

[vsilves]

Rechecked screenshots and yes RL traces were different.  Will repeat with SVA1032X and also NanoVNA F V3 and post results in a couple of days.

> Preset the VNA
> Calibrate with Load A and then measure Load B and record the trace.
> Preset the VNA
> Calibrate with Load B and then measure Load A and record the trace.

Looking at RL trace of cheap load (China1) after calibration with ANNE, I see -40 dB only to less than half a division (10 divisions for 3.2 GHz) so less than 160 MHz.  Then ramps up roughly -15 dB around mid span, and maybe -8 dB at 3.2 GHz. First picture below.

Conversely, looking at trace of ANNE load after calibration with China1, S11 starts at slightly better than -30 dB at near DC, ramps up to maybe -18 dB at mid-span, and ends at around -9 dB at 3.2 GHz.  Second picture below.

Picture 3 shows the specs for SVA1032X.  Could you comment how you interpret the limitations for the dynamic range of S11 measurements on SVA mode and what S11 RL levels the SVA1032X should be able to measure?

Picture 4 is the ANNE S11 RL after calibration with a presumable decent/regular load (other than ANNE itself).  Shows that it stays below -40 dB across entire span.

Picture 5 is the cheap (China1) S11 RL after calibration with same decent/regular load (Long1).  I don't have a China1 RL after Long1 cal, but picture below shows China2 (very similar set of 5 China bad loads).

Again thanks for kind and helpful comments!

[hendorog]

Something is off with the measurement in your second picture.

Can you try the process again:

Preset then calibration with China1 and then measure ANNE and take a screenshot
Preset then calibration with ANNE and then measure China1 and take a screenshot

Make sure you don't change anything - use the same open and short and choose the same cal kit in the SVA for both.

[hendorog]

Picture 3 shows the specs for SVA1032X.  Could you comment how you interpret the limitations for the dynamic range of S11 measurements on SVA mode and what S11 RL levels the SVA1032X should be able to measure?

The S11 'dynamic range' is called Effective Directivity.

That is the directivity that the VNA is capable of measuring after calibration.
VNA Calibration corrects the directivity error in the hardware, which is why the VNA can more accurately measure RL than you can using the SVA tracking generator and SA mode.
However the error correction of directivity is _directly_ dependent on the return loss of the calibration load.

Aside: You can cheat that rule on an SVA, if you have someone with a real cal kit measure your OSL standards and give you s1p files. Then you can take your own measurements on your SVA, save them as s1p files, and then do the calibration offline on a PC.

Effective directivity can be measured, but requires an airline.

The better the effective directivity, the more accurately you can measure return loss.

The Effective Directivity spec is 40dB. If your ANNE load is better than that and you turn on averaging at least as much as the spec says you should achieve that.
Attached is an image from the pdf I pointed to earlier. It shows the measurement error for a given Directivity value.
See the 40dB directivity trace on the chart.

[vsilves]

@hendorog, I did the cal A measure B, then cal B measure A, in order to compare the reciprocal return loses.  See PDFs.  The RL of the opposite load is shown in step 5.

First PDF is ANNE vs China1 (bad bad load in my perception) and the RL of each other does not look identical.

Second PDF is ANNE vs Long1 (decent load in my perception, from NanoVNA most likely).  The RL of each other DOES look identical!

Thanks for your analysis and comments.  I hope this means China1 is messed up and my SSA/VNA is not.

[vsilves]

This is the other experiment I wanted to do.  OSL cal with ANNE then measure some attenuators to observe expected vs actual RL obtained.

I think the observed RL meets roughly the expected RL for 3, 6, 10, and 13 dB attenuators (got roughly 6, 12, 20, and 26 db RL). Maybe no enough dynamic range for 30 dB attenuator.

The idea would be to repeat this after first OSL cal with China1 to see how the expected RL is "not" obtained.

[hendorog]

@hendorog, I did the cal A measure B, then cal B measure A, in order to compare the reciprocal return loses.  See PDFs.  The RL of the opposite load is shown in step 5.

First PDF is ANNE vs China1 (bad bad load in my perception) and the RL of each other does not look identical.

Second PDF is ANNE vs Long1 (decent load in my perception, from NanoVNA most likely).  The RL of each other DOES look identical!

Thanks for your analysis and comments.  I hope this means China1 is messed up and my SSA/VNA is not.

Ok thanks for all of that effort. It that looks more sensible - not identical but the return loss charts are now fairly similar.
Because there is still a difference, some inconsistency must be causing that.

The Long1 pdf looks much better than the China1 pdf, both in terms of RL and the symmetry.

Likely there is some sort of mechanical issue with the China1 load, the trace you get when you put it back on and measure it is quite different to the others.
Other things to check, the N-SMA adapter, the connections are not clean, something is not tight, have you left to warm up for an hour or so?
Check that nothing changes when you tap/wiggle the load (you will need to turn off averaging and have a low number of points to see anything)

The 180 degree phase rotation which is visible on your Smith chart images is correct I believe.

You can do the same test with the attenuator instead of the China1 load.
It does not matter if you Cal with the ANNE load and measure the attenuator, or Cal with the attenuator and measure the ANNE load. You should get the same trace in both cases.

If you don't get the same trace, then I think something must be wrong.

If you have time/can be bothered to take measurements of all of your SMA devices (incl OPEN and SHORT), and save them as s1p files with appropriate filenames, we can run the calibration offline on a PC.
Please ensure the SVA has been turned on for an hour or two first and is in a place with no draughts or direct sunlight and a stable temperature.
No calibration is required.
50 averages and 1000 points (adjust points if you need to keep the measurement time under 10 seconds)

[vsilves]

The 180 degree phase rotation which is visible on your Smith chart images is correct I believe.

This is a great tip, along with original tip to measure A->B and then B->A.  May try with attenuator vs ANNE, but also have second (similarly good) Long2 50Ω load to try. I have a LPF, Combiner/Splitter, and a Directional Coupler from Mini-Circuits to about 2GHz and may try to match spec sheet graphs (to prove cal load is ok if specs are matched), if I can figure out how to measure.  Also have a mixer but a bit harder to figure how to measure for now.

Of all the possible pitfalls you mentioned, I think everything is steady and clean, except yes I may have violated not enough warm up time of instrument (for China1).  May repeat, but I have observed similar behavior after 1+ hour warm up (in stable basement temp, no drafts).  Two of my 5 China 50Ω loads (not this China1) intermittently show as "open" if I wiggle the core while holding the nut. This does not happen with any of my other loads. So a mechanical problem is very likely for all China# loads.

If you have time/can be bothered to take measurements of all of your SMA devices (incl OPEN and SHORT), and save them as s1p files with appropriate filenames, we can run the calibration offline on a PC.

Very interested if you could analyze any .s1p files I may generate for OSLs.  It appears SVA1032X can NOT save .s1p files, only .csv files.  I will figure out how to pull S11 traces using python, and generate .s1p files (which may be faster than saving .csv and processing).  With that end, what would be more convenient to process for offline calibration on a PC?:

1. # MHz S DB R 50
frequency logmag phase

or

2. # MHz S RI R 50
frequency real imag

Once more, thanks for your time and comments, which help immensely a hobbyist like me.

[hendorog]

The 180 degree phase rotation which is visible on your Smith chart images is correct I believe.

This is a great tip, along with original tip to measure A->B and the B->A.  May try with attenuator vs ANNE, but also have second (similarly good) Long2 50Ω load to try. I have a LPF, Combiner/Splitter, and a Directional Coupler from Mini-Circuits to about 2GHz and may try to match spec sheet graphs (to prove cal load is ok if specs are matched), if I can figure out how to measure.  Also have a mixer but a bit harder to figure how to measure for now.

Of all the possible pitfalls you mentioned, I think everything is steady and clean, except yes I may have violated not enough warm up time of instrument (for China1).  May repeat, but I have observed similar behavior after 1+ hour warm up (in stable basement temp, no drafts).  Two of my 5 China 50Ω loads (not this China1) intermittently show as "open" if I wiggle the core while holding the nut. This does not happen with any of my other loads. So a mechanical problem is very likely for all China# loads.

If you have time/can be bothered to take measurements of all of your SMA devices (incl OPEN and SHORT), and save them as s1p files with appropriate filenames, we can run the calibration offline on a PC.

Very interested if you could analyze any .s1p files I may generate for OSLs.  It appears SVA1032X can NOT save .s1p files, only .csv files.  I will figure out how to pull S11 traces using python, and generate .s1p files (which may be faster than saving .csv and processing).  With that end, what would be more convenient to process for offline calibration on a PC?:

1. # MHz S DB R 50
frequency logmag phase

or

2. # MHz S RI R 50
frequency real imag

Once more, thanks for your time and comments, which help immensely a hobbyist like me.

Copy, that makes sense with the mechanical load issue.

On the SSA if I hit the File button then Save Type there is an option called SNP. If that is available in the SVA then that will save an S1P file.
Otherwise yep the CSV method works and either of the two formats is fine.
I think from memory you need to be in Smith Chart mode to get it to export both R + I or gain + phase in a single csv.

I will use scikit-rf to do the cal, and I use it in a Jupyter notebook.

The cal and then measurement can be done the two different ways on just the one set of files. That means you don't get the connection/reconnection error and so the two different cals will result in perfect overlaps.
If you end up coding the extract s1p then maybe do several disconnect/reconnects of the loads and extract multiple files so that error can be viewed.

[vsilves]

Here are the .s1p files for a few of my 50Ω loads and my open and short SMA connectors, all in .s1p format and descriptive file names.  Some with repeated removal and reinsertion (test 1, test 2, etc).  Uncalibrated instrument.  I had never opened any .s1p file before, but I installed a free Mac app named splotRF, and was able to open the .s1p files and verify that the Smith chart trace shown in the app looks identical to the instrument's Smith chart. Hoping I did not code anything wrong on the output files.

Need to improve code.  Currently pulls FREQ, LOGMAG and then PHASE, PHASE, in two goes, so it messes with any AVG when changing mode from LOGMAG to PHASE, but only takes like a second to pull the 1000 points.  Need to figure out an SCPI command that will return FREQ, LOGMAG, PHASE (or FREQ, REAL, IMAG) all in one go to keep existing AVG.

[vsilves]

The hint of using scikit-rf was both surprising and great.  Scikit-rf looks very intimidating and I know nothing about it (probably need to study more theory before I can use it).  However, plotting multiple .s1p files side by side as Smith, LogMag, or Phase charts using scikit-rt was so easy!  I prefer this kind of rudimentary interface to fancy programs.

In any case, trying not to bombard the forum with files, but I fixed my SCPI script and now I can pull RE, IMG data at once from Smith Chart view (like you said) and that pulls the 50 AVG trace or whatever trace 1 has to the .s1p file.  Quick to pull (under 1 sec?), but based on your drift comment I should not let the averaging itself take too long either before I pull a trace (so reduce number of points). With some practice I should be able to pull more robust .s1p traces under various conditions.

[vsilves]

Just playing with scikit-rf a bit more and I think I figured out how to do offline calibrations using .s1p files.  Results in attached PDF for ANNE, long1, and china1 seem to meet expected traces (identical logmag and 180º shifted phase for A <--> B).  Generated traces look a lot like the actual measurements I remember.  Pretty amazing that you can generate all that from only the uncalibrated .s1p files for the OSL.  I had no idea this was possible until hendorog mentioned it. 

Unfortunately, this says nothing about the quality of the loads (without a known good reference).  But I think loads of similar response are, duh, similar.  Just don't know (for sure) which ones are the good and which ones are the bad.

Would like to get example OSL .s1p files with good descriptions to play with. Not sure if manufacturers or users post this somewhere.  I will be searching eevblog ;-)

[hendorog]

Nicely done!

Sorry I have been a bit too busy to get into this, but you have done exactly what I would have done.

It was messing around with scikit-rf that lead to discovering the reciprocal relationship and the 180 degree phase change. I think it provides a nice sanity check I think.

I have not seen that mentioned anywhere before, this is the first time I have posted it and thank you for verifying it.
That probably means it is so obvious that everyone except me knew about it already

The maths side seems to work out too.

[joeqsmith]

...
Would like to get example OSL .s1p files with good descriptions to play with. Not sure if manufacturers or users post this somewhere.  I will be searching eevblog ;-)

I'm not following but it would be easy enough to synthesize data for a set of standards.  You could make them anything you like.  You could then use those to calibrate your virtual VNA and then measure some virtual parts.   When playing with METAS, that is basically what I was doing.   
   
I doubt this will help but when I attempting to tighten my measurements with the LiteVNA, I characterized the standards I use with it.   This was done using my Agilent PNA which was calibrated using an ECAL that was characterized to some standards I loaned.  The short and open standards are basically what was supplied with the LiteVNA64.  The load is a mini-circuits ANNE I hand selected using the PNA and metrology grade kit.   You can find the touchstone files using the dropbox link in my signature.  Goto VNA_Support/Solver64/Calibration/LiteVNA64.  That directory contains the data for the SOLT.  These were swept from 300k to 9G.  1601 data points.  500Hz IFBW.  0dB.   Format is Freq S11 dB/Ang.   

I used METAS to derive the coefficients for these standards.   That data is hardcoded into Solver as a default.  You can view it with that tool, or just recreated it with METAS.   

Again, I noted little difference between using the raw Touchstone files and the coefficients METAS selected.   Measuring a T-Check that performed well on the PNA to 9GHz, using the same coefficients and calibrated with the same cheap standards, falls apart with the LiteVNA around 4GHz.  I used the same software when running both VNAs, same math.  That was also checked against Agilents built-in firmware.

[vsilves]

Still loving skrf and got many nice plots, but looking at Mini-Circuits ZFDC-10-5-S+ (directional coupler) I'm struggling to plot "directivity" correctly using their posted .s3p file. The first picture is my logmag plot of their .s3p file. The second picture is their posted graph for directivity (a bit more detailed than the small graph in the main spec sheet).  The third picture is me trying to plot directivity from the .s3p file (all in dB, plotted with a minus sign in order invert the traces and directly compare with picture 2):
- First trace in red is S31-S32 (simple definition).
- Second trace in blue is S31-(S32+S21), which I figured from the handbook of microwave component measurements, where it is defined as = isolation / (coupling * loss).
- Third trace in green is S31+S21-S32 (exactly as stated in the Marki document that hendorog pointed at earlier).

Comparing visually pictures 2 and 3 I'm disappointed as they do not seem to match very well, only very generally. Are the calculations wrong or is there no way to calculate directivity directly from S parameters and the only way is to measure S21 (logmag at CPL port) with out port open/short and then out port terminated, and then subtract?

[hendorog]

Did you use the ANNE load to terminate the unused port?

[vsilves]

To highlight something that does work here is a simple DTF graph.  Even though skrf does not have TDR/DTF that I know of, using a simple numpy IFFT or CZT on the Siglent VNA .s1p file (about 8.5 cable here) yields results similar to Siglent's internal DTF tool.

On another note, if anybody has both (1) a known good SMA 50Ω calibration load and (2) a Siglent SVA1032X (or better yet a converted Siglent SSA3021X Plus like mine), it would be great to get two .s1p files (one with uncalibrated SVA and one with OSL calibrated SVA using same load for calibration).

I found lots of .snp files in this forum and around from helpful members but have not identified any .s1p files that cover a load measured both with and without first calibration of the SVA, let alone the SVA being an SVA1032X.

[vsilves]

Did you use the ANNE load to terminate the unused port?

The .s3p file is the one posted by Mini-Circuits here, so I did not perform any measurements.  Just trying to reproduce their results.  I see other offsets between their own tech spec sheet graphs and their actual posted .s3p file data.  If I try to measure with only the 2 port Siglent SVA I will try and thing what I terminate and what I do not terminate :-)

https://www.minicircuits.com/WebStore/dashboard.html?model=ZFDC-10-5-S%2B&srsltid=AfmBOoo9m8hNMxr51KLYybor5wlXTa6vgoVi1vr7YaLZAOkPx3aapP_V

[hendorog]

Oh sorry now I understand.

Hmm, that is interesting.

[Miek]

Even though skrf does not have TDR/DTF that I know of

It does by the way, here are a couple of examples:

https://scikit-rf.readthedocs.io/en/latest/examples/networktheory/Time%20domain%20reflectometry%2C%20measurement%20vs%20simulation.html
https://scikit-rf.readthedocs.io/en/latest/examples/networktheory/Time%20Domain.html