VNA for cable characterization

[joeqsmith]

If we calibrate using a stop frequency of 4 and 2GHz you can see how the measured impedance of the connector is less.  We just don't have the resolution to show what is actually going on but 9GHz is certainly going to provide a much clearer picture than 2GHz. 

For the two added data sets, I hand tightened the standards during calibration and hand tightened the cable to measure it. 

[joeqsmith]

No change other than loosening the test cable's connector and re-tighten by hand (of course, the VNA cables moved, along with the test cable).   That foil tape and sticking things to the table don't seem so silly anymore....

[Pinörkel]

Sorry for the late reply. I could not access my computer for the last week.

Thinnest doesn't tell us much.  Materials will play into it.  Even knowing that, it still offers nothing.  You need to measure the torque then provide that detail.

Yes, materials play a role, but the information I was after was not to determine the exact torque of my hand tightening. Over-tightening can damage connectors, under-tightening can only mess up measurements. I just wanted to use the 3D printed torque wrench as an upper bound to check, if my hand tightening force was likely to over-tighten and that information was easy to get by using the weakest version and applying only 2/3 of the force the inventor of this design did.

If you like to guess, by all means, fingers work fine.   You mentioned early on you have one of the books I read.  Have you taken the time to read it?  It's a pretty good introduction into how to use a VNA, and is filled with tips.  Sadly, I have not found a way to absorb the materials outside of taking the time to read them.

I read about 100 pages of the book. But since it was kind of late that night, I will have to re-read it selectively.

The coax of course is MUCH longer than the connectors.  So now the upper frequency has less of an effect on the impedance and we should get the same values if there were no other variables.   The fact yours changed so much tells me there are other errors with your setup.   

Yes there definitely seems to be some kind of error in my setup. The longer I play with this, the more repeatable the issues get. I am suspecting some systematic errors, but systematic errors should be killed by the calibration, according to the book. So, I started to check the basics like instrument warmup times and checked the calibration results.

I found the time of 30 minutes that I let the liteVNA warm up to be too short. The S11 noise floor of my device seems to transition into a steady state between 60 and 90 minutes. While checking that, I noticed something strange. The S11 trace on the display of the liteVNA and in the Solver64 software did not match when using the exact same sweep parameters. What I mean by that are not some small noise differences, but qualitative shape differences of the curves and an approximately 10dB lower noise floor on the liteVNA display. I redid the calibration many times to rule out calibration issues to cause this difference, but re-calibrating had no visible effect on this. For cross checking, I downloaded the VNA View Qt software and found that the display there matched the one on the device. So I could simply switch the software while leaving everything else untouched to compare the VNA View and Solver64 software output. The differences were still there, and then they somehow mostly disappeared for the newer calibrations and I could not find the reason for this. From there on the noise floor looked like this:

VNA View noise floor:


Solver64 noise floor:


Another thing, I noticed while using the up to 9 GHz range, proposed by joeqsmith, was that the noise floor went up significantly above 6.3 GHz, which is of course to be expected on this device. However, when checking the validity of the different calibrations, I found the ones going up to 9 GHz to be affected by a lot of noise, when viewing the short, open and load conditions in the Smith chart. With the calibrations up to 6.3 GHz these result in a small dot at the expected positions for short, open and load. With the ones up to 9 GHz the result is more like an angry ball of wool. This can be reliably reproduced by re-calibrating and just changing the upper sweep bound between 6.3 GHz an 9 GHz.

Calibration test with load 300k-9G 1x magnification:


Calibration test with load 300k-9G 10x magnification:


Calibration test with load 300k-6.3G 1x magnification:


Calibration test with load 300k-6.3G 10x magnification:

[joeqsmith]

Over-tightening can damage connectors, under-tightening can only mess up measurements.

I tend to move things around when using the equipment.  This means the cables can end up with a rotation force.  If the connectors are not tight and the cable rotates, the two center pins can have rotation.   For the cheap cables and connectors, maybe you don't care too much.  Maybe if you start using precision parts, you care a little more.  I posted some photos from another website showing some high grade HP connectors that were damaged this way.   My advice is form good habits.   

I found the time of 30 minutes that I let the liteVNA warm up to be too short. The S11 noise floor of my device seems to transition into a steady state between 60 and 90 minutes.

Depending how stable you need it, it may not ever be good enough.  There was a reason I modified the one LiteVNA to use an external low phase noise clock.

While checking that, I noticed something strange. The S11 trace on the display of the liteVNA and in the Solver64 software did not match when using the exact same sweep parameters. What I mean by that are not some small noise differences, but qualitative shape differences of the curves and an approximately 10dB lower noise floor on the liteVNA display. 
I redid the calibration many times to rule out calibration issues to cause this difference, but re-calibrating had no visible effect on this. For cross checking, I downloaded the VNA View Qt software and found that the display there matched the one on the device. So I could simply switch the software while leaving everything else untouched to compare the VNA View and Solver64 software output. The differences were still there, and then they somehow mostly disappeared for the newer calibrations and I could not find the reason for this. 

I have not used QT and don't run the LiteVNA stand alone.  That said, it's just some basic math and I would not expect there to be much difference if at all.  Of course, this assumes that in all three cases,  the errors due to the three calibrations are ZERO.   My guess is it is a user problem in not knowing all the fine details, some of which is documented in that large software thread.   I'll take a stab at it and say you have not changed Solver64's IFBW to match the LiteVNA's default.  I mentioned this in that thread as well as a few videos why I use the wider bandwidth and sacrifice noise for speed.  Then there is the isolation. 

My advice is take some time to read more about some of these details so you have a better understanding of what is going on.  If you really feel my software has some math problem compared with the LiteVNAs firmware, I can take the time to run a few tests using their CF card (no software).   QT I know has at least one math bug as I changed my software to handle importing their Touchstone files.   

Another thing, I noticed while using the up to 9 GHz range, proposed by joeqsmith, was that the noise floor went up significantly above 6.3 GHz, which is of course to be expected on this device. However, when checking the validity of the different calibrations, I found the ones going up to 9 GHz to be affected by a lot of noise, when viewing the short, open and load conditions in the Smith chart. With the calibrations up to 6.3 GHz these result in a small dot at the expected positions for short, open and load. With the ones up to 9 GHz the result is more like an angry ball of wool. This can be reliably reproduced by re-calibrating and just changing the upper sweep bound between 6.3 GHz an 9 GHz.

Of course, you could use a lower IFBW before running the calibration to try and improve it.  This is really up to you.   The 9GHz is really to squeeze that little bit of extra resolution from the time domain measurement.   The dynamic range at 9GHz is maybe 10dB or so.  I've shown some data using the LiteVNA in the X-band compared with my PNA.  You can see something, sort of, that kind of makes sense.   

Anyway, good to hear you are making some progress with it.   

[joeqsmith]

Using the same settings as Solver64 with the LiteVNA64's firmware running standalone, then saving Touchstone to SD card and importing with METAS.  I'm not seeing these big differences you claim.   

I've been told I have math errors before.  The last time was when using it with waveguides.  Again, it is certainly possible there are problems with it.  Actually, I know there are as some features I started to work on never panned out.    But that basic software is almost two decades old and is used on a regular basis.  I've also frequently compared the results with a completely different VNA, cables, standards, software.  Even in this thread.   What you are describing would be a major problem.   

If you do feel there is a problem, you will need to provide enough details for me to reproduce it.   

[joeqsmith]

This is the article I mentioned showing some damaged precision connectors. 

It's a wise idea to label the kit with a notice about being drawn and quartered if one rotates the standards while tightening them.

https://www.microwaves101.com/encyclopedias/how-to-not-trash-a-calibration-kit

Link to the discussion on calibration errors the last time it came up.  I ran several tests, comparing the LiteVNA64 with Solver against my Agilent PNA (standalone).   
https://www.eevblog.com/forum/rf-microwave/experimenting-with-waveguides-using-the-litevna/msg4797743/#msg4797743

[Pinörkel]

I tend to move things around when using the equipment.  This means the cables can end up with a rotation force.  If the connectors are not tight and the cable rotates, the two center pins can have rotation.   For the cheap cables and connectors, maybe you don't care too much.  Maybe if you start using precision parts, you care a little more.  I posted some photos from another website showing some high grade HP connectors that were damaged this way. My advice is form good habits.

My measurement setups usually don't go for a walk. I read through most of the interesting stuff you linked on microwaves101.com. Now I see why some calibration gear connectors are so delicate. I think, on the cheap liteVNA calibration kit, the most critical factor why not to rotate the pin during insertion would be gold coating abrasion. The female center contacts there consist only of two simple halves. However, I was surprised about the good quality of the two SS405 50cm SMA cables that came with the device. I already tried to find where to get more of those, but had no luck finding the exact type. There are some that look really similar, but the connector quality is much better on the supplied ones, than on the ones I found online.

I have not used QT and don't run the LiteVNA stand alone.  That said, it's just some basic math and I would not expect there to be much difference if at all.  Of course, this assumes that in all three cases,  the errors due to the three calibrations are ZERO.   My guess is it is a user problem in not knowing all the fine details, some of which is documented in that large software thread.   I'll take a stab at it and say you have not changed Solver64's IFBW to match the LiteVNA's default.  I mentioned this in that thread as well as a few videos why I use the wider bandwidth and sacrifice noise for speed.  Then there is the isolation.

Bingo. I found no way of changing the IFBW in the liteVNA or the VNA View Qt software, and was not able to find documentation about the liteVNAs default setting, but setting the IFBW in Solver64 to 1k produced very similar results.
VNA View Qt result:


Solver64 result with 4k IFBW:


Solver64 result with 1k IFBW:

I've been told I have math errors before.  The last time was when using it with waveguides.  Again, it is certainly possible there are problems with it.  Actually, I know there are as some features I started to work on never panned out. But that basic software is almost two decades old and is used on a regular basis.  I've also frequently compared the results with a completely different VNA, cables, standards, software.  Even in this thread.   What you are describing would be a major problem.   

If you do feel there is a problem, you will need to provide enough details for me to reproduce it.

I did not intend to imply that there was something wrong in Solver64. I just had the impression that the signals were processed in another way before being displayed. This turned out to be true: I did not set the right IFBW values to get comparable results. Regarding repeatability, I think I am on a good way to finally gain some trust in my measurements.

The next step will be to decide what additional gear I need to make some useful cable measurements. At the moment, I have two main construction sites:
1. I have only cables of the exact same materials, regarding cable and connectors. I will need to build some different cables for making comparisons.
2. My current set of adapters is bad. To fix that, I will need to reduce the amount of adapters to an absolute minimum and find a calibration setup that eliminates almost everything except my BNC cable DUT from the setup. If I use the current calibration kit, I will have at least one unnecessary SMA-F to BNC-F adapter in my soup. To eliminate that, I could get a SMA-M to BNC-F pigtail for putting it between the liteVNA and the test cable. Calibrating that out would require me to build a BNC-M calibration kit, although I already found an article proving that BNC connectors make poor calibrations standards, it may perhaps be sufficient for my application. The alternative would be to use the liteVNA cal kit or some better SMA calibration standards, that have the disadvantage to force the inclusion of an additional adapter into my final measurement. The last component to get would be a good quality BNC-F 50Ω terminator to put at the end of the cable. The Fairview ST3B-F looks quite decent. Too bad I could not also use that as the calibration load without using an additional adapter.

[joeqsmith]

I think, on the cheap liteVNA calibration kit, the most critical factor why not to rotate the pin during insertion would be gold coating abrasion.

They may not be gold.  Anything of value would be very thin so expect a short life.   I check these cheap connectors under the microscope before using them.  One little defect and that rotation becomes more than a wear problem.  Again, good habits = longer life = repeatable results.   

You may be able to buy the OEM cables direct from them.     

I did not intend to imply that there was something wrong in Solver64. I just had the impression that the signals were processed in another way before being displayed. This turned out to be true: I did not set the right IFBW values to get comparable results. Regarding repeatability, I think I am on a good way to finally gain some trust in my measurements.

Solver performs all of the processing and expects raw data from the VNA.   For better or worse, it typically does what you program it to.   Because you are playing with at least three different ways to collect the same data, my advice is not to fall into the trap where you expect Solver to work like any other software.  That's caused a few people to have problems.     

***

I remember another user who had a similar problem:

https://www.eevblog.com/forum/rf-microwave/nanovna-custom-software/msg4067353/#msg4067353

Yes, that was the problem, Sweep must be turned on before you start the calibration.
It wasn't in the manual (at least not the parts that I read) and it is the opposite of what the Saver app requires, Saver even gives a warning that the Cal cannot start until you disable the continuous sweep.
I think that makes sense since the Cal procedure wants to take control of a few single sweeps, but obviously you can write code however you want, works just fine when you know how to set it up.

https://www.eevblog.com/forum/rf-microwave/nanovna-custom-software/msg5249997/#msg5249997

[pdenisowski]

I could get a SMA-M to BNC-F pigtail for putting it between the liteVNA and the test cable.

The quality of the pigtail (and especially the BNC part, as you mentioned) then becomes a potential issue:  I'm not sure how phase (or, frankly, amplitude) stable any commercially-available SMA-BNC pigtail would be.   This is where tape becomes handy

[joeqsmith]

Calibrating that out would require me to build a BNC-M calibration kit, although I already found an article proving that BNC connectors make poor calibrations standards, it may perhaps be sufficient for my application.

They are available along with a gage kit. 
https://www.maurymw.com/wp-content/uploads/2023/11/2Z-069.pdf

***
One option may be to replace the custom connector from the equipment with something more common.   

[joeqsmith]

LiteVNA64 (latest hardware) with supplied cable.  Cable fitted with SMA F/F Amphenol PN# 132169.    Cal standards applied to the end of the adapter.  Again, ideal model was used.  When testing the BNC adapters, the standard load (sorted Mini Circuits ANNE) was attached to an Amphenol PN# 24103 SMA F to BNC M.   This device is rated to 4GHz.  The IFBW set to 2k, sweep range 50k to 4GHz for all tests.    Three different combinations were measured.   From the photo, A+B,  A+C and A+D.   B is made up of two Kings adapters.  The BNC F/F is silver plated.   C&D are Amphenol adapters.

I would expect you to have similar results even with your cheap adapters.   

[Pinörkel]

At the moment, my current lousy set of stuff unfortunately only allows to compare two combinations of adapters this way: The liteVNA calibration load on the calibration set SMA F/F and an ultra cheap SMA-M to BNC-F with a fitted ancient BNC 50Ω termination of unknown quality. I need to get at least a few half decent adapters BNC-BNC, SMA-BNC and some 50Ω terminators. Unfortunately, Mini-Circuits seems to be kind of the only reasonably priced supplier that is not afraid to provide datasheets with electrical properties like return and insertion loss. However their BNC related stuff is mostly rated only up to 2GHz.

Setup


Measurement


btw.: If you ever come across wanting to visualize Solver64 data with LibreOffice you may find this monster useful, to convert the postfix letter style decade format into something, Libre Office can understand:

Code: [Select]

=IF(ISNUMBER(VALUE(RIGHT(A2)));A2;LEFT(A2;LEN(A2)-1)*10^(IFERROR(FIND(RIGHT(A2);"kMGTP")*3;IFERROR(FIND(RIGHT(A2);"munp")*-3;0))))

[joeqsmith]

I have two BNC 50 ohm terminators.  All of the ones I have inspected use an axial part.    Shown after SOLT, then attaching SMA / BNC adapter and BNC loads.   Similar results to what you measured with your setup. 

I imagine that the cables mating connector on the test equipment is also custom made to get the return loss they were looking for.   I assume also that you have no plans to swap it for a standard BNC or different type all together.  I wonder if mixing the parts will cause other issues.

[Pinörkel]

I have two BNC 50 ohm terminators.  All of the ones I have inspected use an axial part.    Shown after SOLT, then attaching SMA / BNC adapter and BNC loads.   Similar results to what you measured with your setup.

While your terminators and mine seem to behave extremely similar, the datasheets of the Mini-Circuits BTRM-50+ and the Fairwave ST3B-F seem to perform much better (I have drawn the curves into my measurement sheet for comparison).

I imagine that the cables mating connector on the test equipment is also custom made to get the return loss they were looking for. I assume also that you have no plans to swap it for a standard BNC or different type all together.  I wonder if mixing the parts will cause other issues.

There is indeed something special regrading the BNC female connectors on the critical equipment ports. They were made by Specialty Connector Co. and do not have a solder connection at the back. Instead, a miniature Tektronix Peltola coax connector can be fit directly into the back of the BNC female in order to have a fully shielded RF path with minimal impedance changes. However I do not know, if they also optimized the impedance value in some way. Those connectors can be found in a lot of Tektronix 500, 5000 and 7000 gear.

[joeqsmith]

I understand the problem with swapping it out now. 

You show a M and F terminator.  I assume you want a female.   This part is spec'ed to 4GHz with a return loss of 20.82.  So slightly better than the Mini-Circuits part at roughly 3X the cost. 
https://www.pasternack.com/images/ProductPDF/PE6256.pdf

The male version is also spec'ed to 4GHz with a return loss of 23.15. 
https://www.pasternack.com/images/ProductPDF/PE6TR008.pdf

Of course, there was the Maury Microwave kit.  Or if you just want a load:
https://www.fairviewmicrowave.com/content/dam/infinite-electronics/product-assets/fairview-microwave/product-datasheets/FMTR1064.pdf

***
Not seeing something like this being any value but it is inexpensive.
https://www.ebay.com/itm/155393728567

[Pinörkel]

Sorry, the purchase of a calibration kit costing several thousand euros is unfortunately off the table, as it is completely disproportionate to its potential future use. I did not know that there are pre-assembled china calibration kits like that. However, I already thought to buy some of these PCB BNC connectors, since they seem to have a comparatively well defined reference plane and could be a good base for experimenting with building an own BNC calibration kit.

Apart from that, I got disabled by a corona infection and I am simultaneously loosing my mind trying to get test equipment here in Germany. It is apparently near to impossible to buy specific stuff from Pasternak, Fairview Microwave and other suppliers here, because there are no distributors that deliver to Germany. Or, they do not sell to consumers, or only if you buy part counts larger than 100... In addition to that, German ham radio shops seem to be a concentrated accumulation of incompetence. I tried to buy some cabling and connector stuff: The first shop tried to sell me custom made H155 50Ω cables with 75Ω connectors, insisting that they were 50Ω. The second one sold me crimp connectors with missing center pins and shipped significantly less cable length than what I payed for, the third one packed the wrong cables and stuff I did not even order...    long story short: It will take much longer to get hold of a few test cables and adapters than I originally imagined.

At the same time I was able to play with the Solver64 software a little more. I stumbled onto some things, I am still trying to understand.
One thing is the step and impulse diagrams in the time domain. As far as I understand it, those should be convertible into each other by means of integration and differentiation. I had the impression that the step response had kind of a delayed integration reaction to the impulse response, as if there were a horizontal offset, but that may well just be my imagination, playing a prank on me.

Regarding calibration accuracy, I found some calibration kit constants for the liteVNA calibration kit on the web an tried to enter them into Solver64, but I could not really match the information with the Solver64 input fields for that. For the male load, I could e.g. not find input fields for R, L and C in Solver64.
https://groups.io/g/liteVNA/topic/calibration_kit_data_for/102476618
For checking calibration quality, I read that I could maybe have a look at how to do a T-check. For this I would most likely only need an SMA all female t-piece.

Also in the main tab of Solver64 I potentially stumbled upon a minor GUI bug (or maybe this is intentional). In the graphical views, there is the possibility to deactivate auto scaling for the X and Y axis via the context menus, which I find quite useful for looking at zoomed portions of the graphs. While this works perfectly in the time domain graph in the advanced tab, for the graphs in the main tab (like Impd Rectagular and Reflection Coeff), the context menu is missing the check marks in front of the auto scale entries. They also do not appear when switching auto scaling on and off for the vertical axis. At the same time, horizontal auto scaling is reset to auto for every sweep. So, for looking at a horizontally zoomed portion of the graph (e.g. 0 to 500M out of 0 to 4G), one needs to disable the sweep first, or all manually entered values will get sweeped away.

[joeqsmith]

I would have thought in Germany you would have access to all basic components like cable and connectors.   

For the TDR/TDT, I attempt to follow HP.   
https://www.testunlimited.com/pdf/an/5989-5723en.pdf

Solver 64 uses the same coefficients HP uses.  My attempts to make a T-check have not worked out.  You don't want a T with an added stub and connector.  I think I have documented some of that in the NanoVNA custom software thread.   

The main graph does allow you to disable the vertical autoscale.   I normally just set the sweep range for what I need.  So, locking out the horizontal manual scale was intentional but it could certainly be changed.

[Pinörkel]

I would have thought in Germany you would have access to all basic components like cable and connectors.

It usually is, IF you try to get stuff as a company or if you want to get jellybean stuff. Getting specific high quality stuff from outside the EU, especially from the US or UK can be very troublesome or be associated with ludicrous shipping costs, if you are just a lowly non-commercial customer. We have a really well sorted supplier for all kinds of of custom-made cabling and cable components here. It just does not sell to private customers.

For the TDR/TDT, I attempt to follow HP.   
https://www.testunlimited.com/pdf/an/5989-5723en.pdf

Interesting paper. The inconsistencies I see are most likely due to the tendency of my brain to reject math stuff without a good visualization. Nonetheless I have captured an image of what I mean:

To my understanding the white step response should be largest, where the gradient of the red pulse response is the highest, which in this picture is clearly not the case. Looking at the horizontal sampling point distance of the curves I have the impression that the red curve should be shifted right by one sampling interval which could be an off-by-one index error in the software or just my brain hallucinating bogus. ;-)

Solver 64 uses the same coefficients HP uses.  My attempts to make a T-check have not worked out.  You don't want a T with an added stub and connector.  I think I have documented some of that in the NanoVNA custom software thread.   

Ah, you mean the PCB-style T-check fixtures? I just read from someone that a T-check could be performed with really simple materials and you do not even need a high quality load. However, I do not know whether that is true.

The main graph does allow you to disable the vertical autoscale.   I normally just set the sweep range for what I need.  So, locking out the horizontal manual scale was intentional but it could certainly be changed.

Ah, ok. I will try that, but does this not require frequent re-calibration, because the sampling points are no longer at the same spots?

[joeqsmith]

Nonetheless I have captured an image of what I mean:

Attached was taken from the paper I previously suggested.

I had included a through termination on the cal boards but these can't be used for very high frequency.  The last one I made was built without any circuit board.  It performs better but still was not very useful.   

I keep things fairly well organized and typically reload previous calibrations.   The software does provide interpolation as well. 

[Pinörkel]

Attached was taken from the paper I previously suggested.

Yes, your picture shows exactly what I mean: there, the peaks of the pulse response are exactly at the horizontal positions where the slope of the step function is maximal/minimal. In the screenshot that I took, this does not seem to be the case.

[joeqsmith]

Attached was taken from the paper I previously suggested.

Yes, your picture shows exactly what I mean: there, the peaks of the pulse response are exactly at the horizontal positions where the slope of the step function is maximal/minimal. In the screenshot that I took, this does not seem to be the case.

Sadly, as I have stated previously the LiteVNA does not have enough BW for decent resolution.  Shown is looking at the Beatty standard Touchstone file included.