NanoVNA Custom Software

[joeqsmith]

Rather than the sensitivity of my hybrid to the impedances presented to its ports, it's the VNA's self reported port 2 input impedance of 53.5 ohms that's my concern. That's 7% from nominal. I would like the port 2 input impedance to be within a tenth or two of 50 ohms. People have also mentioned maybe the cables or even the thru are contributing to that, although I have yet to read about anybody claiming that the cables or thru that come with the nano VNA to be anything other than 50 Ohms.

I'm presently dealing with VHF centered around 146 MHz.

You have a standard you use for a load.  Guessing a bit of an unknown.  Then you have a port match which I am guessing is also a bit of an unknown.   I say guessing as I assume if you had access to better equipment and standards, you would have already narrowed down the source of errors.  Consider if they are each 5%, and both swing to the worse case your measurement is better than that.   

Consider that the load is what you cal to.  You are telling the VNA, that is 50 ohms at 146 MHz.  Let's say that the part is really 25 ohms.  The VNA will measure that as 50.  And if your port is now 25, it will also display 50.   Then there is the problem that you can't just measure the load with your DMM and expect to get anything meaningful.   

All that said, I use a load that I had sorted from Mini-Circuits for my standard.  I bought several of them and compared them with a set of metrology grade standards.   I picked the best two, the second best being what I use with these low cost VNAs.  I use the cheap supplied standards for the short and open.   Next, I use the ideal model which assumes there are no parasitics and the standards are perfect.   I know the VNA will require some warmup  time but lets ignore that.    Obviously, the errors are adding up already and I haven't even started.     

Next, I calibrate the LiteVNA from 140-150MHz using the above standards.  I did not torque anything.   Looking at "Loads.jpg",  I then measured the original supplied load (yellow), my home made load made up of 2X1206 resistors on FR4 (red), and finally reattached my Mini-Circuits load.   It's off by ab out 0.6%.   

Just as an FYI, I did measure the DCR of the Mini-Circuits at 50.042 ohms and supplied load at 51.029 ohms.  I'm a bit surprised to see them read about 1 ohm difference at 145MHz. 

Next, using the supplied cables, I measure port 2  "Port2.jpg".   It measures roughly 51.5 ohms or about 2.6%.    Now, think about it.  Had I used the supplied load as my reference standard rather than the  Mini-Circuits ANNE I sorted, the measurement would have been reported to be much tighter.  You could fool yourself into thinking it was better but without knowing what these errors are, you are just guessing. 

It's nothing that spending some cash couldn't solve.  Or, if you have access to some better equipment and standards, you could start to sort things out.     

This video had a lot of down votes, I am guessing from some very ignorant viewers.  To me, it clearly demonstrates the direct effect of using a poor load for a standard. 

[szoftveres]

This video is quite entertaining but its content is questionable; it doesn't demonstrate how a slightly bad calibration kit affects the calibration.
In the video, the presenter uses open as short, short as open, and 2dB attenuator as load. These are quite extreme and unrealistic examples for representing the effects of a bad calibration kit. Not sure how the down-votes are visible, but this could be the reason.

[tautech]

This video is quite entertaining but its content is questionable; it doesn't demonstrate how a slightly bad calibration kit affects the calibration.

Maybe you don't know who Mr Kirkby is ? 

IMO it was a very clear example why trusted Cal standards are needed and experimentation with self made vs Cal standards can produce similar results.

In the video, the presenter uses open as short, short as open, and 2dB attenuator as load. These are quite extreme and unrealistic examples for representing the effects of a bad calibration kit.

Extreme they are but the results are as just expected.....absolute crap !

[Fred_B]

Well, I'm already kind of on that. I made a some loads just for the hybrid, because I bought some cheap 2 Watt supposedly 50 Ohm RF SMA resistors for the job and then found they varied by over ohm. So for the hybrid, I got some male pcb thu-hole SMA connectors and soldered a couple of 100 Ohm 0.01% resistors on them. I made three. One of them came out bad. I'll have to take it apart and redo it. The other two test within a tenth of a ohm of each other and vary in reactance by just a little bit as far as I can tell.

I have parts to make a female SMA calibration set the same way. I just have yet to get around to doing it. I also have s parameter touchstone files for them, although they are on my computer that has a damaged video card. So, I have to wait to get a new video card before I can get to them.

Here's some pics of the loads I made for the hybrid.

***

Come to think of it, maybe that's why the supplied load with your device is 51 Ohms. 

Metrology is a wiley bugger.

[tautech]

Here's some pics of the loads I made for the hybrid.

Done exactly the same and they should be fine for LF stuff.

A buddy is right into VNA's and has some fine Cal standards that I checked my DIY's against then decided to splash out and get a half decent SMA set.
It's only money.......

[Fred_B]

My current project is in the VHF and UHF bands. I spent a good while simulating the loads before deciding to make them. They should do alright for this project. If I had money, I get a good machine and standards, although since I lack money such is pretty much unobtainium for the foreseeable future.

[joeqsmith]

This video is quite entertaining but its content is questionable; it doesn't demonstrate how a slightly bad calibration kit affects the calibration.
In the video, the presenter uses open as short, short as open, and 2dB attenuator as load. These are quite extreme and unrealistic examples for representing the effects of a bad calibration kit. Not sure how the down-votes are visible, but this could be the reason.

Hey, my waveguide calibration expert.  Good to see you posting again.  Hopefully I cleared up any misunderstanding you had.

Kirkby's demonstration shows the equipment isn't smart and does exactly what you told it to do.  It was very predictable.  Using such an extreme example as you put it, by telling the VNA a short is an open and an open is a short,  it's obvious what the outcome will be. 

I have posted data where I do not make any attempt to calibrate the VNA.  It is normally the last step I take, after the results are in the ball park (close).  Of course, that is a case of knowing your equipment.  I can get away without cal using my old Agilent and the LiteVNA in many cases.

Well, I'm already kind of on that. I made a some loads just for the hybrid, because I bought some cheap 2 Watt supposedly 50 Ohm RF SMA resistors for the job and then found they varied by over ohm. So for the hybrid, I got some male pcb thu-hole SMA connectors and soldered a couple of 100 Ohm 0.01% resistors on them. I made three. One of them came out bad. I'll have to take it apart and redo it. The other two test within a tenth of a ohm of each other and vary in reactance by just a little bit as far as I can tell.

I have parts to make a female SMA calibration set the same way. I just have yet to get around to doing it. I also have s parameter touchstone files for them, although they are on my computer that has a damaged video card. So, I have to wait to get a new video card before I can get to them.

...

Metrology is a wiley bugger.

When I first attempted to use my 1070's HP VNA above a few GHz, I made my own standards and was pretty pleased with the results.  Then I learned that my standards were garbage and all I really accomplished was wasting a lot of time.  Learn from my own ignorance! 

I purchased eight Mini Circuits ANNE terminators for about $10/ea.  One of them had really good return loss.  For $100, it was cheap.  I had the equipment to measure it and friends who loaned me the standards that I used to measure them.  The second best terminator used with the low cost VNAs was an added bonus.   

Datasheets for the ANNE show typical numbers.  That should have been obvious but I can tell you dealing with a few, it's not.  See the following:

https://www.eevblog.com/forum/projects/20db-rf-attenuator-seeking-feedback-to-improve/msg3093968/#msg3093968

There are several threads talking about the low cost VNAs and their standards.  Here are a few from the search:

https://www.eevblog.com/forum/testgear/nanovna-v24-sma-cal-load/
https://www.eevblog.com/forum/rf-microwave/nano-vna-reads-low-swr/


***
Forgot....  When I made the open and short for my Agilent, a local member had helped me characterize them.   The following shows their MatLab scripts I used.  The open was a pain to get right.  Took a few trips to the lathe to get it to where the script would converge. 

https://www.eevblog.com/forum/projects/20db-rf-attenuator-seeking-feedback-to-improve/msg2987982/#msg2987982

[szoftveres]

Hey, my waveguide calibration expert.

Hahaha I hope that I didn't stir up your emotions too much by pointing out something in your video, and this slightly sarcastic greeting is for something else.
Have a good day my friend! (Yeah, if you calibrate a vna backwards, it will show backwards values)

[joeqsmith]

I was hoping I didn't upset you by point out you were wrong.  Or, at least you never pointed out where I was wrong after I provided you with a step by step. 

[Fred_B]

I mulled the situation over a bit and decided to do a some testing and simulation.

I warmed up the VNA for about an hour. Then calibrated with the open, short and thru that came with the machine, although I used my load, which I am confident is very close to 50 ohms DC resistance.

Then, I attached the cable connected to port 1 to port 2 directly. This yielded a port 2 impedance of 47 Ohms and 1.59nH. Phase reported was very close to 0 degrees, meaning that the measurement impedance plane is well aligned with the end of the cable. At least I think it does.

Next, I disconnected the port 1 cable from port 2 and used the supplied thru and second cable to reconnect port 2, using both both cables and the thru supplied with the machine.  This yielded a port 2 impedance of 53.6 Ohms at -78.22 degrees. 

After this I built a spice model of the VNA and cable arrangement using the data from the VNA.  Since the first cable is effectively calibrated out of the results, I omitted the cable from the spice model.  I set up the model with a voltage source with 50 ohms impedance and a load with a 47 ohm resistance in series with a 1.59 nH inductor. The results of the run came out as expected, with a input impedance for the load of 47 ohms with a little inductive reactance.

For the second run included a 300 mm cable plus 12 mm additional for the length of the thru. I had to adjust the impedance of the cable a bit to achieve a good match with the results from the VNA. With a cable impedance of 50.04 Ohms the port 2 input impedance appears as a 53.6 Ohm impedance to port one of the VNA due to the impedance transformation effect of the cable, and phase agrees within 1 degree of the VNA result.

So, it would seem that the cables and the thru provided with the machine are of reasonable good quality. The 47 Ohm port 2 input impedance I believe is reasonable accurate. Now I just have to figure out a way to correct for that so that my hybrid sees 50 ohms at the end of the port 2 cable.   

[joeqsmith]

Hello Fred,

Good to hear you are continuing your efforts.

... which I am confident is very close to 50 ohms DC resistance. ...

There are a few recent posts about some through terminators and how poor their performance it.   I mentioned the one I had from Paternak used an axial TH device.   You really want to know how the parts behave at the frequency you are using them at, not at DC.   

It appears you are using the VNA standalone.  I really have no idea how their firmware works as far as the calibration.  If you had a way to characterize your standards, I am not sure they offer a way to enter them.   My software supports setting the coefficients which in the past has caused some problems.   The current release allows users to enter a value of 0 for coefficients and avoids a divide by zero (NAN) condition.     

When you calibrate at the end of the cable, I assume you are using the supplied F-F barrel.   Then you removed the barrel to attach the cable directly to Port2.   I have never checked but understand some of the supplied barrels are very poor.  It's the internet so for what it's worth.  Obviously you are also changing your reference plane when you remove it. 

*** See link for mention of barrels performance
https://groups.io/g/nanovna-users/topic/very_poor_thru_barrel/34447489?p=
***

You're measuring from 47 ohms to 53.6 ohms going from 1 cable to 2 for me seems unexpected.  Shown is the LiteVNA 64 with the same calibration as last time.  Again, this is with the supplied cables.   However, I have a decent thru that I leave attached to the Port1 cable.   

Yellow is the with the two cables in series.     

Red is with a single cable.   I inserted a M-M barrel to replace the second cable.  While you measure 6.6 ohms difference, I measure 0.2 ohms. 

Again, nothing was torqued, old calibration, no warmup, ideal models, poor standards.....   That said comparing the difference is a bit of a relative measurement, so let's make the same measurement without calibration.  I get about 0.3 ohms difference.  FYI, colors are swapped.

I'm sure if I took the time to make these measurements properly, things would tighten up.   From what I see, if you really require better than 10%, you need to sort out the cause of the main contributors. 

***   
Also, if you like, I could measure the LiteVNA's Port2 using my Agilent to compare results.  From previous time I have compared results for passive circuits, they have been fairly close.

[Fred_B]

Hey Joe,

I bought a supposedly better F-F barrel that is a little longer, nickel plated and with a flat for a wrench in the middle. In the course of my testing I swapped the one that came with the kit and the one I bought later. Their performance was very similar. So, either they are both similarly bad or similarly good. I tend to think they're on the good side, because of the SPICE simulation results. In my simulation of the loads I made I found they behave pretty much as DC loads up to about 3-4 GHz when they start becoming somewhat reactive.

The SPICE simulation show the impedance transformation of the transmission line that the coax forms. The transmission line is what ever wavelength 312mm is in a cable with a velocity factor of 0.695 c, which is the figure I found for RG-316. The conversion of the 47 Ohm load to 53.6 Ohm input impedance of the 50.04 Ohm transmission line model is exactly what is expected by transmission line theory. The SPICE result verifies that. I've modeled a bunch of quarter wave transmission line impedance transformers for Wilkinson power splitters and impedance matching harnesses for antennas. I know transmission line theory well enough to work with it, although less well enough to explain it all. 

As far as your results go, I make sense little of the first one, with the ~43.6 Ohm impedances. Is that the uncalibrated one?

The second results kind of makes sense, if the port 2 input impedance is significantly closer to 50 Ohms than the port 2 on my machine.

Because I am working on a new computer I have yet to set up your software on it. I installed VNA-QT, although it displays a discontinuity in the sweep results that I have yet to sort out. I like the VNA-QT for the one reason that it allows me to load s parameter touchstone files for the calibration loads. I can produce them from my simulation software with which I designed and modeled my loads. Converting the S parameter touchstone files to coefficients numbers I have yet to figure out.

Cheers,

Fred

[KedasProbe]

Anyone else having a problem to install the 64bit version of NI-VISA (ni-visa_20.0.0_offline.iso)

32bit and 64bit are present in one install (unfortunately) and it always forces a 32bit install, I tried on 2 windows 10 64bit and one Windows 7 64bit in all cases it want's to install a 32bit version. Any idea how to force a 64bit install?

(The 64bit labview runtime installed without problem tough, and does work with solver64)

Edit: it turns out that it works without installing NI-VISA, only ni-labview-2020-runtime-engine_20.1.1_offline.iso setup on Windows 10 64bit. (LiteVNA64 HW64-0 3.1 FW1.3.07, Build Time Sep17 2023)

[joeqsmith]

Fred,

You should post more details about your SPICE simulations.  I would take the time to follow along.   

I bought a supposedly better F-F barrel that is a little longer, nickel plated and with a flat for a wrench in the middle. In the course of my testing I swapped the one that came with the kit and the one I bought later. Their performance was very similar.

Its possible that you have so many other problems, they are masking what the barrels contribute to the error.   

As far as your results go, I make sense little of the first one, with the ~43.6 Ohm impedances. Is that the uncalibrated one?

Yes, the file labeled "1and2cables_uncal.jpg" is with no attempt to calibrate the LiteVNA.  Of course the absolute values are off a bit but the relative values are very close as we would expect.   

The second results kind of makes sense, if the port 2 input impedance is significantly closer to 50 Ohms than the port 2 on my machine.

Hard to say but we shouldn't have the amount of error you are seeing.  Something is wrong. 

Because I am working on a new computer I have yet to set up your software on it. I installed VNA-QT, although it displays a discontinuity in the sweep results that I have yet to sort out. I like the VNA-QT for the one reason that it allows me to load s parameter touchstone files for the calibration loads. I can produce them from my simulation software with which I designed and modeled my loads. Converting the S parameter touchstone files to coefficients numbers I have yet to figure out.

Some of the last changes I made to my software were to work around problems when importing data collected with the QT software.  I believe we were posting about that problem in this thread.

I've thought about adding support for databased standards but have not looked into it enough to know if all the vendors follow a standard.  The standards I had access to were from Agilent using the polynomial fit.  For the low cost VNAs, I pretty much use the ideal model.  The software supports my other VNAs.  It's these other VNAs that drove the ability to change the coefficients.  It's in the common code, so all of the software goes along for the ride.     

[Fred_B]

You should post more details about your SPICE simulations.  I would take the time to follow along.   

Attached is the LTSpice model I made. The model itself is fairly simple, although it implicitly deals with Transmission Line Theory, which is a whole other matter. There's a 50 Ohm source that feeds a transmission line model of the 300mm RG-316 cable that came with my VNA. I added 10 mm for the length of the thru. I used the 0.695 velocity factor of RG-316 plus the lengths of the cable and thru to determine the time delay for the combination. For the load I used the figures from the VNA results of 47 Ohms resistance and 1.59 nH of reactance, which works out to be about 1.46 Ohms and is basically negligible in this case. You'll need an understanding of transmission line theory to get why the cable transforms the 47 Ohm load into a 53.6 input impedance at the other end of the cable. That's all I can say about that, except that the SPICE model verifies the transformation rather exactly and confirms the cable is very close to 50 Ohms impedance.

Some of the last changes I made to my software were to work around problems when importing data collected with the QT software.  I believe we were posting about that problem in this thread.

I've thought about adding support for databased standards but have not looked into it enough to know if all the vendors follow a standard.  The standards I had access to were from Agilent using the polynomial fit.  For the low cost VNAs, I pretty much use the ideal model.  The software supports my other VNAs.  It's these other VNAs that drove the ability to change the coefficients.  It's in the common code, so all of the software goes along for the ride.     

Yes, we were, although this is different. I import into VNA-QT the s parameter touchstone files I exported from the EM simulation software I used to design my loads.  I have yet to build the calibration set I designed in physical form, so I modeled the standards that came with the VNA together with the thru barrel, and exported those touchstone first. That resulted in the very good noise floor measurement I got for the nano, which you may recall and which approached what 0W0 claimed the nano would do.

The touchstone files have a standard. It's rather simple. Links for the standards are on this page.

https://en.wikipedia.org/wiki/Touchstone_file

[joeqsmith]

I went back and read all your posts for context.  I didn't realize you were the person that I was working with using QT and exporting Touchstone files with zero'ed data.  Agree, the format is simple enough.  As far as supporting the database, it's more a question on what the companies supply.  I think I would start by asking the major manufactures for examples to get a feel for how similar they are. 

[Fred_B]

That a good point. What would be convenient for you as a coder as far as the touchstone files go? Would it be good to have the touchstone files exported with frequency entries that match the frequency span and number of points in the calibration files produced by Solver? I can change those details before exporting to match. I lack any concern at the moment as far as the database goes, because for now I am just making my own calibration sets and producing the related touchstone files.

Anyway, I got Solver running, did a cal, and repeated the measurements. The results came out the same. Attached are screen shots of the whole app so you can see the settings.

[joeqsmith]

As far as the database standards,  I'm not sure it would matter as long as the companies all do the same thing.   Most of the code is pretty simple.   I spend more time doing research in how functions should work than I do coding.

I think if you watched my series on rectangular waveguides where I talk about how you can't use an open for a standard, and instead how an offset short is used to simulate the open by shifting the phase,   you catch on real quick what is going on with the transmission line.  I am just surprised your VNA would be that bad.  So, I pulled out my V2Plus4 purchased directly from OWO.  We just want to know what the effects of the VNA are, so using the same cables, standards and procedure, I repeated your test using 1 & 2 cables.  What do you know, it is about as bad as what you are seeing with the Professional one.   

Well, at least it gives us more confidence in your measurement.

[joeqsmith]

I was going to attempt a sanity check with the PNA for you but ended up learning something new instead about these low cost VNAs...

Let the PNA warmup and ran a SOL using my poor mans cal standards (female) that I had characterized against the Agilent standards.   I am using the best load of the eight ANNE terminators I had purchased.  This load is attached to a good barrel.   

I then checked the OEM load that was supplied with the low cost VNA using another good barrel.   Its a little off from what I read with the Lite.

[joeqsmith]

I then connected directly to port 2 of the LiteVNA and measured 57 ohms.   Brain kicks in and says, idiot, power that thing up.   Here lies the rub.  We can't simply measure it as the are doing some switching.  Thinking it may be an artifact from sweeping,  I tried to select the CW mode from their built-in firmware but it has no effect.    I repeated this with the V2Plus4 with the same results.

Attached plots show the effect for both low cost VNAs.  What is interesting is how the LiteVNA changes about an ohm as it chops,  compared with the V2Plus4 at about 2 ohms.

Something else to be aware of. 

[FPSychotic]

Sorry by ask , I cannot find the changes at the moment, but I'm trying to update the software and cannot find the github nevermore.
Also found a cloud folder ,on which it looks the last version is solver64, but it was said was abandoned, so I'm really confused.
I have working solver32 and solver64 in windows 11, can say both work and coexist, but no idea what I should to use, what is still active and if the litevna64 firmware has any importance.
Thanks and sorry I read some pages but couldn't find the changes

[gf]

We can't simply measure it as the are doing some switching.  Thinking it may be an artifact from sweeping,  I tried to select the CW mode from their built-in firmware but it has no effect.    I repeated this with the V2Plus4 with the same results.

Attached plots show the effect for both low cost VNAs.  What is interesting is how the LiteVNA changes about an ohm as it chops,  compared with the V2Plus4 at about 2 ohms.

They have only a single receiver/mixer, but they need to measure 3 quantities for each frequency point. So the 3 readings need to be multiplexed. The source currently being measured is connected to the receiver, and the other sources are connected to a terminator. Then the receiver is switched to the next source... OTOH, the original NanaoVNA has 3 receivers/mixers. Maybe it's the better choice for 150MHz anyway?

[joeqsmith]

I mentioned CW not having any effect, but you can also disable the sweeps in the firmware and it will still switch.   

[Fred_B]

Attached plots show the effect for both low cost VNAs.  What is interesting is how the LiteVNA changes about an ohm as it chops,  compared with the V2Plus4 at about 2 ohms.

Something else to be aware of.

You've brought up something there that I am actively wondering about, the switching. I had a look at the schematic for the V2Plus4. Apparently it switches back and forth between measuring S11 and S21 for each frequency point. If I comprehend your post correctly, you're measuring the low cost VNA, with a quality PNA. I fail to make sense of what your are showing, a switching that is apparently every few MHz or so across the frequency sweep, and we fail to see that when self-measuring the port 2 input impedance of the low cost units. That's too slow for the S11/S21 switching. So, what do you think that is?

Just for giggles, could you use the PNA to measure the output impedance of port 1 on the low cost devices?

A couple more notes about the SPICE model. The original calibration cable attached to port 1 has been omitted from the model because for all practical purposes it has been calibrated out of the measurement. Also, the magnitude of signal at port 2 in the model is reported as close to -6 dB, that's because LTSpice normalizes the signal level at the internal input to the voltage source, which in this case has been set to 1V, before the source output impedance is applied to the signal, rather than to the external output of the voltage source after the source impedance has been applied to the signal. That means in VNA terms the signal level at port 2 is -0 dB relative to that level at port 1, which can be checked by looking at the signal level at node P1 in the model. Basically, V(P1) = V(P2) meaning there's zero loss in the "lossless" transmission line model.

[joeqsmith]

Apparently it switches back and forth between measuring S11 and S21 for each frequency point.

Three measurements.  Port1 incident and reflected along with Port2.  I did not try to disable the S21 measurement as I didn't see it in the firmware.   

If I comprehend your post correctly, you're measuring the low cost VNA, with a quality PNA.

Yes

I fail to make sense of what your are showing, a switching that is apparently every few MHz or so across the frequency sweep, and we fail to see that when self-measuring the port 2 input impedance of the low cost units. That's too slow for the S11/S21 switching. So, what do you think that is?

It's the asynchronous switching of the low cost VNAs, which I can't seem to disable from firmware. 

Just for giggles, could you use the PNA to measure the output impedance of port 1 on the low cost devices?

It's a 2-port, 1 path VNA.  Port 2 is an input only.  Port 1 will drive the output.   So no.