NanoVNA V2+4 SMA Cal load

[virtualparticles]

The uncertainty of reflection measurement is entirely dependent on the return loss of the load standard.

It is maybe useful to add that this is usually true for good-quality, characterized calibration kits, as well as generally for low-reflection DUTs. For some of the improvised/home-made calibration kits discussed in relation with the NanoVNA, the residual source match error (relevant with more reflective DUTs) is likely to be set by the phase error on the open/short, as well the return loss of the load.

Excellent point

[Mechatrommer]

I missed you were talking about the OEMs software.   I thought they had a load and save cal was all.  It's a pain to run on my PC.  I'll fire up the old laptop and have a closer look at what this database is that you referred to...
You may find the Keysight software here:
http://na.support.keysight.com/pna/apps/applications.html

you need to have your cal kit's database ready though. ie their profile plot in s1p files from very well calibrated VNA and Cal Standard. supposedly these profile will be used as transfer standard (if i term it correctly) so your measurement on other VNA calibrated with data-based CAL kit (lets call it secondary or tertiary transfer standard) will be similar to what is measured by the source VNA that generated the data-based files in the first place. if you have say Agilent Cal Kit with their s1p profiles, you may play with that NanoVNA-QT SW. ymmv cheers.

I learned that lesson with the V2+, you need to account for leakage (crosstalk) terms or it falls apart above 3GHz.

thats why the good thing that it spits raw data back to PC, so at least we can do/study something about it in PC and math... should be a good learning process given we have time for that. afaik my Deepace only saved out processed data, so i'm not sure how to cancel back its internal systematic error/leakage. maybe later i can play further with it by disabling calibration data altogether so maybe i can get raw measurement data. fwiw.

[joeqsmith]

I don't provide private emails details on a public forum, that's just irresponsible and violates the other parties trust. Sure I could issue a request, but why don't you ask yourself if you question my explanation??

Yes, I question your explanation as to why the supplied load resistor would be 51 ohms.  There are no facts associated with it.  I have no details from the source and you are reluctant to supply it.  My guess is they found some 51 ohm parts with a better return loss over the working range of the Nano than 50 ohm parts.    I find it far more believable the return loss is better because of the quality of the part they found rather than its DC resistance. 

Why don't you contact the OEM and quit wasting my and everyone else's time. I've provided the info on the V2+ OEM above, they are source as mentioned, it seems you are reluctant to contact them??

I doubt this was a fundamental design choice, why would you want to use 51 ohms instead of 50 ohms?? More likely a compensation means to improve the performance of an already pretty good product, maybe they had a higher than expected RL in the upper frequency ranges. That's the reason I asked why the 51 ohms, I initially thought it was an out of spec 50 ohm load, and got the response because it makes the higher frequency RL better. After finding you and Doug also have 51 ohm loads with the V2+, then this makes sense as it is intentional and is good enough answer AFAIC.

Right, lots of maybes.  Little facts.   This is why I question your comments.

Again why are you reluctant to contact the OEM yourself, maybe do a little research on your own rather than ask others to do it for you!!

You seem to have an interest in SWR, or at least mention it a lot, down to three places.   Is it really this big of a deal for you to try and measure SWR this many places out?  If so, why? 
It's just easier to calculate in your head, simple as that.

I find swagging to 1 place beyond the decimal easier than sorting out 3 places.   SWR to 3 places out makes no sense to me.

 Well if you are into "swagging" measurements then this wouldn't make sense to you, even claiming to not worry about doing a short cal, so one decimal place is probably fine!!   

Agree, generally this 49-51 range is fine but I wanted to understand the root reason for the 51 ohms, was this a sloppy load or was it intentional. Now I know the answer. Leaving things "out in the bush" has a habit of allowing Murphy in at some point, and he will bite and bite hard, I know I have "teeth" marks to prove it 

So having a cal load of 51 ohms could have some impact on precision measurements at some time, and it's better to know the reasoning behind the decision rather than sweep in under the rug and open one's self up for a possible issue later. Sure I wouldn't rely on the V2+ for any precision work, but might use it to verify another measurement from a precision instrument as a sanity check. ...

True, it's good to understand any tools you use.  But we still do not know why the 51 ohm was chosen.   I doubt is has anything at all to do with the V2+'s hardware but just the quality of the parts they found.

Not we, its seems like you don't know, so actually more speculation on your part, maybe it's time to contact the OEM so you can enlighten yourself!!

I was thinking about that short.  So I made up an end cap and soldered to the back side.  Fits a standard 5/16" wrench.
 
Like this??
[ Attachment Invalid Or Does Not Exist ]

Notes above in purple.

Best,

"The 51 ohm load just happaned to be the best load we can find in this cost range, and it is not specific to the V2 or V2Plus4."


A good book for those who haven't read it. 
https://www.amazon.com/Death-Expertise-Campaign-Established-Knowledge/dp/0190469412

[switchabl]

I think the answer is probably simple and maybe slightly disappointing: if you take it apart/cut it open, you will likely find a single resistor (hopefully SMT) inside. 51 ohm is a E24 value and widely available. 50 ohm RF resistors exist but are more expensive.

Btw if you open up a quality 50 ohm load rated for higher frequencies, you will instead find a tantalum nitride (TaN) or similar thin-film, possibly laser-trimmed, deposited on an RF substrate.

[joeqsmith]

I had been following this thread which was an interesting read: 

https://www.eevblog.com/forum/testgear/kirkby-calibration-kit-alternatives/msg1599934/#msg1599934

I missed you were talking about the OEMs software.   I thought they had a load and save cal was all.  It's a pain to run on my PC.  I'll fire up the old laptop and have a closer look at what this database is that you referred to...
You may find the Keysight software here:
http://na.support.keysight.com/pna/apps/applications.html

you need to have your cal kit's database ready though. ie their profile plot in s1p files from very well calibrated VNA and Cal Standard. supposedly these profile will be used as transfer standard (if i term it correctly) so your measurement on other VNA calibrated with data-based CAL kit (lets call it secondary or tertiary transfer standard) will be similar to what is measured by the source VNA that generated the data-based files in the first place. if you have say Agilent Cal Kit with their s1p profiles, you may play with that NanoVNA-QT SW. ymmv cheers.

I had seen that it had support for Touchstone format.  Is this what you are calling a database?   I will do a search and see if I can find some cal files.  Anything I have done is based on that Agilent format. 

I learned that lesson with the V2+, you need to account for leakage (crosstalk) terms or it falls apart above 3GHz.

thats why the good thing that it spits raw data back to PC, so at least we can do/study something about it in PC and math... should be a good learning process given we have time for that. afaik my Deepace only saved out processed data, so i'm not sure how to cancel back its internal systematic error/leakage. maybe later i can play further with it by disabling calibration data altogether so maybe i can get raw measurement data. fwiw.

I like the original Nano in that I can view and use the touch screen while it's running.   Normally, I just clear the cal and pull the data.   I heard they had some fudge factors coded into it but it hasn't caused me any problems.   Added support for interpretation but need to test it. 


Occam's razor strikes again.  Not sure why that was even a question as it seemed obvious.  Oh well.     

[mawyatt]

I think the answer is probably simple and maybe slightly disappointing: if you take it apart/cut it open, you will likely find a single resistor (hopefully SMT) inside. 51 ohm is a E24 value and widely available. 50 ohm RF resistors exist but are more expensive.

Btw if you open up a quality 50 ohm load rated for higher frequencies, you will instead find a tantalum nitride (TaN) or similar thin-film, possibly laser-trimmed, deposited on an RF substrate.

Yes 51 and 47 are the standard 5% values round 50, it may even be a leaded resistor since the back has a tall "cap" that covers whatever resistive element they are using

1% 49.9 ohm SMD resistors are not expensive, but they chose 51 instead. The 3 independent DCR measurements show this is 51 ohms spot on, so likely they are not using a 5% resistor.

I would expect a quality thin film resistor deposited on an alumina substrate in a quality load, but not at this price and performance level. Would also expect this to have a DCR of very close to 50 ohms.

I'll have known good references after the holidays when I go to university lab, then I'll be able to verify IF the 51 ohms actually does improve the HF RL on the V2+4 like the OEM suggests.

Best,

[Mechatrommer]

@joe yes afaik from literature reading, the data-based means using measured data of call kit s11 for calibration. I cant think of anything other than the s1p file... back then, vnas were using model based ie their C polynomials coefficients because digital storage was a premium. Now storing thousands of measured data points is cheap...any modern vna should be able to do it. Its more real representation. Polynomials implementation today is only following the legacy method.

@mawyatt, 2x 100ohm smd is cheap

[Hydron]

Mine was also 50.9something, measured using adaptors relative to the short as a zero reference.

[switchabl]

I mean, in principle it is possible to use the parasitics as a matching network so that 51 ohms ends up close to 50+0j ohms at high frequency. But even if you managed to pull that off, it would be quite sensitive, so you might well end up having to fine-tune every load by hand. At 3GHz just building something that has lower parasitics in the first place really looks like a much simpler option.

But, it turns out I actually measured the supplied load when I experimented with the NanoVNA v2 (without +) a few months ago. So let's look at some data. The reference was my working load standard, which was selected from a bag of Huber&Suhner 18GHz SMA loads. So not ideal, but should be good enough to get a qualitative picture (incidentally, that one measures almost exactly 50.00ohms DCR on my 34401A).

There is always a chance they changed suppliers in the mean-time, so yours may differ. In any case, at low frequencies, it is 50.8 Ohms, so seems to be 51 Ohm as well. Overall performance is however dominated by the reactive part. To get an idea if the 51 Ohm are actually beneficial, I fit a model consisting of a piece of transmission line, terminated by R || C. Now we can compare to, say, R=49.9 Ohm: the real part of the impedance is indeed closer to 50 Ohm at higher frequencies, but that is not really helpful as the return loss is still worse. Needless to say, the low frequency performance is degraded with 51 Ohm.

[mawyatt]

@mawyatt, 2x 100ohm smd is cheap

Yes I know, probably not even $0.01 each for 1% SMD resistors. Recall someone did some home-brew loads and found that two 100 ohm SMD performed the best overall with 1, 2 or 3 parallel resistors in these moderate frequency ranges.

Best,

[joeqsmith]

@joe yes afaik from literature reading, the data-based means using measured data of call kit s11 for calibration. I cant think of anything other than the s1p file... back then, vnas were using model based ie their C polynomials coefficients because digital storage was a premium. Now storing thousands of measured data points is cheap...any modern vna should be able to do it. Its more real representation. Polynomials implementation today is only following the legacy method.

@mawyatt, 2x 100ohm smd is cheap

I was thinking something else.  Except for the QT software you mention, I don't have anything that supports it.    The real benefit for me would be the education from implementing software for it.   

My first VNA has no computer.  Thumb wheel switches, some overlays and a grease pen.   I used an ideal model when I wrote the software for it which was a huge improvement and good enough for the experiments I came up with. 

After playing with the V2+ /4 for the last several weeks in the 1-4GHz range, even with an ideal model I'm impressed.  For $120, it's a great learning tool.       

[mawyatt]

I mean, in principle it is possible to use the parasitics as a matching network so that 51 ohms ends up close to 50+0j ohms at high frequency. But even if you managed to pull that off, it would be quite sensitive, so you might well end up having to fine-tune every load by hand. At 3GHz just building something that has lower parasitics in the first place really looks like a much simpler option.

But, it turns out I actually measured the supplied load when I experimented with the NanoVNA v2 (without +) a few months ago. So let's look at some data. The reference was my working load standard, which was selected from a bag of Huber&Suhner 18GHz SMA loads. So not ideal, but should be good enough to get a qualitative picture (incidentally, that one measures almost exactly 50.00ohms DCR on my 34401A).

There is always a chance they changed suppliers in the mean-time, so yours may differ. In any case, at low frequencies, it is 50.8 Ohms, so seems to be 51 Ohm as well. Overall performance is however dominated by the reactive part. To get an idea if the 51 Ohm are actually beneficial, I fit a model consisting of a piece of transmission line, terminated by R || C. Now we can compare to, say, R=49.9 Ohm: the real part of the impedance is indeed closer to 50 Ohm at higher frequencies, but that is not really helpful as the return loss is still worse. Needless to say, the low frequency performance is degraded with 51 Ohm.

Using the parasitics as a means to achieve a certain goal seems like a recipe for disaster IMO, especially if these parasitics are what measurements are based from since parasitics aren't generally well controlled!

Nice plots that show the VNA readings monotonically move away from the DCR value as parasitics come into play as the frequency increases and at very low frequencies look like the DCR value.

As a quick experiment I just calibrated with the V2+4 supplied 51 ohm load (DCR 50.931 ohms). Then measured SMA load (DCR 49.687), the VNA reports 48.7 ohms @ 1MHz & 10MHz. This VNA reading agrees with simple Ratio (50/50.931)*49.687= 48.78 ohms and VNA reading*(50.931/50) = 49.61 close to the DCR value. Did a few more values below.

DCR        NanoVNA 51 ohm cal {Ratio 50.931/50*Reading}     NanoVNA reading & {Ratio (50/50.931)*DCR}               

49.687                  48.7         {49.61}                                             48.7     {48.78}                                                 
51.395                  50.4         {51.34}                                             50.4     {50.46}
50.882                  49.8         {50.73}                                             49.8     {49.95}
50.182                  49.2         {50.12}                                             49.2     {49.26}
49.643                  48.6         {49.50}                                             48.6     {48.74}                                               
 


So it seems that using a load cal of something closer to an ideal 50 ohms for LF work might improve the measurements around 50 ohms, or at least provide a "hint" at what the DUT might really be.

Of course the best solution is just use a quality 50 load and be done with it and not have to worry about all this!!

Best,

[joeqsmith]

@joe yes afaik from literature reading, the data-based means using measured data of call kit s11 for calibration. I cant think of anything other than the s1p file... back then, vnas were using model based ie their C polynomials coefficients because digital storage was a premium. Now storing thousands of measured data points is cheap...any modern vna should be able to do it. Its more real representation. Polynomials implementation today is only following the legacy method.

@mawyatt, 2x 100ohm smd is cheap

I was thinking something else.  Except for the QT software you mention, I don't have anything that supports it.    The real benefit for me would be the education from implementing software for it.   

My first VNA has no computer.  Thumb wheel switches, some overlays and a grease pen.   I used an ideal model when I wrote the software for it which was a huge improvement and good enough for the experiments I came up with. 

After playing with the V2+ /4 for the last several weeks in the 1-4GHz range, even with an ideal model I'm impressed.  For $120, it's a great learning tool.     

Digging into it a bit I see both the Cooper Mountain and Keysight manuals refer to it as a database.   It looks some of Keysights PNAs support it.  Mine is too old.   

https://coppermountaintech.com/wp-content/uploads/2018/05/Using-a-Databased-SOLT-Calibration-Kit.pdf
https://www.rosenberger.com/fileadmin/content/headquarter/News/Application_Notes/AN002.pdf

I didn't find many details about it.   I would guess that there isn't a standard and companies may supply different resolution files.  I would guess then that the software has to interpolate the data for the range you want to use.   Maybe one kit has data every 10MHz where  the other is ever KHz.   If you find any detailed documents on it, post a link and I will have a look.   

For the fun of it, I have added support for both interpolation and offset loss.  I doubt it would make any difference but in case you want to attach your $15,000 standards and cables to your Nano, you have the option to enter all the coefficients.   

[mawyatt]

@ joeqsmith

It seems the latest NanoVNA-H4 version is also using a different cal load that is above 50 ohms (50.78). Here's what nano-users Hugin (OEM for the H4) says. Same answer I got from V2 Plus4 OEM HCXQS.

https://groups.io/g/nanovna-users/topic/nanovna_h4_calibration_loads/78920711?p=,,,100,0,0,0::recentpostdate%2Fsticky,,,100,2,0,78920711

From Hugin,
The load attached to the H4 has been re-matched to perform better at higher frequencies. The 49.9 ohm load used earlier performed quite badly above 2GHz, but the new and improved load can get S11 below -30dB at 6GHz. For measurements that require higher frequencies, the new load better. For UHF measurement, I suggest you use the load included with H4 for calibration.

BTW the H4 (50.78 ohms) load pictured in the link above does not look like the load supplied with the V2+4 (50.93 ohms) we have, they are not the same SMA loads.

The V2+4 loads are supplied from,
https://item.taobao.com/item.htm?id=607518859769

So it seems the NanoVNA OEM Hugin is recommending the same approach as we hinted, use a cal load closer to 50 ohms for lower frequency work and the supplied higher than 50 ohms cal load for higher frequency work.

Best,

Edit: Very interesting reference "About the Extra Ohm in 50 Ohm Calibration Loads" from Avian's Blog. Shows the  ~51 ohms cal load selection is to offset the lower HF cal impedance and meet an overall better worst case RL over the entire frequency span, but sacrifices the lower frequency RL.

https://www.tablix.org/~avian/blog/

[joeqsmith]

Again, a good book for those who haven't read it.

https://www.amazon.com/Death-Expertise-Campaign-Established-Knowledge/dp/0190469412

[switchabl]

From Hugin,
The load attached to the H4 has been re-matched to perform better at higher frequencies. The 49.9 ohm load used earlier performed quite badly above 2GHz, but the new and improved load can get S11 below -30dB at 6GHz. For measurements that require higher frequencies, the new load better. For UHF measurement, I suggest you use the load included with H4 for calibration.

A likely explanation is that they found one from a different manufacturer with better RF performance and that one just happens to be 51 ohms. Maybe they already had a lot of those or they found a cheap resistor model that works better at higher frequencies but is not available as 49.9/50 ohms (or would cost a few cents more). Hard to know for sure.

I would very much assume that if they could get the exact same construction but with a 50 ohm resistor, performance would not only be better at low frequencies, but at the very least no worse at higher frequencies.

Edit: Very interesting reference "About the Extra Ohm in 50 Ohm Calibration Loads" from Avian's Blog. Shows the  ~51 ohms cal load selection is to offset the lower HF cal impedance and meet an overall better worst case RL over the entire frequency span, but sacrifices the lower frequency RL.

https://www.tablix.org/~avian/blog/

That makes a lot of (questionable) assumptions. If you look at my data, you can also see that 51 ohms results in the real part of the impedance shifted closer to 50 ohms. What it doesn't do is improve the return loss since that is totally dominated by imaginary part. Again, you could imagine the parasitic C and L forming an L-match to actually get (close) to 50 + 0j ohms, but I don't expect that to be practical and would be very surprised if they could pull it off consistently. It certainly doesn't work that way for the one I have.

So it seems the NanoVNA OEM Hugin is recommending the same approach as we hinted, use a cal load closer to 50 ohms for lower frequency work and the supplied higher than 50 ohms cal load for higher frequency work.

There is the concept of a low-band load, that has very, very good return loss but only at low frequencies (say <2 GHz) and may be supplemented e.g. with a sliding load that only works for higher frequencies. But in this case you would rather look for something that is better across the whole range of interest. I mean it's not like the included one is particularly good at higher frequencies either (at least mine is not).

Again, a good book for those who haven't read it.

https://www.amazon.com/Death-Expertise-Campaign-Established-Knowledge/dp/0190469412

I am not entirely sure what you are trying to say. If it is "don't believe something just because a chinese seller or a random blogger says so", I concur.

On the more immediate subject matter, the "Handbook of Microwave Component Measurements" by Joel Dunsmore is something I have found quite helpful.

[virtualparticles]

@joe yes afaik from literature reading, the data-based means using measured data of call kit s11 for calibration. I cant think of anything other than the s1p file... back then, vnas were using model based ie their C polynomials coefficients because digital storage was a premium. Now storing thousands of measured data points is cheap...any modern vna should be able to do it. Its more real representation. Polynomials implementation today is only following the legacy method.

@mawyatt, 2x 100ohm smd is cheap

I was thinking something else.  Except for the QT software you mention, I don't have anything that supports it.    The real benefit for me would be the education from implementing software for it.   

My first VNA has no computer.  Thumb wheel switches, some overlays and a grease pen.   I used an ideal model when I wrote the software for it which was a huge improvement and good enough for the experiments I came up with. 

After playing with the V2+ /4 for the last several weeks in the 1-4GHz range, even with an ideal model I'm impressed.  For $120, it's a great learning tool.     

Digging into it a bit I see both the Cooper Mountain and Keysight manuals refer to it as a database.   It looks some of Keysights PNAs support it.  Mine is too old.   

https://coppermountaintech.com/wp-content/uploads/2018/05/Using-a-Databased-SOLT-Calibration-Kit.pdf
https://www.rosenberger.com/fileadmin/content/headquarter/News/Application_Notes/AN002.pdf

I didn't find many details about it.   I would guess that there isn't a standard and companies may supply different resolution files.  I would guess then that the software has to interpolate the data for the range you want to use.   Maybe one kit has data every 10MHz where  the other is ever KHz.   If you find any detailed documents on it, post a link and I will have a look.   

For the fun of it, I have added support for both interpolation and offset loss.  I doubt it would make any difference but in case you want to attach your $15,000 standards and cables to your Nano, you have the option to enter all the coefficients.   

To add a "database" for a calibration piece, one just needs the s1p file for the open, short or load, or s2p for the thru. Using the calibration kit menu, one simply defines each of the pieces by these files. Of course these data files must be created by a carefully calibrated and accurate VNA. It doesn't matter how many frequency points are used, but the more the merrier of course. The VNA will interpolate as necessary. As long as the calibration pieces themselves are mechanically/thermally stable with repeatable results, this method gives very good results.

One can implement 12 term calibration in Python or MatLab without too much trouble. The mathematics for it is in this slide set:

http://emlab.uiuc.edu/ece451/appnotes/Rytting_NAModels.pdf

[joeqsmith]

To add a "database" for a calibration piece, one just needs the s1p file for the open, short or load, or s2p for the thru. Using the calibration kit menu, one simply defines each of the pieces by these files. Of course these data files must be created by a carefully calibrated and accurate VNA. It doesn't matter how many frequency points are used, but the more the merrier of course. The VNA will interpolate as necessary. As long as the calibration pieces themselves are mechanically/thermally stable with repeatable results, this method gives very good results.

One can implement 12 term calibration in Python or MatLab without too much trouble. The mathematics for it is in this slide set:

http://emlab.uiuc.edu/ece451/appnotes/Rytting_NAModels.pdf

I could make a Touchstone file with ten points or thousands.   I could have one point every GHz or one every 10Hz.  I have to imagine the companies making these low cost standards use a common resolution.    I would also assume, say it's a 100KHz to 100GHz standard and you want to measure 1MHz to 1.1MHz.  You use what ever few points from the touchstone file and maybe fit it.    I'll read the paper and see if it covers some of the basic details.   I have had no luck with my searches.  I was thinking to write Kirkby as I assumed he would use touchstone but I couldn't find where he supports it.   

I have not seen were anyone is using touchstone for the metrology grade standards, only for the low cost references.

[Mechatrommer]

I have had no luck with my searches.  I was thinking to write Kirkby as I assumed he would use touchstone but I couldn't find where he supports it.

i'm not sure what you are trying to search. even though the VNA Kirkby is using may not support databased cal data, it may produce touchstone (s1p) data for measured DUT i assume. or he may has a tool in PC to do format conversion.

I have not seen were anyone is using touchstone for the metrology grade standards, only for the low cost references.

probably due to metrology grade VNAs are the old boat anchors that only support polynomials coefficients? if you have friend with Keysight FieldFox unit (more modern VNA), you may ask how they do calibration, using s1p files or still using coefficients? the drawback with databased (s1p touchstone) method is if we are measuring on noisy VNA, such as Nano and Deepace VNA. the noise/error will be reflected in the s1p file and will be brought to other VNA thats using the s1p file for calibration. one of the trick will be to smoothen/averaged the database/s1p/touchstone file with hope to attenuate the noise/systematic error in the measurement.

[virtualparticles]

1601 points is pretty common the the data base file. As long as the cal piece isn't changing quickly then interpolation works well. Rytting is the go-to guy for all things In VNA calibration.

[switchabl]

The most common application of data based calibration is probably e-cals. But you usually don't see the data because it is stored on the unit and loaded automatically. Common mechanical SOLT cal-kits at reasonable frequencies are well-behaved enough that coefficient based characterization is normally fine. Once you get beyond 20-30GHz or so, there is more benefit from data based standard definitions. Mind you, that data doesn't have to be measured data, it often comes from CEM modeling.

Touchstone files are the standard for exchanging S-parameter data. VNA manufacturers have their own proprietary calkit files that contain values for several standards along with extra info on type of standards, frequency range, maybe uncertainty data etc. But you should always be able to import individual touchstone files into the cal-kit editor on your VNA if you need to define a custom one.

The minimum number of data points will depend on how smooth the data is. And well, there must of course be enough to avoid phase ambiguities (but that is a low bar). There are different ways to go about the interpolation, if you interpolate real and complex parts separately, that is fine but may need more data points than interpolating in polar coordinates. Evaluate by doing a measurement at higher resolution and compare to the lower point-count interpolated version.

You should also do multiple measurements and average them.

[joeqsmith]

I have had no luck with my searches.  I was thinking to write Kirkby as I assumed he would use touchstone but I couldn't find where he supports it.

i'm not sure what you are trying to search. even though the VNA Kirkby is using may not support databased cal data, it may produce touchstone (s1p) data for measured DUT i assume. or he may has a tool in PC to do format conversion.

Min data points, resolution... all the things I've mentioned.  Just the basics.  For Kirkby, I just didn't see where he offered anything beyond the standard coefficients.  If he were doing this as a normal service, I would have asked about the format. 

1601 points is pretty common the the data base file. As long as the cal piece isn't changing quickly then interpolation works well. Rytting is the go-to guy for all things In VNA calibration.

Good info.  I really had no idea.   So the thought is what ever the calibration standards are, you would always have 1600 points?   Is this documented somewhere or is it just typical of what you have seen?   

From what you have found, do they commonly provide you with data outside of what the kit is rated for? For exampled 900K to 10.1MHz for a 1 to 10MHz standard? 

I read the paper but again, they don't provide any details about what the companies who are making the standards are doing, which is really what I was after.     

The most common application of data based calibration is probably e-cals. But you usually don't see the data because it is stored on the unit and loaded automatically. Common mechanical SOLT cal-kits at reasonable frequencies are well-behaved enough that coefficient based characterization is normally fine. Once you get beyond 20-30GHz or so, there is more benefit from data based standard definitions. Mind you, that data doesn't have to be measured data, it often comes from CEM modeling.

The two papers I linked above led me to believe the primary reason was to reduce cost.  Each kit would have it's own baby papers to ensure accuracy. 

It makes sense that as you go up, this technique would provide better performance.   

Anything for work I would imagine is e-cals now.  I doubt many places are using mechanical standards on their 4-port + VNAs.     

Touchstone files are the standard for exchanging S-parameter data. VNA manufacturers have their own proprietary calkit files that contain values for several standards along with extra info on type of standards, frequency range, maybe uncertainty data etc. But you should always be able to import individual touchstone files into the cal-kit editor on your VNA if you need to define a custom one.

The minimum number of data points will depend on how smooth the data is. And well, there must of course be enough to avoid phase ambiguities (but that is a low bar). There are different ways to go about the interpolation, if you interpolate real and complex parts separately, that is fine but may need more data points than interpolating in polar coordinates. Evaluate by doing a measurement at higher resolution and compare to the lower point-count interpolated version.

You should also do multiple measurements and average them.

I am not surprised to hear they would have some sort of custom format.   

Anyway, interesting topic.  I would like to know more about it.  If people find materials, please post them up.   Know of a good book that covers it, ISBN would be great. 

Thanks.

[Mechatrommer]

I have had no luck with my searches.  I was thinking to write Kirkby as I assumed he would use touchstone but I couldn't find where he supports it.

i'm not sure what you are trying to search. even though the VNA Kirkby is using may not support databased cal data, it may produce touchstone (s1p) data for measured DUT i assume. or he may has a tool in PC to do format conversion.

Min data points, resolution... all the things I've mentioned...
For Kirkby, I just didn't see where he offered anything beyond the standard coefficients. If he were doing this as a normal service, I would have asked about the format.

Kirkby provided s1p touchstone format spanning from 50MHz to 12GHz (6GHz rated cal kit), besides a master (text) file for coefficients and other text files compatible with various VNA's brands derived from master file. see attached as a teaser, i've provided RL plot for the Load earlier from s1p measured data. whats not shown is countless literatures he included in usb drive (mostly HPAK's that you can easily download (i've downloaded most) from the net) that i've backupped in my external drive. for your other questions, let the persons you replied to answer them. short answer... Kirkby provided 1601 points of what i believe as measured data (averaged maybe) but not derived from modelled data, if they are, my little tool will plot exactly matched between measured and modelled.

[virtualparticles]

1601 points is pretty common the the data base file. As long as the cal piece isn't changing quickly then interpolation works well. Rytting is the go-to guy for all things In VNA calibration.

Good info.  I really had no idea.   So the thought is what ever the calibration standards are, you would always have 1600 points?   Is this documented somewhere or is it just typical of what you have seen?   

From what you have found, do they commonly provide you with data outside of what the kit is rated for? For exampled 900K to 10.1MHz for a 1 to 10MHz standard? 

Our Automatic Calibration Modules (ACMs, same as eCal) use a 1601 point data base over the operating frequency. Data is only taken over the operating frequency range and it is actually measured with a golden VNA. There are more than 4 calibration artifacts used internally as you can see by the block diagrams. A least squares approach is used to perform calibration in the over-determined system of equations. This is helpful as it improves the accuracy of the calibration somewhat with some random variation in the measurements.

[mawyatt]

Again, a good book for those who haven't read it.

https://www.amazon.com/Death-Expertise-Campaign-Established-Knowledge/dp/0190469412

I am not entirely sure what you are trying to say. If it is "don't believe something just because a chinese seller or a random blogger says so", I concur.

On the more immediate subject matter, the "Handbook of Microwave Component Measurements" by Joel Dunsmore is something I have found quite helpful.

If you do a little research you will find the chinese seller you refer to, Hugin is actually one of the original developers of the NanoVNA and also the OEM of various versions of the NanoVNA. The other chinese seller HCXQS the designer and OEM of the NanoVNA V2 Plus and Plus4. The random blogger you refer to is Avian's Blog, I certainly find his blog interesting and informative, maybe not as a reference material like Handbook reference, but still worthwhile.

BTW Dr Joel Dunsmore is a brilliant creative engineer/scientist that has published some excellent references and has numerous patents. I've personally known Joel since 1995 when we both were on the initial advisory board helping create the Center for Wireless and Microwave Information Systems (WAMI) at University of South Florida (USF) where I later was an adjunct. Joel represented HP/Agilent with equipment for the lab, along with Tektronix and Mini-Circuits supplied additional equipment and components respectively, and I represented Honeywell and supplied significant "$ endowments"  to help finance the initial lab setup. We worked closely with the WAMI founders Drs Larry Dunleavy and Tom Weller, which later founded Modelithics at USFs research center. WAMI has been a huge success over the years and a credit to Larry, Tom, Joel and other supporters!

http://wami.eng.usf.edu/about/mission.htm


Best,