Calculating correction data for an unknown VNA calibration kit

[TheSteve]

A user on the forum asked for help measuring his Times Microwave OE5108 N male calibration kit that had no correction data.

I have experimented with this a little in the past so I thought I'd give it a go.

The process is as follows:

Warm up my VNA for several hours
Calibrate the VNA with a known calibration kit
Measure S1P (touchstone) data for each of the unknown standards (open, short, load)
Find best fit correction data
Load the correction data into the VNA, perform a calibration and verify performance

Equipment used:

VNA - Keysight FieldFox N9918A 26.5 GHz VNA
Cable: Gore Phaseflex 3.5mm cable
N adapter: Keysight precision 3.5mm to N female adapter

I have three calibration kits on hand to compare against:
Keysight 85032E 6 GHz mechanical O/S/L
Keysight 85092-60005 9 GHz ECAL
Maury Microwave 8850P 18 GHz mechanical O/S/L

The DUT is a Time Microwave OE5108 18 GHz O/S/L N male kit

The target VNA the OE5108 kit is to be used on is a Keysight 8753C with 6 GHz option.

The higher the frequency the harder it is to not only get good data but also calculate the best fit correction data. With this in mind I will be starting with a frequency range of 300 kHz to 6 GHz using the Keysight 85032E mechanical calibration kit.

To perform the math required to calculate the best fit correction data I will be using METAS VNA Tools II.

Attached are the following files/pictures:

Measurement setup
Touchstone(s1p) files with the measured data for the OE5108 cal kit pieces.
Graphs of the measured data and calculated best fit correction curves.
Calculated correction constants

This is a first pass so far, I've not yet done any verification of the calculated values. I have let METAS VNA Tools optimize all parameters other than impedance for now.
I'd like suggestions on the best way to verify the cal kit and anything in the process that may be improved. I can then post updated data and verification results. If 6 GHz goes well I will repeat the process for 18 GHz.


Here are the current calculated correction values(also attached as a file):

Agilent Definitions:

Standard Type: Open (N male)
Offset Z0 (Ohm):   50
Offset Delay (ps):   117.251566
Offset Loss (GOhm/s):   0.619522
C0 (E-15 F):      338.152624
C1 (E-27 F/Hz):      8100.514019
C2 (E-36 F/Hz^2):   -4468.405384
C3 (E-45 F/Hz^3):   719.894636
RMS Error   Max Error
0.004862   0.009313

Standard Type: Short (N male)
Offset Z0 (Ohm):   50
Offset Delay (ps):   73.521902
Offset Loss (GOhm/s):   0.206593
L0 (E-12 H):   1065.970546
L1 (E-24 H/Hz):   16965.387578
L2 (E-33 H/Hz^2):   -3271.896260
L3 (E-42 H/Hz^3):   1787.549767
RMS Error   Max Error
0.003897   0.006611

Standard Type: Load (N male)
Offset Z0 (Ohm):   50.000000   
Offset Delay (ps):   132.802668   
Offset Loss (GOhm/s):   0.000000
RMS Error   Max Error
0.013200   0.021613

[Joel_Dunsmore]

The graphs show a yellow curve and a black curve.  Can you explain what each curve is? (or maybe I missed it).

[TheSteve]

Yes, I forgot that. The black(magnitude) is the measured data, yellow is the "best fit" calculated curve. These are plots generated by METAS VNA Tools.

[Joel_Dunsmore]

Yes, I forgot that. The black(magnitude) is the measured data, yellow is the "best fit" calculated curve. These are plots generated by METAS VNA Tools.

That's what I inferred. The deviation from ideal is 1.006 vs expected (just below 1), so that 0.006 error represents a -44 dB term which is almost certainly error in the calibration. Normally I would expect Ecal to be significantly better than a mechanical cal kit, unless you use a sliding load cal kit.   In some ways the best fit kind-of "smears" the error and so likely gives a reasonable result, but for sure, any value greater than 1 must be an error (or you have a perpetual motion machine).   

[Bud]

Is it worth trying to mechanically fixate the N adapter of the cable to prevent cable movement ?

[Joel_Dunsmore]

One more thing: this kit does not have a connection for the center pin of the open. Thus, the model of the open depends a bit on the exact nature of the SMA-to-N adapter that you use.  The electrical-model of the open presumes a 6 slot socket (F) end, but most adapters use a 4 slot end, and thus have a slight lower capacitance than the 6 slot end (the 4 slot has longer slots and tapers to a smaller value).   The gold adapter might be a precision adapter so it might have the proper socket (F) end.  Without a center pin, the socket side impedance is a bit higher than expected, which is accounted for in the model. The Ecal of course will have a center pin that will will not be affected by the time of SMA-to-N adapter.  I notice the Times kit also has a pin for the open standard. 

[Bud]

One more thing: this kit does not have a connection for the center pin of the open.

You mean the reference Keysight 85032E cal kit?

[TheSteve]

The 3.5mm to N adapter I am using is the 6 slot end version. I will keep experimenting with the various cal kits I have here but so far the results look most optimal with the 85032E kit. I am also testing things on a VNA with native N female connectors so no adapters of any kind are used.

[hendorog]

At a more basic level, I've had better luck in the past with DIY cal kits by chopping the freq range into two segments:
e.g.
300kHz to 3.5GHz
3.5GHz to 6GHz

Less convenient to use of course but might be worth a crack.

[TheSteve]

I'm frustrated so far with the results I'm seeing with the various N cal kits I have. When I have done this before with 3.5mm gear I've had some amazing results(and some so/so results). I'm not giving up though.

[Joel_Dunsmore]

One more thing: this kit does not have a connection for the center pin of the open.

You mean the reference Keysight 85032E cal kit?

Yes, so the length of the (F) socket can affect the calibration, and it should be gauged.  If you had a (M) Ecal, I would think it should give the best results if the Ecal connector isn't damaged and isn't too old (the Ecal can drift over time, but this mostly affects the S21 thru, not so much the OSL definitions.

It might be the Times kit has an offset impedance from 50 ohms; if so it can give you that odd shape as well; but most likely it is an error in the cal.  For example, the short should be very near 1.000 as it is a total reflection, regardless of anything else.  The fact that it shows a ripple is almost for sure a consequence of the initial type-N calibration.  Maybe try calibrating with Ecal and then measuring both 85032E standards and the Times standards.  See if anything becomes apparent.

[TheSteve]

I've tried the ECAL and it performs a cal but it certainly isn't any better. It likely needs a cal itself. It is a real struggle to find really good N ECAL units for hobby level pricing. I have excellent 3.5mm ECAL's though

[Bud]

At a more basic level, I've had better luck in the past with DIY cal kits by chopping the freq range into two segments:
e.g.
300kHz to 3.5GHz
3.5GHz to 6GHz

Less convenient to use of course but might be worth a crack.

Certainly is less convenient, but better than nothing.

@TheSteve - can we just do 0.3-3GHz   and   3-6 GHz sets of coefficients?

[TheSteve]

We can absolutely do that, but I haven't given up on a 6 GHz set yet.

[Joel_Dunsmore]

I've tried the ECAL and it performs a cal but it certainly isn't any better. It likely needs a cal itself. It is a real struggle to find really good N ECAL units for hobby level pricing. I have excellent 3.5mm ECAL's though

When you say isn't any better, can you plot the Ecal calibrated results (of the Times kit measurement) and the 85032E results on the same plot.  It would be interesting to see if the results are both bad in similar ways.  And, just for fun, measure the 85032E standards using the calibration from the Ecal.

The question in my mind is whether the Times kit is a bit odd, thus having a response that is not well modeled by the simple polynomial model, or if it all relates back to the reference calkit. But, since I don't see a ripple associated with the cable length (1 m should give a ripple on the order of a couple hundred MHz), then it means the raw source match and directivity of the Fieldfox is being removed.   

On the topic of splitting the cal standards: it won't do much good to get a "better match" if the reference data isn't correct.  So the thing to determine is if it the Times standards that are unusual, or if it's the reference cal kit.  The 3 GHz ripple pattern points to something like a 5 cm length, which I wonder if it is consistent with the actual length of Times kit. If it is, it might be the nominal center conductor impedance is off and so the delay term should have a non-50 ohm result.  The fact that the values you obtained has the impedance as exactly 50 ohms seems to indicate to me that the chance that the center-conductor to outer-conductor impedance error is not being considered in the fitting. Otherwise I would expect numbers like 50.00315 or 49.9990; given that you have 9 significant digits on the delay, for example.  Maybe you have  to fool around a bit with the offset delay value.

[Bud]

The serial number of the Times kit is very low:  # 0019. Could it be a prototype ?

[G0HZU]

I don't know if this helps but a few years ago at work I checked out the health of an old 85032E N type cal kit and I managed to find the project file for it today.

See below for a Genesys simulation after using the 85032E to initially calibrate an Agilent E5071C VNA up to 6GHz. The 85032E open and short was measured after a calibration and I then applied the C0 C1 C2 C3  corrections and the offset delay corrections via Genesys. The results are as below. I didn't include the Offset Zo or Offset Loss corrections so the plots of the reflection coefficient are what the VNA measures if presented with the 85032E after calibration. You can see that the phase response gives a flat line if I apply the official C0 C1 C2 C3 corrections and the offset delay correction via Genesys. The magnitude of the reflection coeffcient is shown for both the open and the short.

Genesys also plots the fringing capacitance vs frequency for the combined C0-C3 corrections which may be of interest.

I've also included the info in the E5071B user manual for the 85032B/E cal kit. I have an old E5071B VNA here but these tests were done several years ago at work using a works E5071C VNA.

I've never had great results when doing critical stuff with N type cal kits or N type test fixtures so I'm interested to see how this thread progresses. I have a home made N cal kit here (male and female) and I'd be keen to try and improve it. So far I've generated C0-C3 and L0-L3 coefficients for it using Genesys but I think it would be better to use the META VNA Tools for this stuff.

It's worth checking out the way your VNA treats the gender of the cal kit descriptions. On the E5071 VNA an (f) means female for the analyser connector and the cal kit is fitted with a MALE connector. I'm not sure why Agilent did this but it can lead to confusion.

[TheSteve]

Thanks for posting the info. Newer VNA's now base everything off of the "DUT" connector/gender. They even show a picture of the connector you've selected and gender.
As I said above I've found this much more challenging with N connectors so far vs 3.5mm. In the past I calibrated using a known good 3.5mm ECAL and then measured the female Keysight 85052B set. I cheated and used the official impedance, delay and loss figures and let METAS VNA calculate the rest. The values it came up with were extremely close to the official Keysight values, shockingly close even. It is hard to reproduce to that level though, each connector mating is critical. If cables are used they can't be moved at all etc. This was also to the full 26.5 GHz.

[G0HZU]

Thanks. I can't remember now but I think I may have calibrated the E5071C using a 85032F cal kit when I did the above tests of the 85032E. There's a project file for the 85032F cal kit in the next folder and this was the full male and female cal kit. At work I think the production area now has a N type Ecal so I could try measuring my homebrew N cal kit with that at some point and I can re-check the 85032E if I can find it at work.

I'm planning on early retirement so I won't have access to new calibrated equipment in the near future... I have a decent 3.5mm Ecal module here at home but nothing decent for N connectors apart from a fairly dodgy Anritsu T cal kit.

[TheSteve]

Performed a cal using the 85032E kit, then saved S1P data for both the 85032E and the OE5108. This is using a different VNA with slightly different parameters, 2 MHz to 6 GHz.

Attached are the graphs from the measured 85032 open/short and the 5108 open/short. The S1P files are also attached.
So far to me the measured data looks pretty reasonable. Black again is measured data, orange is the best fit correction curve METAS VNA Tools has calculated.

I didn't attach plots but I also checked the Maury open/short after the 85032 cal and the lines were very reasonable(very similar to the 85032 kit), no big curves or dips. It seems only the OE5108 kit has them. Perhaps this kit can't be well characterized by the simple polynomial model as Joel mentioned.

[TheSteve]

As mentioned in my previous post I've been experimenting with correction values and then performing a calibration using the EO5108 cal kit.

Here are best results so far:

We have the return loss of the open female N connector on the VNA (0.2 dB/div) - this should ideally be a flat line.
Phase of the open female N connector
Phase with the Keysight 85032 open connected(looks 99.9% the same as with nothing connected which is correct).
Phase with the Keysight 85032 short connected.

I'd like to see flatter lines, but at least it is a result - Joel what are your thoughts?

[KE5FX]

You've seen Claudio, IN3OTD's Octave script, right?

[TheSteve]

I've experimented with his software as well in the past. I found METAS VNA easier to use.

[Mechatrommer]

If 6 GHz goes well I will repeat the process for 18 GHz.

best fit for 6GHz may not be the best fit for 18GHz and vice versa... source: me (or i've read or heard from other sources but have forgotten from where)... so you are free to reject the idea... cheers.

[TheSteve]

If 6 GHz goes well I will repeat the process for 18 GHz.

best fit for 6GHz may not be the best fit for 18GHz and vice versa... source: me... so you are free to reject the idea... cheers.

Reject what idea?

[TheSteve]

the idea that... best fit for 6GHz may not be the best fit for 18GHz and vice versa... cheers.

Of course the correction values will not be the same. The polynomial model is really only suited to a single curve, not a line with multiple curves. So measuring the kit to 6 GHz should get a better fit with the standard correction model vs going to 18 GHz.
In an ideal world with an ideal cal kit the error would be small across the entire range to 18 GHz and it wouldn't matter. If it was a nice curve with negative magnitude the correction would be quite simple. But it seems nothing is ideal

[Mechatrommer]

So measuring the kit to 6 GHz should get a better fit with the standard correction model vs going to 18 GHz.

i mean, you can measure once up to the maximum range your VNA can provide from the beginning (and how far your standard CAL kit is characterized).. because measuring up to 6GHz vs measuring the entire range (18GHz or anything your VNA can provide), you'll get the same measurement result, except  beyond that 6GHz. once the technique/method is established, its just a matter of feeding the curvefit function with any subset, ie a particular portion from the universal set (entire range measurement) to get different curve fit coefficients. i'm sorry if i'm not clear enough, its quite hard to explain it textually... (edit: except.. my point will be invalid if you want to argue about poorer resolution when measuring the entire range, ymmv cheers)

[TheSteve]

OK, I understand what you are saying now. I would only calibrate with a 6 GHz max as it makes better use of the number of sample points. If I used the same number and went to 18 GHz I'd only have one third of the sample points to 6 GHz. And yes I could just add more points but that slows things down.
The sample image you listed is the load, so I wouldn't worry about the graph to much. As for the the opens/shorts, they can only match the measured trace if it can be expressed by the polynomial equation used to generate the calibration correction values. If it can't match it perfectly then it just does the best it can.

[gf]

btw, i wonder why the curve fit estimates are too far off to my eyes in few of your images such as this? maybe something i'm not well versed in yet...

The black(magnitude) is the measured data, yellow is the "best fit" calculated curve.

I wonder in particular whether the shape of these yellow curves can be the result of a 3-parameter load model as given in the OP

Standard Type: Load (N male)
Offset Z0 (Ohm):   50.000000   
Offset Delay (ps):   132.802668   
Offset Loss (GOhm/s):   0.000000

for any (real) values of the model parameters Z0, Delay and Loss?

[Joel_Dunsmore]

btw, i wonder why the curve fit estimates are too far off to my eyes in few of your images such as this? maybe something i'm not well versed in yet...

The black(magnitude) is the measured data, yellow is the "best fit" calculated curve.

I wonder in particular whether the shape of these yellow curves can be the result of a 3-parameter load model as given in the OP

Standard Type: Load (N male)
Offset Z0 (Ohm):   50.000000   
Offset Delay (ps):   132.802668   
Offset Loss (GOhm/s):   0.000000

for any (real) values of the model parameters Z0, Delay and Loss?

For sure, with a load element that has either some net inductance or capacitance, you cannot get a good fit if the delay impedance is fixed to 50 ohms.  I am not familiar with METAS tool for fitting, but if it does not allow a change of delay impedance, it will have issues fitting some responses I think.

[Bud]

Sorry what is "delay impedance" ? Is it the impedance of the delay element of the Load model?

[Joel_Dunsmore]

Sorry what is "delay impedance" ? Is it the impedance of the delay element of the Load model?

Yes, it is the impedance of the delay element and it can be used to compensate for first order parasitic L or C. I discuss it in some detail in pages 738-741 of the 2nd edition of my book.

[TheSteve]

METAS VNA does allow impedances other than 50 ohms. I had fixed it to 50 for my initial run of calculations(you can fix any parameter you want). I have since changed this for the open and short, I can also try it for the load. I hope to have more time to experiment this weekend(supposed to rain, so heli flying is out).

btw, i wonder why the curve fit estimates are too far off to my eyes in few of your images such as this? maybe something i'm not well versed in yet...

The black(magnitude) is the measured data, yellow is the "best fit" calculated curve.

I wonder in particular whether the shape of these yellow curves can be the result of a 3-parameter load model as given in the OP

Standard Type: Load (N male)
Offset Z0 (Ohm):   50.000000   
Offset Delay (ps):   132.802668   
Offset Loss (GOhm/s):   0.000000

for any (real) values of the model parameters Z0, Delay and Loss?

For sure, with a load element that has either some net inductance or capacitance, you cannot get a good fit if the delay impedance is fixed to 50 ohms.  I am not familiar with METAS tool for fitting, but if it does not allow a change of delay impedance, it will have issues fitting some responses I think.

[Marsupilami]

This thread is getting long and I wasn't following it so forgive me if it was said and/or I'm missing some detail but do I get it right that you're trying to characterize an N cal kit with 3 other N cal kits, none of which you're sure about they're accurate?
I'm wondering if you'd be better off cal'ing at the end of your 3.5mm cable with your known good 3.5mm ECal. Adding the 3.5-N adapter will still add 3 unknowns but maybe you can make better educated guesses about those than the who knows what's wrong with them N cal standards.
I'm not very versed in the traditional cal standard models, I've always been using data based standards (always meaning for the 5 minutes I've been working with VNAs) but maybe you would still be able to try to curve fit through the adapter, in which case your parameters would be obviously off but potentially show smaller error compared to the measurement result, proving that the your N calibration was at fault.

[TheSteve]

After experimenting I am very pleased with the calibration the 85032E is providing. I have gathered the S1P data for the Times Microwave kit many times now. It is pretty repeatable. I have calculated correction constants to 6 GHz. From 300 kHz to 5.5 GHz the maximum magnitude error after calibrating with the Times Microwave kit is 0.2 dB or less. From 5.5 to 6 GHz it increases to 0.3 dB. Phase error is also looking reasonable to 5 GHz with some increasing error after that, but still usable.

For reference the magnitude error from my ECAL is also around 0.2 dB - should be lower but the male connector appears to be quite worn. The magnitude error from my Maury Microwave kit is around 0.1 dB.

I've attached pictures of the measured Times Microwave OSL data. Black is measured data, orange is best fit correction.
The load picture has extra colors as it is also calculating best fit parameters for Anritsu and R&S equipment and they aren't the same for the load.

The open and short phase pictures are after calibrating with the Times Microwave kit and then measuring the Keysight 85032E open and short.

The zip file has the S1P data, graphs, correction values and files for various Keysight VNA's to make loading the constants easy.

Final data values:
Agilent Definition
Standard Type:      Open
Offset Z0 (Ohm):   50.151807
Offset Delay (ps):   110.016566   
Offset Loss (GOhm/s):   1.265273
C0 (E-15 F):      505.266674
C1 (E-27 F/Hz):      -13751.991616   
C2 (E-36 F/Hz^2):   2877.240920
C3 (E-45 F/Hz^3):   628.099576
RMS Error:      0.002981
Max Error:      0.007227

Standard Type:      Short
Offset Z0 (Ohm):   50.474695
Offset Delay (ps):   73.979612
Offset Loss (GOhm/s):   1.250892
L0 (E-12 H):      1001.065848
L1 (E-24 H/Hz):      28754.821253
L2 (E-33 H/Hz^2):   4696.023892
L3 (E-42 H/Hz^3):   254.983229
RMS Error:      0.002161
Max Error:      0.004829

Standard Type:      Load
Offset Z0 (Ohm):   49.587352
Offset Delay (ps):   159.997260
Offset Loss (GOhm/s):   2.795148            
RMS Error:      0.011838
Max Error:      0.023109

Workflow was to calibrate with the 85032E cal kit. Save S1P files for the open/short/load of the OE5108 kit and then load those files in METAS VNA to find the best fit correction data. The correction data was then loaded in the Agilent cal kit editor and saved. From there it is transferred into my VNA so I can perform a test calibration with the new kit defs.

I'm hoping to repeat this process with an unknown R&S 26.5 GHz 3.5mm kit in the near future. it will be referenced from a 26.5 GHz 3.5mm ECAL unit. I will start a thread for that one as well.

[Joel_Dunsmore]

This looks quite reasonable.  I do notice that this set of coefficients as the delay impedance as not exactly 50 ohms; did you do something different to get the curve fitting to accept non-50 ohm for this parameter.  I'm guessing the variation you saw was associated with the non-50ohm section. I also not some earlier plots had S11 greater than zero (which for sure indicated a calibration issue). Did you do something different in the calibration to get the better results for the measurement of Times open and short?

[G0HZU]

To see what my old homebrew N type cal kit can do up at 6GHz I brought it into work on Friday and measured it with an E5071C VNA and an 85032F cal kit. See below for the phase and s11 response after I computed the correction coefficients using Genesys for the male and female parts of the kit. The complete cal kit is made using male and female N type bulkhead connectors. These are very good quality connectors (expensive!) but I've been lazy and left the open 'open' as it has no end cap so it has some radiation losses up towards 6GHz. However, the results were quite good. When I first made this cal kit I only expected to use it up to about 3GHz. It's tempting to find a couple more bulkhead connectors and try fitting end caps to the male and female OPENs. It might work better up towards 6GHz if I do this.

The second set of plots are for the female half of the kit. These results seem a little better up at 6GHz in terms of s11. Note that the s11 response is raw/uncorrected as the results below are obtained in Genesys after optimising just C0 to C3 and L0 to L3 and the delay offset.

It is quite rewarding to manually tweak the delay and C0 through C3 (and L0 and L3) to optimise the response and I get the best results by then swapping across to the Genesys optimisation tool where I ask the optimiser to tune for minimum phase difference. However, it would be nice to try the IN3OTD and the METAS Tools to see what I can really achieve with this cal kit but I've tried both and they aren't intuitive to set up and use.

[G0HZU]

Note that these weren't ordinary N bulkhead connectors that I used in my homebrew cal kit. They were salvaged from some old pre-production prototype notch filters I designed at work 12-15 years ago. The filter had to have ultra low insertion loss to 6GHz and these bulkhead connectors were very expensive and have 8 slots in the female centre. They are physically very short and provide a direct launch into a PCB.

Here's an old plot of the notch filter. It had  about a 60dB notch around 1500MHz and what made this filter 'special' was that the insertion loss was very low on the high side of the notch all the way up to about 8GHz. I originally designed this notch filter using Sonnet as a multilayer structure. Since then thousands have been made and they can be used on the output of various wideband transmitters.

[TheSteve]

This looks quite reasonable.  I do notice that this set of coefficients as the delay impedance as not exactly 50 ohms; did you do something different to get the curve fitting to accept non-50 ohm for this parameter.  I'm guessing the variation you saw was associated with the non-50ohm section. I also not some earlier plots had S11 greater than zero (which for sure indicated a calibration issue). Did you do something different in the calibration to get the better results for the measurement of Times open and short?

I switched to another VNA with native precision N connectors on it. I don't trust my 3.5mm to N adapter to be 100%. I then allowed METAS VNA to optimize all parameters including impedance. Adapters, cables etc all look good until you set things to 0.1 dB/div.

[TheSteve]

However, it would be nice to try the IN3OTD and the METAS Tools to see what I can really achieve with this cal kit but I've tried both and they aren't intuitive to set up and use.

METAS VNA tools is quite easy to use for correction if you provide it the S1P files manually. I'm not sure all versions handle it though, I am using version 2.0.


Do you have the S1P files from your cal kit as measured after performing the cal with the 85032 that you can share? I'll pop them into METAS and see what it can determine.

[G0HZU]

Thanks. See below for the female OPEN and SHORT part of the homebrew N kit.  I'm going to remake the male OPEN as this connector has seen previous use and was soldered to a test enclosure at one point. This might explain why it seems worse than the others in terms of s11 loss.

This cal kit has been very useful over the years and it's probably good enough for the stuff I use it for. However, it would be interesting to see if it can be improved. I still feel that the various uncertainties associated with the limitations of the E5071C VNA and the 85032F cal kit mean that there isn't much point in trying too hard to fine tune these corrections. At some point the overall uncertainty dominates.

I downloaded METAS VNA Tools and I think it is version 2.58. It could be that I've downloaded the wrong version because I couldn't work out how to make it generate these corrections.

Note that the s1p files below cover up to 8.5GHz but I think it would be best to limit the optimisation to just 6GHz. I struggled to get good results above 6GHz when I tried.

[G0HZU]

If you need the 50R load data for the homebrew kit then I kind of cheat here because I used a male Anritsu 50R N type termination in series with a decent F-F N adaptor. This adaptor is very high quality and the female centres have 8 slots. I've attached it below if you need it. I do keep meaning to try and make a homebrew 50R N termination but at the moment I use the Anritsu termination.

[TheSteve]

I processed the open and short.

The black trace is the measured data, red trace is the best fit Agilent correction.
The error looks very low, interesting how it finds the ideal parameters are quite a bit off from 50 ohms.

Here are the suggest cal parameters:

Agilent Definition

Standard Type:      Open
Offset Z0 (Ohm):   54.323655
Offset Delay (ps):   52.659936
Offset Loss (GOhm/s):   2.376092
C0 (E-15 F):      -51.559719
C1 (E-27 F/Hz):      7194.698536
C2 (E-36 F/Hz^2):   -7632.558772
C3 (E-45 F/Hz^3):   660.074834

Agilent Fit Error
RMS Error:      0.000823
Max Error:      0.001326


Standard Type:      Short
Offset Z0 (Ohm):   34.781713
Offset Delay (ps):   18.304420
Offset Loss (GOhm/s):   2.913411
L0 (E-12 H):      1545.445695
L1 (E-24 H/Hz):      9342.640691
L2 (E-33 H/Hz^2):   -26590.509714
L3 (E-42 H/Hz^3):   3071.521470

Agilent Fit Error
RMS Error:      0.000578
Max Error:      0.001237

[G0HZU]

Thanks! The numbers for Z0 do seem a bit strange and I'll have to do some studying to make sure I understand what the results mean.

When I made this kit I really didn't expect it to work that well despite the huge cost of the connectors. In order to provide IP sealing the last couple of mm of the connector has a PTFE dielectric so there is a transition from air to PTFE right at the point it launches to a PCB. Presumably Suhner will have tried quite hard to optimise this transition but maybe this partly explains the strange results for Z0.

[TheSteve]

If I fix the impedance to 50 ohms the error doesn't change that much. Anyway, maybe give the values a try and see how it performs.

[Mechatrommer]

Thanks! The numbers for Z0 do seem a bit strange and I'll have to do some studying to make sure I understand what the results mean.

its explained in Agilent app note AN 8510-5B as transmission line's impedance after calibration plane. so i assume:
1) its inside the cal kit, and it was assumed Zc (transmission line's impedance before calibration plane) = Zo = 50, when older VNAs dont support modeling it.
2) all offsets... Zo, delay, loss are physical, they can be measured with proper tools, the problem is there is no easy "proper tools", so now with advance computing, they are included in the curve fit modelling.
3) Steve's VNA support it, so thats why he included the parameter in the modelling..

sometime i wonder... sometime we can leave fringing capacitances parameters to all 0s and tune the offset delay (or with offset loss) alone to get "good enough" model, or set the delay to 0 some other value and Z0 = simply 50 ohm... and tune the C and/or L 0,1,2,3 for Open/Short to get similarly good enough model. so i guess whats important is the model parameters calculated by curve fit function. the actual physical offset value (Zo, delay, loss) is less important/impact/preferable (due to the difficulty of actually measuring them), as long as the model function used in the VNA (given the parameters calculated by the curve fit function) can give very close to the actual measured S11. with data-based VNA, curve fit is unecessary, only interpolation. ymmv.

[TheSteve]

Thanks! The numbers for Z0 do seem a bit strange and I'll have to do some studying to make sure I understand what the results mean.

its explained in Agilent app note AN 8510-5B as transmission line's impedance after calibration plane. so i assume:
1) its inside the cal kit, and it was assumed Zc (transmission line's impedance before calibration plane) = Zo = 50, when older VNAs dont support modeling it.
2) all offsets... Zo, delay, loss are physical, they can be measured with proper tools, the problem is there is no easy "proper tools", so now with advance computing, they are included in the curve fit modelling.
3) Steve's VNA support it, so thats why he included the parameter in the modelling..

sometime i wonder... sometime we can leave fringing capacitances parameters to all 0s and tune the offset delay (or with offset loss) alone to get "good enough" model, or set the delay to 0 and Z0 = simply 50 ohm... and tune the C and/or L 0,1,2,3 for Open/Short to get similarly good enough model. so i guess whats important is the model parameters calculated by curve fit function. the actual physical offset value (Zo, delay, loss) is less important/impact/preferable (due to the difficulty of actually measuring them), as long as the model function used in the VNA (given the parameters calculated by the curve fit function) can give very close to the actual measured S11. with data-based VNA, curve fit is unecessary, only interpolation. ymmv.

You can certainly exclude any parameter you like. METAS VNA will indicate what the expected error is when it has calculated the best fit model. Of course I've never done anything to verify the math or modeling they use, but I have no reason to doubt it.
The Keysight N 85032B/E doesn't include any inductance values for the short and uses only delay, loss and impedance.

[Mechatrommer]

yeah i got a chance to read some of their app notes this morning making me realize how sloppy i am still... their setup up to 100+ GHz with narrow window of uncertainty is something beyond my capability. i've downloaded the setup file and app notes a while a go but dont have a chance to really use them,,, anyway i got to install and play with their VNA tool, i'm not sure if i'm doing it right but it seems too simple to use... just go to Cal Std Model Fit -> Load Data -> Start Optimization and thats it... yeah some parameters can be left fixed (uncheck Mask).. attached are 2 optimizations, all parameters optimized (1) vs only fring. Cs optimized with offsets fixed (2)... s1p used is provided by Steve (5108-open-85032-6ghz.s1p) fwiw... edit: but unfortunately when i tried to fit my diy (imperfect) Load kit (more specifically... Arbitrary Standard.. ie near but not exactly 50 ohm termination), META tool fitting still has some limitation there...