Kirkby calibration kit alternatives?

[G0HZU]

To demonstrate this I dug out an old library file of a 1000R 0805 chip resistor and this was measured with my DIY cal kit up to 3GHz.

Note the scale for resistance is 5 ohm/div and it is measuring parallel Rp and Cp (not series Rs and Rs).
So this should look like 1000 ohm in parallel with about 0.08pF up to about 1GHz before the parasitics begin to change things. This is with a cal kit that does not have Lx corrections for the SHORT so I had to optimise the SHORT by other means.


How would this compare to others on here? Have you ever tried to measure a chip resistor like this? You need a very good VNA and an accurate cal kit and SMD test fixture to do this and get data as smooth as this over a wide bandwidth

[G0HZU]

Here's an old plot of a precision 32.4k chip resistor. I'm pushing the VNA to its limits here so the data is noisy and there will be a fair bit of uncertainty in the setup. But it managed to measure 32k4 with a flat response up to 100MHz. With a 100k resistor it gets really noisy but it does still manage to give a reasonable measurement.

This is with a DIY SMA cal kit made from scrap parts but with careful manufacture and a decent USER cal file.

[Mechatrommer]

The results for the short aren't as good because there aren't any Lx corrections in the data. But it is still good considering it is using SMA based connectors in the SHORT. It would be easy to correct this for VNAs that support Lx coefficients and if a VNA supports this feature then I guess these would be included in the data?

i cant see any inductance effect for SHORT in any kirkby calibration data including the master file from where everything else are derived. so i guess his measurement setup is not supporting that, or consider the effect negligible. albeit not perfect, i can say ±0.5 degree error is good enough for me i guess (my VNA systematic error will be much worse from what i figured out so far). thanks for all your reports and findings..

[G0HZU]

That's OK. Note that I had some free time today and had a look at your male OPEN and SHORT data in your custom correction file and the bundled S1p files. The male SHORT doesn't look as good as the female SHORT in your kit. There's a bit more phase shift in the male SHORT and although it's going to be better than an uncorrected kit, I'm not so impressed with it.

I generally prefer to use female cal kits because this is good for calibrating for non insertable devices that have SMA female connectors. But for insertable devices it's common to use both kits for a calibration. So I kind of ignored the male standards in your kit until today. There may be a way to improve this performance but I'd have to have one of these kits here to play with it.

[hendorog]

This is a bit left field, but bear with me as it could be a simple method to get a better characterised cal kit ...

I have an SDR kits cal kit which is only supplied with electrical delay values. It has no fringing capacitance coefficents.
The open standard in the kit is just a dust cap. It has no effect on the measurement, at least not significant enough to matter.
Therefore the open standard is just the open female SMA adapter which I have on the port of my test set.

Taking a punt here and assuming that the female SMA adapter (which came with the Rosenberger SDR kit) I have is very similar to someone elses.

Therefore, could someone with a nice cal kit (G0HZU?) do me a big favour and sweep an open female SMA, and post the S1P file?

Then I'll take the file, try and run it through the script and see what C0,C1,C2,C3 values come out. Then put them into the VNA and try it out.

The short and load are already covered - there are no L0,L1,L2,L3 fields in the old VNA's for shorts, so I can't input those anyway.
Since I already have the delay for the short, so that is the best I can do for now.

I already have the SDR Kits load, as well as several non-precision 18GHz Inmet loads from eBay, out of which I have selected a few which measure quite close to 50 ohms at DC. The best I can do here is hope that they are still close enough to 50 ohms up to 6 GHz. At least they all compare sensibly with each other and with the SDR Kits load.

Anyone still with me?

[G0HZU]

Probably the most controlled way to try your experiment is to see if you can generate Cx coefficients for the Kirkby kit OPEN s1p file posted below. Then post up your Cx values here and I'll see how they perform in terms of flattening the phase? Will I see the same flat result as in post #47 for example?

[hendorog]

Probably the most controlled way to try your experiment is to see if you can generate Cx coefficients for the Kirkby kit OPEN s1p file posted below. Then post up your Cx values here and I'll see how they perform in terms of flattening the phase? Will I see the same flat result as in post #47 for example?

Will do.

I will need to setup the NLopt stuff first for the script to work. I had it setup in the past but it must have been on the laptop which died on me.

[G0HZU]

How long does it take to run?

I've added some auto optimise limits in the RF simulator to see if it can do a better/faster job than me. For my open and short I tweaked the Cx and Lx coefficients manually whilst looking at the phase. It took a minute or so to get it quite flat. With this attempt at an auto optimiser I gave it a 'bad' starting point with the phase deliberately bad and so far it hasn't managed to beat my manual attempts and it seems stuck in a rut. I think I need to tweak the optimiser a bit...

[hendorog]

Probably the most controlled way to try your experiment is to see if you can generate Cx coefficients for the Kirkby kit OPEN s1p file posted below. Then post up your Cx values here and I'll see how they perform in terms of flattening the phase? Will I see the same flat result as in post #47 for example?

Ok here is the output of the first successful run. I can see C1 has hit its limit so not expecting much...
I didn't change any of the defaults for the initial settings. I suspect I will need to give it an accurate offset delay.

It took several minutes, but I didn't hang around to watch. I was out of the room for 10 minutes or so.
The installation of NLopt is pretty painful. There was a identifier which had changed (Octave_map vs octave_map), and some messing around with include paths etc.

Attached are the graphs it produced, and below is the final output.

Code: [Select]

1st optimization ended with status: Optimization stopped because xtol_rel or xtol_abs was reached. (NLOPT_XTOL_REACHED)
2nd optimization ended with status: Generic success return value. (NLOPT_SUCCESS)

Optimization results:
offs_delay = 50.000000 ps
offs_loss = 5.000000 Gohm/s
offs_Z0 = 50.000000 ohm
C0 = -172.920826 * 1e-15 F
C1 = 10000.000000 * 1e-27 F/Hz
C2 = -1238.347686 * 1e-36 F/Hz^2
C3 = 19.888550 * 1e-45 F/Hz^3

Residual RMS error : -47.880563 dB


[hendorog]

Second run - I opened up the limit for offset delay, as the first run had hit the limit of 50ps.

This completed in less than 1 minute:

Code: [Select]

?1st optimization ended with status: Optimization stopped because xtol_rel or xtol_abs was reached. (NLOPT_XTOL_REACHED)
2nd optimization ended with status: Generic success return value. (NLOPT_SUCCESS)

Optimization results:
offs_delay = 55.447558 ps
offs_loss = 5.000000 Gohm/s
offs_Z0 = 50.000000 ohm
C0 = -265.965357 * 1e-15 F
C1 = -2793.805192 * 1e-27 F/Hz
C2 = 1240.335578 * 1e-36 F/Hz^2
C3 = -194.306089 * 1e-45 F/Hz^3

Residual RMS error : -53.982524 dB


[hendorog]

Best one so far - in terms of residual. I opened up the Z0 limit, to range from 48 to 52 ohms.

Code: [Select]

Optimization results:
offs_delay = 55.465329 ps
offs_loss = 5.685029 Gohm/s
offs_Z0 = 49.172627 ohm
C0 = -286.232017 * 1e-15 F
C1 = -2372.098441 * 1e-27 F/Hz
C2 = 1891.981433 * 1e-36 F/Hz^2
C3 = -246.050762 * 1e-45 F/Hz^3

Residual RMS error : -62.879680 dB


[G0HZU]

The last two look the best but I'm not sure if I'm modelling the offset Zo correctly. I found that tweaking Zo improved the male SHORT Kirkby kit and that was the thing I would have experimented with if I had this kit here as per post #53.

Are you optimising to 3GHz or 6GHz? Your data looks flat to 3GHz but there is a tiny slope up to 6GHz. But it still looks good.

However, some of your values look strange. Normally the fringing capacitance is about 50fF but your optimiser has put in a negative capacitance of -286fF. I guess that will still be OK in the world of mathematics but can you set realistic limits for this capacitance of maybe 30fF to 100fF and run it again?

On the Agilent kit this capacitance only varies by a tiny amount away from 50fF across frequency.

[hendorog]

The last two look the best but I'm not sure if I'm modelling the offset Zo correctly. I found that tweaking Zo improved the male SHORT Kirkby kit and that was the thing I would have experimented with if I had this kit here as per post #53.

Are you optimising to 3GHz or 6GHz? Your data looks flat to 3GHz but there is a tiny slope up to 6GHz. But it still looks good.

By default it is sweeping over the range in the Dr Kirkby S1P file - from 50MHz to 12GHz.

However, some of your values look strange. Normally the fringing capacitance is about 50fF but your optimiser has put in a negative capacitance of -286fF. I guess that will still be OK in the world of mathematics but can you set realistic limits for this capacitance of maybe 30fF to 100fF and run it again?

On the Agilent kit this capacitance only varies by a tiny amount away from 50fF across frequency.

No problem, I'll constrain the C values to that range and run it again. Could be just that the Dr Kirkby standards are not as mechanically 'refined' as the Agilent ones?

[G0HZU]

Here's what appears to be the custom coefficients for that Kirkby female OPEN and C0 looks to be about 44fF.

Female open. (Use OPEN(M) or OPEN -F- depending on model of VNA - see notes above) **
Minimum frequency:  0 GHz
Maximum frequency:  7.0 GHz
Offset Zo: 50.0 ohms
Offset loss: 3.0  G ohm/s
Offset delay: 39.8  ps
C0: 44.36
C1: 156.42
C2: 1159.1
C3: -131.92

I just spotted your attached results files.

The optimiser tool you are using looks very powerful. I tried downloading it and the other sparam tool etc and I already have GNU Octave but I wasn't sure what to type in the command line. I tried an edited version of the default example but it didn't like it. I'm a newbie with GNU Octave so I probably haven't installed something correctly. I assume all the plots come from this optimiser tool?

[G0HZU]

Try running it up to 6GHz or maybe 7GHz as Fmax. It might give a flatter result?

[hendorog]

Here's what appears to be the custom coefficients for that Kirkby female OPEN and C0 looks to be about 44fF.

Female open. (Use OPEN(M) or OPEN -F- depending on model of VNA - see notes above) **
Minimum frequency:  0 GHz
Maximum frequency:  7.0 GHz
Offset Zo: 50.0 ohms
Offset loss: 3.0  G ohm/s
Offset delay: 39.8  ps
C0: 44.36
C1: 156.42
C2: 1159.1
C3: -131.92

I just spotted your attached results files.

The optimiser tool you are using looks very powerful. I tried downloading it and the other sparam tool etc and I already have GNU Octave but I wasn't sure what to type in the command line. I tried an edited version of the default example but it didn't like it. I'm a newbie with GNU Octave so I probably haven't installed something correctly. I assume all the plots come from this optimiser tool?

Yes it does look very impressive. The docs are here:
https://nlopt.readthedocs.io/en/latest/NLopt_Matlab_Reference/

If you are on Windows, there is a page which summarises the steps. However I would not recommend trying the install just yet. I will document what I had to do as that will make it much much easier for you and anyone else to get it going.

[hendorog]

Try running it up to 6GHz or maybe 7GHz as Fmax. It might give a flatter result?

Aha, I did not notice this earlier. The graphs were being output with a hardcoded and therefore incorrect frequency range which did not match the input file.

I am changing the script now to use the fmin and fmax values from the S1P file.

The optimiser itself uses the correct frequency range, and so this won't change the results.

Here are the results from a run with the Cx values limited to the range 30-100 units. Doesn't look very good at the higher freq's.

Code: [Select]

Optimization results:
offs_delay = 35.951981 ps
offs_loss = 10.000000 Gohm/s
offs_Z0 = 49.000000 ohm
C0 = 100.000000 * 1e-15 F
C1 = 99.011405 * 1e-27 F/Hz
C2 = 30.000000 * 1e-36 F/Hz^2
C3 = 30.000000 * 1e-45 F/Hz^3

Residual RMS error : -22.299742 dB


[G0HZU]

Yes, that looks bad at the top end. The 30.0000 values look suspicious. Has this hit a limit?

I don't think you need to alter Zo away from 50R on this OPEN because DrKirkby managed to get a really flat line with Z0 = 50 in my earlier post.
Can you try locking in some of his values and see if it works better with fewer variables? I'm off to bed soon so can't do much more tonight...

[hendorog]

Yes, that looks bad at the top end. The 30.0000 values look suspicious. Has this hit a limit?

I don't think you need to alter Zo away from 50R on this OPEN because DrKirkby managed to get a really flat line with Z0 = 50 in my earlier post.
Can you try locking in some of his values and see if it works better with fewer variables? I'm off to bed soon so can't do much more tonight...

Yes that one hit all of the limits - I constrained it to 30-100 on all of the C values.

I've opened them back up again and I will lock the Z0 back at 50.

[hendorog]

Also, thanks for your help

[suj]

For comparative purposes I've downloaded S1P for female open given by G0HZU in post #55 and checked what definitions will be generated by the tool I use. It's METAS VNA Tools II.
https://www.metas.ch/metas/en/home/fabe/hochfrequenz/vna-tools.html
This is a powerful metrological tool focused on VNA measurements, and one of its parts is a tool for optimizing the definition of standards. The results are similar to those obtained by hendorog in post #60.

[hendorog]

Thanks, that is interesting. They have definately found the same point.
Maybe try the next one too if you have time:

I did another run after removing the data > 7GHz from the S1P file. This fits much better and I think it makes more sense to use a < 7GHz fit for this kit on an 8753 for example which can't go above 6GHz.

I've attached the output pdfs for a couple of new graphs I added to the script. These show the difference between the model and the s1p file and I think these are comparable to what G0HZU shows in post #47. The scale on the phase graph is +/- 1 degree though instead of +/- 5 degrees

I get something similar to one or the other of the following results :-

Code: [Select]

Optimization results:
offs_delay = 52.360635 ps
offs_loss = 5.470724 Gohm/s
offs_Z0 = 49.000012 ohm
C0 = -226.929303 * 1e-15 F
C1 = -1964.414279 * 1e-27 F/Hz
C2 = 2232.392337 * 1e-36 F/Hz^2
C3 = -236.119457 * 1e-45 F/Hz^3

Residual RMS error : -67.897497 dB


Code: [Select]

Optimization results:
offs_delay = 34.432636 ps
offs_loss = 5.643170 Gohm/s
offs_Z0 = 49.000000 ohm
C0 = 138.424072 * 1e-15 F
C1 = -1291.076067 * 1e-27 F/Hz
C2 = 2442.710614 * 1e-36 F/Hz^2
C3 = -217.883669 * 1e-45 F/Hz^3

Residual RMS error : -67.731741 dB
[code]



[suj]

After removing the data f>7 GHz from S1P:





Here, I normalized the graph to the data from S1P:

[hendorog]

Very nice, quite similar to the first result.

Also, I just realised that METAS VNA Tools II tool you are using is free. 

[Mechatrommer]

Also, I just realised that METAS VNA Tools II tool you are using is free. 

but i need to email someone to get the download link. if click the download link, setupping outlook will popup  i just tried copy paste the email address into yahoo mail.