hello,

We’re trying to develop a set of SOLT cal kit. When defining fringe capacitance for the open standard. We used Ansoft HFSS to simulate the open to get capacitance vs frequency reading from the imaginary part of Z(1,1).

We noticed that as the open gets longer, the imaginary part of Z gets to inductive. This also can be showed by Smith Chart where the curve starts at the right most point and turns close wise and reaches inductive area, the upper half of the circle.

In Agilent PNA, the open’s fringe capacitance has to be represented by a polynomial equation vs frequency with coefficients of c0 to c3.

If it gets to the inductive area, I am not sure how to do that properly. Is there any solution there?

Thank you!

We’re trying to develop a set of SOLT cal kit. When defining fringe capacitance for the open standard. We used Ansoft HFSS to simulate the open to get capacitance vs frequency reading from the imaginary part of Z(1,1).

We noticed that as the open gets longer, the imaginary part of Z gets to inductive. This also can be showed by Smith Chart where the curve starts at the right most point and turns close wise and reaches inductive area, the upper half of the circle.

In Agilent PNA, the open’s fringe capacitance has to be represented by a polynomial equation vs frequency with coefficients of c0 to c3.

If it gets to the inductive area, I am not sure how to do that properly. Is there any solution there?

Thank you!

I'm afraid i need more clarification.

1. the group delay input to the ckt file need accout for the fringe capacitance or not?

2. it seems the fringe capacitance calculation has to use the numbers with the contact length de embedded? is this the right understanding?

3. see attached for what i meant for #1 and #2.

thank you!

## Attachments

C0, C1, C2, C3 (polynomial terms for the phase vs freq response of the fringing capcitance

Delay, Z_delay, Loss_delay: This represents in offset delay line in front of your open; in your case it would be approximately 10mm (or its group delay eqivlaent as delay is in ps).

Z_delay is for the impedance of the delay line, presumeably 50 ohms, loss_delay is the loss of the offset delay and can almost always be ignored, use 2 Gohms/sec for a reasonable number for 3.5 mm.

The open model is over determined, such that you can for example reduce C0 and increase delay to get the same result.

10mm air line causes dely of 66.7 ps withou fringe effect. if it addes fringe effect, the dealy would be 71.9 ps.

which number should i input to the ckt file as the delay?

thank you!

when you say "... what fitting works best"? Are there some criteria to justify whether one setting works well? For example, some people try to see if RL goes to zero with different c0 and delay combinations? is this the right way to follow?

thank you!

what i'm thinking is to use two offset shorts and one load. this will prevent using an open. i believe that will give much less percentage of error when defining the fringe parameters. the short inductance has very little effect on the effective dealy time. see if you have comment.

when you say minimum ripple, is there any criteria of how much is considered acceptable?

thanks again.

0.2 dB is a good goal for p-p ripple.

i did a simulation with an air line with diameter of 3.04mm. then i de embed the whole length of the contact to come out the very end effective fringe capacitance. after that, i input the number to get c0 to c3. the results look a lot bigger than what i can see from other cal kit coefficients.

see attached for information.

c0= -1.70 E-15 F

c1= 85511.11 E-27 F/Hz

c2= -2244.44 E-36 F/Hz2

c3= 261.73 E-45 F/Hz3

do you think this looks the right way of doing this? are these numbers kind of too big, for example c1?

thanks.

## Attachments

Can you show the phase of the open after the delay has been removed? Or maybe send an S1P file of the open (with or without port extension).

yes, the sign is negtive. the plot was done by Ansoft HFSS. however, i did input positive numbers into the calculation. if i did not remove the delay, the plot would not be that linear.

i have attached the phase plot without the delay. there are also s1p files.

how did Agilent come out these numbers in the ckt files? did you use acutal measurement? or some simulation software?

when you say 0.2 dB p-p is acceptable, do you have some procedures to do manual modifying the numbers, or trial and error process? there are at least five variables and it is very difficult to get to that goal.

thanks.

## Attachments

The 75 ohm open was modelled with a field solver,but verified with long-line measurements. It'll be afew days before I can analyze you data, but already looking at phase vs. freq for the open, it seems that you have not used a sufficiently long offset line. You should increase you offset line until the phase shift is perhaps less than 45 degrees or until someof the phase points go negative (actually possitive, and capacitance goes negative). Even if it does not follow exactly the physical lentgh, it is needed to support the simple polynomial model you are using.

i'm not exactly sure what diffence it makes by increasing the offset line. the reason is that it will have the same fringe capacitance if you have the delay removed. can you explain please?

to get RL ripple of an open as minimum as possible is the ONLY criteria to justify a good cal kit setting? if not what else do we have to try to ensure it is properly configured?

thansk a lot.

ripples on a long-line short (I recommended short, not open, as opens can have some radiation that confuses things), will give you a combination of goodness of c0 and load. If you have a bad load, it will set a baseline ripple, if you have a bac c0, it will add to the ripple.

is there any reply?

thanks.

i'm still getting strange big c1 term. see attached please.

see if you can advise to improve this situation?

thanks again.

## Attachments

## Attachments

## Attachments

And here is the S1P file if you want to compute the capacitance from the phase and align it with the polynomialopenwith141_334psdelay.s1p

## Attachments

## Attachments

can you let me know how you generated these two plots? did you de embed the delay time? or did you use port extension or something similiar?

how did you get the direct capacitance reading directly on the plot? did you input some equations?

thanks again.

Then I set the delay (under scale:electrical delay) to be 66.7ps*2 (if I used port extension, it would automatically know it is a reflectiona and do the times 2 for me, but electrical delay doesn't do that), Put up a second marker, did the same thing. And plotted the second plot.

i read my PNA help file. it seems to me that it cannot recall SNP file. is there any other way to do this?

## Attachments

see attached plots which looks very similiar to what Joel got.

does this mean that i can get a few points at the curve and fit the polynomial and solve it to get coefficients of c0, c1, c2 and c3?

thanks again.

## Attachments

i just set the delay time 141.00ps, and plot the way as you did. attached is the calculation result.

these numbers look normal to me. don't you agree?

is this the right or authoritative way to come out the coefficients?

## Attachments

I forgot to ask: what is the reason you are doing this. Are you developing new connectors?

We are trying to develp a RP (reverse polarity) TNC kit which is not commercially available in the market. We want to have some verification method in our internal laboratory.

I’m wondering why admittance G+jB (inverse smith chart)needs to be plotted. Not the normal impedance Z plot (R+jX, smith chart) where reactance is reflected? i thought normal smith chart reflects the imaginary part of the impedance. doesn't it apply to this case?

thanks again.

i'm a little bit confused. yes, the R+jX plot will convert reflection coefficient to an equivalent series R and C, however the G+jB will convert to a shunt conductance G (in siemens) and susceptance B (also in siemens). PNA's plot of G+jB shows unit in fF? why is that?

## Attachments

If a shunt element, the values are in 1/ohms = siemens, and if in series, they are in ohms. Thus, you have the effective shunt admittance, which is conductance and susceptance (g+jb) or the series impedance, which is resistance and reactance (r+jx). In both cases, the imaginary value is converted to the equivalent capacitance. In the case of only a capacitor, it will give the same answer. In the case where something, anything else, is in series or shunt, the answer will vary and looking at it to get a constant capacitance you must choose to view it a series or shunt.

i guess i understand it now. as you sai, in both cases, the imaginary value is converted to the equivalent capacitance.

i tried a short today, and i'm getting different result by using R+jX and G+jB mark, though the variation is not very much. which one do you think i should use as the inductance coefficient caculation? what is the reason for the difference?

the s1p file and two screen shots are attached.

## Attachments

As a pair of shunt elements, the same reflection would have a 2 ohm resistance in parallel with a 24 ph inductor. Throw the values in a simulator and test for yourself.

when i try to check why there is high resistance, i found a mistake where when i generated the s1p file, i did not set the material of outer contact to perfect conductor, instead i set it to nickel. this is why there are difference between X and B generated fringe parameters because of higher resistivity of nickel.

i changed the material setting and recalled it in PNA and generated the plot as attached. both X and B plot has the same calculated inductance value now. i also used excel to solve L0 to L3 coefficients and it is showed in the attached.

these numbers look normal to me. do you agree?

## Attachments

Also, because of the ripple, I wonder if you have the impedance of the line set to exactly Z0? Did you include the skin-depth effect when computing the diameters? The ripple, while small, indicates the impedance may be very slightly off.

## Attachments

I used the approximation equation of Z0=138/(e)^1/2*log(d1/d2) to get 50 ohms impedance. When plugging in the numbers, I got 49.987 ohms. I also set both outer and inner contacts with perfect conductor in Ansoft HFSS. So there should be no problem of skin depth of R in the impedance equation. Also the conductivity between the outer and inner conductor is negligible. I thought the 49.987 ohms is good enough because in reality, you can never get precisely 50 ohms considering the physical tolerances.

How much does this ripple affect the calibration? As you said the total phase shift is very tiny, less than 0.05 degrees. The peak to peak ripple is extremely small. I guess this would not have much impact on the fringe coefficients determination. Do you agree?

My understanding is that it can be set at a rounded number of 17.5 ps and this will adjust the fringe inductance coefficients, and this will be as good as it is according to my understanding of what said in the thread. Is this the correct interpretation?

## Attachments

yes, you are right. the impedance was not precisely 50 ohms. it is 49.987 ohms according to the approximation equation for coax transmission line with ideal perfect conductors for both outer and inner contact.

i'm curious on how you know that the impedance is off by 0.01 ohms? how did you come to this conclusion from the factor of the ripple on the plot?

thanks again.

However, if there is an impedance mismatch, it is very common to see such phaseing in and out, as the electrical delay gets long. So I simply loaded up a into ADS your inductance model, and created the curve for phase that matched your model (which I already was suspect of, as the DC value was negative which defies logic and indicates an overall misapplication of the inductance model). Then I changed the inductance to 0 and the impedance to 50.1 ohms, and saw and inverse ripple, from which I inferred that the impedance was less than 50 ohms. I then tried 49.95 ohm, the ripple had the correct shape but was too large, and so I continued iterating until I was within 0.01 ohm and obtained a similar ripple, that being 49.99 ohms (Note, an earlier post had listed 44.99 in one spot, that I just corrected).

By the way, this is almost exactly the process I went through to develop the 75 connector open model for the 8752A, about 15 or 20 years ago. It was the first application of finite element analysis to a calkit modeling process.

I also modeled my expectatoin for minimum and maximum capacitance based on manufacturing tolerances: minimum is for shortest center conductor, smallest diameter center conductor, largest diameter outter conductor. Opposite for the maximum. This gives a max and min gamma phase. The model should choose a value for capacitance in the middle, even if it is not nominal, so you get the smallest worst case deviation. And the artan of the phase gives the residual sourcematch of the cal kit as well as residual phase deviatio for transmission tracking. Residual source match also needs to add in worst case load return loss, or course.

this is very helpful. there are two places that i did not fully understand. you said,

1) ... and created the curve for phase that matched your model (which I already was suspect of, as the DC value was negative ...

what value of DC is negative make you think ...an overall misapplication of the inductance model..?

2) ...And the artan of the phase gives the residual sourcematch of the cal kit as well as residual phase deviatio for transmission tracking...

what does this statement mean exactly? what is artan of the phase?

my understanding is that the transmission line before the open or short happens has to be well optimized to have minimum reflection. otherwise the phase plot (with mark function, delay) would have high peak to peak ripples which does nothing good to either the curve fitting or residual sourcematch error. is that the right understanding?

thanks again.

in this case, does the curve fitting still work well? if not, do you think data based characterization is better to do this?

in one of Agilent application note, it says,

"The data-based standard model is a new feature in Agilent’s network analyzers. It allows a calibration standard to be defined by a data file that contains frequency data, S-parameter data and uncertainty data. The data file may be created using actual measured data from a reference metrology laboratory, model data from device modeling software or combinations of both. See Appendix D on page 37 for details on data file formats."

do you have any comment on this?

thanks again.

## Attachments

I don't know if I understood your extraction but you have a value for L0 of -3pH, it should never be negative if it is a realizable short. THis is a result of trying to polynomial match the ripple. You have a bad value for L0 that is over come (except at very low frequencies) to give a nearly correct value of L overall frequencies. But a strong clue that the inductance model is wrong. Mostly, the principle contribution should come from the first term.

2) I should have said "arc-tan", sorry. take Arctan of the phase and you will receiver the magnitude error that caused the phase shift.

Hard to say for sure, but you can do a similar measurment to find out...but to correct for this and the other error, if the impedance of the open delay offset is not 50 ohms, then change it in the model for the open (or short) to represent the true value. Then the ripple we be there as a result of the proper understanding the delay line impedance, and not some model-matching of the L or C terms to accomodate the error caused by the offset impedance.

At the end of the day, it doesn't matter how the model is generated, just that the phase which it returns matches the phase response of the standard.

sorry that i might have not asked clearly, my question was,

as you know there might be some mismatches before the actual open/short happens, such as unpreventable meachnical support or sometimes the interface itself, while the rest of the transmission line is still 50 ohms, in that case does the curve fitting still works well?

or otherwise we can use data based standards characterization? or say is this a better approach to this problem?

thanks again.

i tried to recall a s1p file. it is attached here. the plot frequency range is weirdly up to 28.9741GHz. the s1p file was made up to 6GHz by Ansoft HFSS.

is this a bug? my PNA FW is A.09.33.09.

can you help look into this?

thanks in advance.

## Attachments

Looking at the long line ripple after cal is only partial validation of your model, the source match. The other part you need is a way to evaluate reflection tracking. Having calibrated with offset standards, you can then attach a flush short and look at reflection tracking directly. Set up 2 traces, one in magnitude and the other in phase. Add 180 degree offset to the phase trace because a flush short should be a reflection of 1 at 180 degrees. So perfect reflection tracking would be flat lines on both traces.

Match ripple approximations from peak to peak in dB: 20*Log10(ABS(1-(10^((pkpk/2)/20))))

pk-pk ~source match in dB

.1 44.7

.2 38.7

.3 35.2

.4 32.6

.5 30.7

Drew

when you say "attach a flush short", do you mean a short which is at the same plane as the reference plane? if yes, what about those interface where there are PTFE insulators? in that case, the flush short cannot be realized due to physical constraints.

is this the right understanding?

Drew

if the PTFE does extend the reference plane, then in reality a flush short cannot be made. what do you mean by "the flush short will compact it back in"?

I understand what you are saying. i meant the interface by definition has PTFE extended beyond the reference plane, such as TNC, BNC and MCX. in that case, since you cannot get a flush short how do you check reflection tracking?

aslo, can you please answer my question i asked in the post of Thu Jul 07, 2011 2:08 am, where i asked why the plots look weird?

thanks again.

I upgraded the FW to A.09.42.01, and recalled it again. the problem still exists. see the attached please.

can you help review this and see if there is any other possible issues which caused this weird looking plot?

PS. i did not open the file so there is no extra space or something unnecessary added.

## Attachments

any comment?

I also tried it on A.09.42.01. No problem at all. Maybe you could try downloading the file from this forum website and try opening it again, instead of using your local file.

this sounds weird to me. i did download it from the forum.

besides the frimware, is there any other factor that might affect PNA's recall function?

## Attachments

i'm afraid this really confused me. i noticed you have added ".0" in the file. i recalled your Mod_OF.s1p, the problem is still there. see the attached plot.

what is your PNA model number? does it affect?

## Attachments

I have seen some problems in the past with languange settings affecting the way files are read. I can only say that I read the file in just fine when I tried it last time. Curiously, I think I did see a problem the first time I read it, but I cannot repeat that issue.

Other ideas are: maybe it is something with the size of the file and all the comment lines in the file, which is unusual for an S2P file (comments are typically not inserted in the body of the S2P file, but at the beginning).

It cleaarly appears related to the end of the file, as though the last line is being read in and a new line is created with a frequency matching the last imaginary number. Leads me to believe something is reading in the numbers incorrectly in your case, but as I say, I can't reproduce it. maybe someone else can chime in.

## Attachments

I did a trial where I did not select "Include Gamma and Impedance Comments” when doing data export from Ansoft HFSS. Everything works well now.

I have been reading the Touchstone file format reference material. Here is what I found from the web.

http://vhdl.org/ibis/connector/touchstone_spec11.pdf

But I cannot find description on what gamma data are used for. Do you have any idea on what gamma data is used for? Or what are they?

In other words, if there is no gamma data in the s1p file, is there any impact on its recalled result?

## Attachments

I assume you are saying it dosn't affect the recalled result.

I used the method to generate all fringe parameters and generated the ckt file. I then tried with a pair of adaptor to verify if the kit I have worked on works ok or not.

I have got the weird looking S21 or S12 plot. there are many ripples in the plot. some peaks even go above zero.

do you know why this happens? what can i do to improve this?

## Attachments

It could be a loose or unstable cable during the calibration. The other cause is poor residual source match and load response in the cal kit.

THe ripple pattern has a p-p of about 0.2 dB, indicating a combination of residual source match and raw load match of about -40 dB. If the raw load match is on the order of -15 dB, that means your residual source match plus load match together is on the order of -25 dB, not too spectactular.

You can look at the raw source and load match directly using the calset viewer.

I understand you said there are two patterns of ripple, and long and a short cyble ripple.

I have done this test a few times, every time I have the same reult. so a loose or unstable system cable during calibration can be excluded.

when you say residual source match and load response, do you mean the load is not ideal and have some return loss?

how can i improve this?

thanks again.