Ansoft Planar EM Vs. Sonnet em ???

[JimR]

ansoft comm diff

I'm implementing very simple, single trace, geometries in Ansoft and Sonnet tool sets. For example a 10x1mm through-line microsrtip with a standard edge port on both ends. When I generate Y and S-Parameters and compare, I get results I don't understand. The frequency resonance and curve shapes are similar (in fact the same) but the magnitudes differ by as much as 20dB. I'm thinking that I don't understand the de-embedding and/or port normalization concepts.

1. Can anyone offer simple (even single trace) designs that give matching results between Ansoft and Sonnet software?

2. Can anyone shed some light on why these Y and S-Parameter results can differ in magnitude so significantly?

Thank you
Jimbob

 

[rautio]

ports sonnet em

Hi Jimbob -- My guess is you made an error entering the geometry in one of the two tools. Do a single thru line in both and compare with circuit theory. If you don't have a nodal analysis tool, use the net list nodal analysis tool in Sonnet. Results should be pretty similar. For high accuracy requirements, Sonnet will be better (I work for Sonnet), but major differences are likely just a mistake on input. The two results should be reasonably similiar to each other and to circuit theory.

 

[JimR]

ansoft+effective+dielectric+constant

In Sonnet I simply created a 10x10mm box, placed a 1x10mm rectangle, then added ports to both sides. I then confirmed that the dielectric substrate parameters are that same as used by Ansoft. Geometry entry into Ansoft was equally trivial. Is there something more that should be investigated?

When you say compare with circuit theory. (Of course circuit theory applied to microstip is always an approximation, right?). The way I'm thinking this would be done is to use an estimation for the "effective" dielectric constant and characteristic impedance for the microstrip line (i.e. a qausi-TEM approximation as supplied by AppCad), then calculated the S-Parameters of this transmission line terminated in 50 ohms. Is this anywhere close to what you are suggesting?

Your assistance here is Greatly Appreciated!

 

[rautio]

ansoft theory

Hi Jimbob -- Look at the Zo and Eeff calculated by Sonnet. Then put a netlist in Sonnet (will give you directions if you want, or just check out the manual), and analyze the S-parameters of a line with that Zo and Eeff.

If you use the same dielectric constant everywhere (for example, Erel = 1.0, i.e., air everywhere), now Eeff had better be the same as Erel.

If you want line dimensions that you know to be close to 50 Ohms, use 72 microns wide on 100 micron thick Erel 12.9 (GaAs). S11 had better be close to zero.

 

[JimR]

Here's maybe a dumb question concerning this approach. The Zo has little variation from 1-10GHz yet the Y,Z,S parameters have a deep resonance at 8 GHz. Doesn't this represent a contradiction?

Thanks again for your replies...

 

[rautio]

Hi Jimbob -- By deep resonance, I think you mean S11 goes to zero (-infinity dB or so).

It is a problem if you are dealing with a 50 Ohm line. S11 should be close to zero everywhere, not just at one frequency.

If Zo is not near 50 Ohms, then S11 will go to zero or have a maximum at every frequency where the line is a multiple of 1/4 wavelength long. This is because S11 is determined by two things: 1) The reflected wave from port 1, and 2) the reflected wave from port 2 making its way back and coming back out port 1.

When the two reflections are in phase, you have a maximum in S11. When the two reflections are 180 degrees out of phase, you have a minimum.

Check it out on a circuit theory program, you will see the same thing.

This is a nice trick to remember if you have a very long length of transmission line (like coax in a community cable TV system). If you know the cable is cut somewhere, but you are not sure where, just look for the mininums and maximums of the SWR as a function of frequency.

 

[JimR]

Thank you, that does make things clearer. Following your explanation I derived an equation for the "generalized" reflection coefficient for a three-layered medium (based off of "Waves and Fields in Inhomogeneous Media" but converted to transmission line parameters). I then wrote a small program in Mathematica to accept the effective dielectric constant and Zo generated by Sonnet. When I plot the reflection coefficients they agree exactly with the plots that Sonnet is giving! The analytical formula helps me to see the cause of the shape of the curve; I'm starting to see the light.

Unfortunately there still remains some differences between what Ansoft generates in the low frequency region. Ansoft's S-Paramters get a little squiggly near the low frequencies. Its confusing what could be causing this since their Z and Y parameters don't show this low frequency abnormality. I was thinking that they are probably generating the Y parameters from the numerically calculated current and then converting this to S-Parameters. If this is in fact the case then their conversion from Y to S parameters is adding something strange at low frequencies that I can't determine? I'm also wondering which one is right, Ansoft or Sonnet?

Thanks again for the clarification; it really does help.

jimbob

 

[rautio]

Hi Jimbob -- If you are looking at a length of uniform transmission line at low frequency, i.e., the length is less than 1/4 wavelength at the highest frequency, then there should be no little squigglies. I would guess that the little squgglies you are reporting are either numercial error or de-embedding error. Unfortunately, all we can do is guess. Sonnet is the only EM company that publishes their full port calibration algorithm as far as I know. (If anyone knows of any others that do full disclosure on this matter, feel free to let us know.) Also, as far as I know, we are the only company with exact de-embedding as well, and we can do that because of our perfectly conducting sidewalls that we use as perfect short circuit reference planes for the port calibration. By "perfect" I mean that the port discontinuity is removed to within numerical precision leaving error due to subsection size as the major source of EM analysis error.

BTW, I just got a report that one of our "competitors" (from my viewpoint, we do not have any 100% competitors) just plain can't figure out why I keep talking about error when everyone else always talks about accuracy!

 

[JimR]

Rautio - I figured out why I was getting discrepancies with some geometries with calculated characteristic impedances that differ significantly from 50 Ohms. With the Sonnet tools, the port impedances are set to de-embed to 50 Ohms. (And with Sonnet Lite you can't change this; CSULB is trying to get a complete system in place at this point. They say they have funds but the question is when).

With Ansoft's tools, although the port is set to 50 Ohms, if you don's select the "PostProcess" check box in the port configuration window, then the default process is to de-embed to the calculated port impedance. This sounds like it should have been obvious but as it turns out the geometries I was dealing with were very close to 50 Ohms. When I moved to geometries with 85 Ohm impedances then the root cause of the discrepancy became more obvious and easy to confirm. I can't determine if Sonnet has the same weird behavior at low frequencies when the impedance is matched at 85 ohms as I don't believe that I'm able to change this in Sonnet lite (it's set permanently to 50 ohms?).

Thank you once again for your replies - jimbob

 

[rautio]

Hi Jimbob -- You can change the termination in all versions of Sonnet by going to the data viewer and select Graph->Terminations. You can set each port to whatever normalizing Zo you want.

If you decide to change it, be very careful. In microwave engineering, S-parameters are almost always normalized to 50 Ohms. Anything different is a rare situation.

For EM theory types, it is common to normalize S-parameters to whatever the Zo is of the port connecting transmission line. These are called "generalized S-parameters". I think such a name is strange because there is less infomraiton in generalized S-parameters than in S-parameters for which we know the normalizing Zo.

This can cause a big problem when you get an EM theory type of person talking to a microwave designer. When they each say "S-parameters", then might be talking about two different things!

So, if you are doing microwave work, always use 50-Ohm S-parameters (with a couple of exceptions for very advanced users) and if you have to work with EM thoery types, make absolutely sure you are both talking about the same thing.

If you want to get software for education use, we have a very strong university deal, feel free to contact us directly.