convert an S2p file to spice output file (lib )using Python

[ftn]

can someone help me and tell how to convert an S2p file (network analyzer) to an spice file (LIB file)
using python

thank you View attachment Datei.rar

 

[volker@muehlhaus]

This isn't trivial, because S2P is behavioural frequency domain data. You will need some software that does equivalent circuit model fitting.

https://www.sonnetsoftware.com/products/sonnet-suites/bb_spice.html
https://www.cadence.com/products/sigrity/broadbandspice/pages/default.aspx
https://muehlhaus.com/products/equivalent-circuit-model-fit-for-rfic-inductors

 

[dick_freebird]

They are fundamentally different things. More likely you would
pick a SPICE primitive that functionally matches the device,
and then -fit- the model parameters to get the right S param
behavior.

The VNA knows nothing about topology and Python would
have to take a guess or be educated, neither of these a
small matter.

 

[dick_freebird]

I do have a vague recollection of seeing a couple of s-param
behavioral blocks down in one of the Cadence standard libs,
back when I was using that tool set. In this case you might
be able to just push the VNA values into their appropriate
form-fields. But this is not SPICE, it's part of schematic entry
libraries and probably ends up at a veriloga module of some sort
I expect. Which Spectre is fine with, but some SPICE dialects
may not be.

 

[FvM]

You probably didn't notice, but ftn appended an example lib component comprised of behavioral controlled sources, defined in frequency domain by a FREQ statement.

They are essentially one-to-one conversions of the s2p file and in so far don't involve complicated operations.

But the usability of frequency domain models with only a few data points is very limited and even with a reasonable data point density, you shouldn't overestimate their performance. In time domain anylysis, the models are evaluated by a fourier transform which doesn't work well over a large frequency respectively time scale. So yes, frequency domain models have a purpose for devices that can't be easily represented by functional models, lossy cables are e.g. an example where I have already used it. But in most cases, parameter extraction for functional models is preferable.

 

[volker@muehlhaus]

You probably didn't notice, but ftn appended an example lib component comprised of behavioral controlled sources, defined in frequency domain by a FREQ statement.

Thanks, indeed I missed that.

My experience with that simplistic approach: it just doesn't work well. Even if you have many data points, and even if you go low enough in frequency to have something like a "DC" behaviour included, results for transient analysis are poor. And from what I remember, it really slows down the analysis.