Looking for SPICE model for coupled lossy transmission line

[buenos]

Looking for SPICE model for coupled lossy transmission line.
This would have to work in standard berkeley spice-based simulators without special keywords specific to Altium or LTSPICE.
It has to be lossy, with dissipation factor, or with dB/in.
It has to be coupled, aka differential pair, with even and odd mode impedance.
Is there a free spice model file that I can download, or someone can paste in here in a comment?

 

[pancho_hideboo]

Use https://www.sp.se/en/index/research/emc/Sidor/LC-Calc.aspx

 

[dick_freebird]

I'd be happy to find just one tline .model that seemed credible.

 

[pancho_hideboo]

I'd be happy to find just one tline .model that seemed credible.

”TLINE” is lossless model.
So I don’t think it is possible with Berkeley Spice.

 

[FvM]

The question is less clear than it may seem at first sight. You can easily make a LGRC lumped transmission line model, also coupled lines can be modelled by combining multiple transmission line models.

But LGRC model doesn't represent a lossy line with realistic behavior, except for small band, which is a more or less trivial problem. So I would ask first about the purpose of your simulation.

 

[buenos]

The purpose is high speed channel simulation with QUCS, to generate S-parameters.
The frequency dependent loss and the tline delay are important.

 

[FvM]

If you want a S-parameter model, why are you asking for a SPICE model? A simple behavioral model with constant delay and sqrt(f) proportional losses will fit a real transmission line much better than most lossy SPICE models.

See previous discussions, e.g https://www.edaboard.com/showthread.php?265991-transmission-lossy-line-for-long-Cable

 

[volker@muehlhaus]

The purpose is high speed channel simulation with QUCS, to generate S-parameters.
The frequency dependent loss and the tline delay are important.

It's not that simple to replace missing models. The commercial simulators (e.g. ADS or Microwave Office) have a more complete model library.

 

[buenos]

I want to generate an s-parameter output and display of a complete channel. One of the several elements in the channel is a transmission line with user controlled parameters (length, loss, impedance).
I know ADS is better, but I was trying to make a complete simulation flow using free software.

 

[volker@muehlhaus]

Sure, anyone would prefer not to spend 100k$ or more on that EDA software. The only issue is that there is no simple solution to that. The more complex line models are proprietary, and possibly a technology mix of static EM solvers and circuit models. If you are a true expert in this RF modelling field, you might be able to find your way using free software, but it will take much more time.

 

[buenos]

Actually I built a model in QUCS. 2 half length transmission lines, and an RF equation device in between with the transmission equations being K*DF*f/f0.
Not sure if it's correct modelling.
I cannot put this into a subcircuit, due to a bug in QUCS causing error. So I have to copy over a block of 3 components and wires representing one transmission line, and feed parameters to all 3. Then I have up to 8 of these blocks in one channel, it is a mess.
For now all I need is a linear loss curve. The exact loss number (at 10GHz) is obtained from the PCB fab (previous test coupon measurements), so I dont have model copper roughness and other things separately.

 

[volker@muehlhaus]

2 half length transmission lines

Is the lambda/2 what you have in hardware? It's a magic length because at that exact length, line impedance doesn't matter and there is no impedance transformation by the line, regardsless of Zline. So depending on your actual length in hardware, because this might be a over-optimistic assumption.

and an RF equation device in between with the transmission equations being K*DF*f/f0.

What are you trying to model? Coupling between the lines is both inductive and capacitive, so you can't model it with one lumped path between the lines. Can you show a screesnhot of the testbench?

The usual and reliable lumped representation is cascaded RLCG segments of lambda/10 length. For 2 coupled lines, you then have RLCG for each line, and additional set of C and inductive coupling between the lines. Some EM simulation tools generate such output from field simulation data, for use in SPICE simulators.

For now all I need is a linear loss curve. The exact loss number (at 10GHz) is obtained from the PCB fab (previous test coupon measurements), so I dont have model copper roughness and other things separately.

What do you mean by loss, the dissipation only or the overall signal integrity stuff? I'm asking because reflection and coupling might be bigger issues than losses (dissipation). We have guys here with all the EDA tools and experience to analyze such structures, so maybe you describe your configuration in more detail and someone runs a quick analysis. That would at least give you some reference data for comapring your own results.

 

[FvM]

For now all I need is a linear loss curve.

Typical transmission loss is proportional to sqrt(f) rather than linear in f. It can be well modelled in a behavioral H(jw) or s-parameter description.

For the line coupling, a multi-transmission line model may be useful.

 

[buenos]

What I was trying to do is to make a channel model that i can change easily.
Instead of 3 parts per tline each with lots of parameters, I would have one tline with few parameters (length, imp, loss). Just like in ADS.
I am using coupled nonlossy lines with a equation device to add the loss to it.
Half length because i want both ends of the sub component to represent impedance correctly towards other elements in the channel. So i sandwich the equ device between 2 halves of the tline. nothing to do with lambda.