While S parameters can be quite powerful tools in simulation and analysis, they are also full of stumbling blocks. There are various things that can go wrong in the creation of S parameters, especially for broadband (time-domain) simulation. Check that the exported, combined S parameters are still causal. You might need to specify more data points or wider frequency sweep in the export of the S parameters.
I recommend using S parameters only when necessary. It makes little sense to use an S-parameter model of a transmission line. The T-line model does suffer from the weaknesses of the S-parameter model, such and bandwidth limits, frequency sampling points, passivity, etc. S-parameter models make sense for truly 3-D structures, parametric analysis, or maybe IP obfuscation.
Thank you for your response Weston.
In this particular channel, all the traces are not s-parameters. Only the via models are s-params
extracted from the Hyperlynx 3D field solver.
The circuit simulation runs fine and as expected. Via models pass passivity checks and are
I've had great success doing this before, but all the via models were generated from
Ansoft HFSS, so I expect it has something to do with the generation of the via models from the
HyperLynx field solver since its the only thing different? When the s-param models are placed in the circuit (none of the transmission lines
are s-param models) the simulation results are as expected.
When we export the circuit (with the via s-param models and line sim transmission lines)
and simulate with that new exported model, the result is different from the circuit.
We are trying to pass an s-param model of a circuit segment to a client of ours, but are finding
we may need to go a different route.
Is there anything else you could suggest we might do using the 3D field solver (since it appears to be the
only difference) ? If not, we may change strategy to generate the model, or go back to HFSS for the s-params.
Not quite sure why the exported model of our circuit would be so different. We're expecting the same behavior
using the s-parameter model as with using our linesim circuit.
I understand that the problem is in the final S-parameter export. This is the step that I was talking about. You might need more frequency points or wider frequency band for the final S-parameter export.
The same issue applies to the difference between HyperLynx Advanced Solvers Full-wave Solver. The accuracy of the field solvers is comparable. I suggest that you look at the differences in the default settings for maximum frequency and number of data points for the S parameters that the tools write. Using the same frequency settings will create very similar S parameters from the two solvers, and hence will create very similar effects in simulations.