Hello everyone,

I want to do a transient simulation with S parameters - the reason is that my circuit critically depends on passive structures modeled by an EM simulator such as Sonnet or IE3D. Generation of equivalent RLC models for the structure is often very difficult and time-consuming. Unfortunately, Spectre's convolution method for generating a time domain model is simply not good enough - it produces strange unphysical results. The frequency coverage in the S parameter file is pretty good (down to 1 MHz) - we have tried providing a DC path using DC blocks, adding a line at 0 frequency in the S parameter file etc. We filed a support request with Cadence but it doesn't seem to be getting any better.

Another area where transient with S parameters is present is harmonic balance with initial transient assist. The problem with Spectre here is that even if it is able to find the convolved time-domain model for the initial transient assist, it continues to use that model even when the harmonic balance step starts instead of just referring to the S parameter file.

Does anyone know of a scaleable broadband (or even narrowband) RLC model generator given an S parameter input file? My structure has between 8 to 50 S parameter ports. How has your experience been with transient simulation in circuits containing S parameter blocks using the simulator of your choice?

Thanks,
mohta

Hallo mohta,

I am in a very similar situation like you beside that I use ADS+Momentum and have 50..80 ports.

The generation of RLC models by hand is for this number of ports not possible.

I have as well problems with the convolution simulator (of Agilent).

My s-parameters are also fine and quite well to the measurement.

A added sometimes also a line in the s-parameters for provide a DC response and it helps a lot.

I also issues support requests (with Agilent in my case) and one is ongoing, but in past there was not the right help.

One of my applications is also Transient-Assisted-Harmonic-Balance.
But I can't tell whether Harmonic-Balance uses s-parameters as it should of the convolution impulse model because in my case convolution still stucks.

ADS offers narrow and broad band spice model generators (partly special licenses are required).
Most of the models are frequency depending functions implemented as macro models.
That means they need as well the impulse calculation for convolution.
Agilent claims to correct for passivity. So it might be a help for you.

Further, some people of the ADS support team recommend not to use the broad band model generator but to use convolution directly as it should not be necessary to use the model generator. The convolution simulator contains by now an option to correct for passivity.

However, we the size of our number of ports one gets easily memory and hence resolution problems.
With 50x50 ports the matrix contains 2500 elements.
The impulse calculator for convolution samples the frequency range and in my case even 1024 samples (3 GHz in steps of 2,9 MHz) give a insufficient match.
That gives in total 2.5 Mio values in the time and frequency domain.
And now the software can do some FFT and IFFT on it which runs in my case out of my 2 GByte Windows XP limit.

If somebody has ideas or comments on the whole topic, please let me know.

Wolfgang Doerr