implement random freq domain transfer function

Kevin,

The trouble is that arbitrary frequency-domain transfer functions might be non-physical and lead to causality problems in the simulator (this is one of the challenges with s-parameters that are non-physical - they can end up describing non-causal behaviour and lead to convergence difficulties in the time-domain). It would probably be better to do some kind of fit (maybe rationalfit in MATLAB?) to convert to a form that can be represented with svcvs.

There is also "fracpole" available in Spectre which was an implementation many years ago to help model fractional poles (e.g. for modelling empirical fits of skin-effect). See "spectre -h fracpole". Not necessarily what you want, but just some thoughts on related topics.

Andrew

If you have access to MATLAB, you can generate a pole/zero equivalent of a transfer function using 'rationalfit'. You can choose the order and see if the results are accurate. 

You can also convert the transfer function to s-parameter (with standard terminations) and use nport with those s-parameters. 

Depending on the frequency range over which you have the data(if the data has not rolled off to insignificant values at the end of the range), you can have problems with time-domain simulation(like the ones mentioned by Andrew Beckett).

Nagendra

(I posted this a while back, but looks like it was dumped as spam).