What range of frequency should I select for EM simulation to get S parameter?

I'm currently doing EM simulation by HFSS on stripline to get 4 port S parameter data. The data will be used in a transient analysis to simulate crosstalk. The input voltage will be step voltage with rising time of 1ns. I'm curious about what range of frequency should I select to get a right result.
Thank you for your answer in advance. :)

Thanks a lot.! Is there any numerical way to get the value?

From DC to more than 10GHz.

See https://www.edaboard.com/showthread.php?377517#2

1 ns risetime is corresponding to about 350 MHz bandwidth. Thus 1 GHz simulation bandwidth will give a good estimation.

Wrong.

For time resolution of 1nsec/10, system bandwidth is 10GHz.
In convolution, FIR has to be evaluated for delta_t less than 1nsec/10.
This means inverse FFT requires frequency range from 0Hz to 10GHz.

If you use other model than convolution in transient analysis, e.g. rational approximation model, you can limit frequency range for both fmin and fmax.

I think FvM is not wrong. In order to save simulation time and disk space you can do some approximation. That is:

rise time of 1st order (RC like)
taking the 90% of energy of the signal can be enough

under these conditions we know the rise time can be modeled as Vrise/Vmax = 1-exp(-t/RC)
defining now the rise time as 10% to 90% of Vmax (the final voltage) we will have

0.9=1-exp(-t1/RC)
0.1=1-exp(-t2/RC)

solving for time and combining the two equations rise-time≈t2-t1=2.2*RC
also we know BW=1/(2*pi*RC) then:

BW≈=2.2/(2*pi*rise-time) from which

BW≈0.35/rise-time or BW(MHz)≈350/rise-time(ns)

Using this method you don't reconstruct perfectly the shape of the rising edge (to do this you have to use the pancho_hideboo method), but I think you can get result accurate enough for the cross-talk modelling.

Thanks you all for answering the question! I learnt a lot! I already have a data on 10GHz and understand that is the right range now. Thanks again!