Convolution in frequency domain?

Dear IshanMishra,

IshanMishra said:

so is there a FFT-like method where we can speed up the simulation time by multiplying the frequency response of the input and s-param file and then taking the inverse FT of this multiplicaiton to obtain the time-domain response.

If you are trying to maintain the same level of accuracy and time resolution in the response of the S-parameter based subcircuit to the PRBS input sequence, there is no "magic" in believing that the computations in at frequency based solution will be less than those in a time based solution. There is an equivalence in the time and frequency domains and the need for small time steps equates to the need for very large bandwidths.

Further, unless I am misunderstanding your post, your use of an analogy of an "FFT-like" method does not relate to your basic problem.The FFT is algorithm is a "divide and conquer" algorithm used to solve the DFT in O(NlogN) time in lieu of O(N^2) time.

The issue you are facing is a common one in numerical simulation and is often referred to as a "stiff" set of differential equations. There are other means to try to expedite finding the solution to these types of equations. If you can provide more details about the nature of your input stimuli, your network modeled by the S-parameter file and your goal, maybe I could provide some thoughts.

I am sorry I can not be of more help. I am sure my response is not what you were hoping to read! Perhaps others who have greater expertise than I and monitor this Forum can provide a better or more helpful response IshanMishra.

Shawn

Using an nport with the interpolation method set to "linear" (or "spline" - the difference is just how the interpolation is done; the method is the same) uses a convolution-based approach for modelling the s-parameters in the time-domain. If that's slow, then it may be because your s-parameters are poor quality or non-physical or non-causal.

It's rather hard to give an answer without more detail, so perhaps you should contact customer support?

Andrew

Using an nport with the interpolation method set to "linear" (or "spline" - the difference is just how the interpolation is done; the method is the same) uses a convolution-based approach for modelling the s-parameters in the time-domain. If that's slow, then it may be because your s-parameters are poor quality or non-physical or non-causal.

It's rather hard to give an answer without more detail, so perhaps you should contact customer support?

Andrew