AnsweredAssumed Answered

Optimal FEM padding for broadband FEM simulations (EMPro, ADS)

Question asked by bartboesman on Jun 2, 2017
Latest reply on Jul 24, 2017 by RyanA

Hello all,


Although the FEM solver in EMPro computes a solution in the frequency domain, it is in some cases more opportune the use the FEM solver for broadband EM simulations. When doing so, Empro by default generates a mesh at the highest frequency of concern, unless specified otherwise. Whereas this in general leads to accurate results at the highest frequency end, I have often noticed significant deviations or even unreliable results at the lower frequency end. It is easily understand that in such case, the adaptive frequency sweep will try to fit a set of unrealistic S-parameter data, consequently leading to an excessive number of computed frequency points. 

After several iterations, it has been noticed that the discrepancies observed at the lower frequency end can be reduced by increasing the FEM padding, such that the latter not corresponds to 1/2*wavelength(maxFreq), but rather to 1/2*wavelength(minFreq). We should remark here that the above observation is true when the simulated structure radiates significantly in the entire frequency region of interest (although we should perform additional simulations to make strong statements about the counterpart). Nevertheless, this means that for accurate broadband simulations, the computational efficiency decreases due to the excessively large computational domain (depending on the lowest radiating(?) frequency of interest) in combination with the small mesh cells (depending on the highest frequency of interest). As I see it now, the best approach is to split the entire frequency region of interest into multiple smaller frequency regions, which are simulated successively in separate simulations, each with their own optimal mesh.


In ADS, the story is rather similar, although additional simulations are needed to make strong statement on the similarity between the observations in ADS and EMPro. Nevertheless, I believe that I can say that the observed results will probably correspond very well. Remarkable, however, is that in ADS, the FEM padding by default has fixed values, namely 3.125mm for the LATERAL extension of the substrate, and 5mm for the VERTICAL extension of the substrate. This means that ADS assumes, in case the FEM solver is selected, no frequency dependent padding is necessary.


Both the observations in EMPro and ADS cause me to open the discussion on the optimal selection of the FEM padding parameters, as well in EMPro as in ADS. All comments or remarks on this are very welcome.

Best regards,