微波EDA网,见证研发工程师的成长!
首页 > 研发问答 > 微波和射频技术 > 电磁仿真讨论 > Re: 3D Simulator

Re: 3D Simulator

时间:03-23 整理:3721RD 点击:
Hi itaifrenkel
e.g. I simulate a dual-frequency microstrip antenna in HFSS, resonance frequency are at 2.45Ghz and 5.25GHz.
To get the patterns at two frequency, I must simulate twice at two points respectively.
Thanks!

Dear asdfaaa

in MWS, parallel processing on several CPUs is already implemented for CPUs on the same mainboard. As far as In now, it works pretty much the same way computing in a cluster. Each processor runs (simulates) a smaller part of the total structure. There is no general problem in the FIT Theory which prevents you from doing parallel processing.

It is more a technical issu. For parallel processing for CPUs on different mainboards (cluster) you have to use special toolkits an techniques. It simply has not been impelmented so far in MWS.

Best regards.

F.

Dear Loucy,

Thanks for thinking of me, yes you are right. I learn from a WinNEC95VM and I'm using a SuperNEc demo version. These two programs are useing wire modeling. Eventually, I want to simulate more complex structures so, which software can I investigated on, I want a software good enough to simulate a variety of antenna? Maybe in the futur, I need to deal with frequency dependent materials.
Another question, WinNEC95VM is a frequency domain simulator, so, if I want to transform it in time domain, I know I have to applied a Fast Fourier Transform. But, how can i do it? I have Mathcad and the FFT I have no problem. How many frequencies should I sweep to do a FFT? Take a simple example, a simple resonnance dipole, can you help me on it , can you provide me the steps to achieve this?

Thanks,
Mica

In my opinion,

parallel processing on >1 CPUs on one motherboard =>(indicates) the computing engine is multi-threaded, the data/memory is mostly likely shared by the processors. There is only 1 computation domain, yet the (FDTD) updating is performed by several CPUs. The program is relying on the operation system to "balance" the task. The limit in the size of the problem remain the same as in 1 cpu case.

cluster computing such as the one implemented in RSoft FullWave is based on MPI. the computation domain is decomposed into many sub-domains, the fields are stored(distributed) in separate computers. The FDTD problem can be much larger. It seems CST has not offered this feature yet. They would need to handle the load balance in their own code (in contrast with relying on Windows to schedule the threads).

It looks like XFDTD of Remcom also has the capability to use cluster computing. I am not sure how well it scales.


lets hope CST adds cluster computing in the near future

If you want to see the scaling of simulation speed with the number of nodes in a cluster computing environment check the following link http://www.lumerical.com/fdtd_parallel_fdtd

EdmundZheng,

In your case , you can adapt on the higher frequency (lets say 5.5Ghz) and then use fast sweep to cover the range up to the lower frequency.

There is nothing special that requires more than one adaptation and one sweep.

Itai

I noticed a couple of the original questions were not answered.

A full-wave solver is simply one that solves the full set of Maxwell's equations. Whether the geometry input is 2D, 2.5D, or 3D does not matter. For instance layered media simulators like Sonnet and Ensemble are full wave since they solve the full set of Maxwell's equations even though the allowable geometries are limited. Momentum RF is not full-wave since it assumes that the potentials are independent with respect to frequency.

Bounded vs. unbounded can be tricky. For most people it does not matter. Bounded solvers may not allow surface waves and other parasitic modes to propagate. For instance vias are very good at exciting a parallel plate mode; it's good to know when this is occuring. If the bounded solver has a good PML, I don't think this is a problem. Standard radiation boundary counditions do not work as well for these type of modes (because these are guided modes not radiation).

Hope this helps.

Thank you for your reply!
Fast sweep can get s11 or vswr, however, it can't get the pattern at two frequency points respectively.

EdmundZheng,

I have to check it, but as far as I remember,you can do the following:

1) Use fast sweep with Save Fields checkbox. Then you should be able to draw patterns.
2) Instead of fast sweep you can choose a one frequency discrete sweep. That means that you are re-using the higher frequency mesh, to solve the problem at a specific frequency of interest. So, there is no need for the adaptation process.

Itai

Copyright © 2017-2020 微波EDA网 版权所有

网站地图

Top