Continue to Site

Welcome to EDAboard.com

Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.

3D Simulator(lumped ports, waveguide ports, plane waves?)

Status
Not open for further replies.

Mica188

Member level 2
Joined
Nov 1, 2004
Messages
45
Helped
0
Reputation
0
Reaction score
0
Trophy points
1,286
Activity points
459
3D Simulator

Hi,

I have read some threads in this forum and I found some criteria for the choice of a software.

1) what kind of excitations can be applied to the structure (e.g. lumped ports, waveguide ports, plane waves?)

Can you give me some examples of softwares on these items?


2) is it a time- or frequency domain solver : Which is better time or frequency?

3) does it solve in a bounded or unbounded domain? what type of boundary conditions are possible :

Can you explain more in details bounded or unbounded conditions?

4) what is a full-wave solver ?

I'm a beginner in this domain. Right now, I'm using SuperNec demo version and I would like to know the comparaison with HFSS from Ansoft.

Thanks in advance,

Mica
 

Re: 3D Simulator

Mica188 said:
Hi,

I have read some threads in this forum and I found some criteria for the choice of a software.

Mica

1) what kind of excitations can be applied to the structure (e.g. lumped ports, waveguide ports, plane waves?)

Can you give me some examples of softwares on these items?



The excitation is just the initial boundary condition, how do we excite the model decide which kind of ports we can set to the model. You can find more information in HFSS full manual.


2) is it a time- or frequency domain solver : Which is better time or frequency?


It is very hard to say which one is better or not. Befre we choose which solver, we must know what is our simulation objective, and which parameter we are interested.

For time domain solve, it is fast to get the time domain response, once the time domain response is known, after Fourier tranform we can get the frequency domain information, but since another numerical transformation is needed, so the computational cost is increased in and the accuracy is decreased.
For frquency domain solve, it is very convenient for us to get the frequency domain response, and it should be accureate enough. But the problem is what can we do if we need a very wide frequency band information, the only choice is to run simulation lots of times to get the response at each frequency point, after that, we can use Fourior transform to get the time domain response, but the computation cost is higher than the time domain solver.

so, choose which solver, should base on which parameter we are interesting.
For antenna, filter, such narrow frequecy band device, both solvers should be ok, you can use what ever you want. If you have very strigent requirement of the simulation accuracy, then I prefer frequency domain solver.


3) does it solve in a bounded or unbounded domain? what type of boundary conditions are possible :

Can you explain more in details bounded or unbounded conditions?


Bounded or unbounded condition are based on which formular the solver is using.
Both FDTD and FEM are based on solving differential form of Maxwell equation, that means the unknowns are the filed components, and the wave will definitely propogate to infinity(that means our computational domain is the whole space), so how can we truncate our computational domain? Yes, we are trying to truncate our computational domain without spoiling the simulation accuracy, we intentionally enforced the ABC/PML/Radiation boundary to them.
But for numerical techinques based on intergration formular, such as MoM, we dont have such concern, because of the unkowns are the virtual equvilant current(source) at virtual boundary, once we can get this, then substitute is back, we can get the field.


4) what is a full-wave solver ?


As my understanding, full-wave is utilizing the maxwell equation to describe the relationship between E and H without any pre-assumptions. it should be 3-D, of course.
Some solvers use 2-D or 2.5-D, it assumed the periodicity at the longitudal direction, so at that direction, E/H filed are assumed to be constant or exp(jβt).
And some solver, such as Q3D, it totally broke the relationship between E and H, it use electrical potential to calculat the E, use the current to calculat the H, it is called steady state calculation. The dynamic nature of EM wave was totally spoiled, but the accuracy should be ok for lower frequency, defintely it will fail at high frequency.

Best Regards.
 
3D Simulator

I am very proud that I come from the same country as you.
Admire!^_^
Thanks!
 

Re: 3D Simulator

asdfaaa ,

I disagree with some of your statements.

HFSS is much faster when accuracy is needed. The FDTD solvers have hard time refining their mesh and CONTROLLING the accuracy of the solution. So, if you intend to use (fast) time domain as a less accurate solver, you should add to the simulation time the time you intend to spend in the lab tuning your device. Accuracy control is a time saver. In HFSS you can choose to start solve a fast but less accurate solution, and then choose to solve it accuratly.

Second, HFSS has interpolating and fast sweeps which eliminate the need of solving each frequency point. Both broadband and shortband sweeps are done in an efficient manner.

Now - as for the time domain conversion being slower ....
Let us take a signal integrity problem such as a pulse entering a via. When you use a time domain solver and you need to co-simulate between the spice and the FDTD solver. Each time you cahnge the pulse shape you need to re-run the FDTD simulation !!!. That is unless you convert the time domain results into s-parameters (freq. domain) and then convert them back to time domain in spice.
HFSS creates directly the S-parameters, and it comes with a parametric model export into the circuit simulator. In addition, the circuit simulator can ask HFSS to solve a new variation if you do not want to interpolate between results. Furthermore, this circuit simulator is optimized for large s-matrices. There is no competition here.

I know of only one application that one might choose FDTD over HFSS, and that is if one must plot the fields in time domain.

Just another comment on Q3D. As long as the model is much smaller than the wavelength of the highest frequency it will be accurate. In the current version though, L is not frequency dependent and it does not calculate g (dielectric losses).

Mica188, what is the application you are looking a simulator for ?

Regards,
Itai
 
Re: 3D Simulator

Dear itaifrenkel,

I pretty much disagree with your statements.

First:
Modern Time Domain Solvers (like for example MWS) have adaptive mesh refinements that will give you accurate mesh and solution. As a matter of fact, this mesh refinement might even be better then a freq. domain refinement because the refinement itself is broadband! The HFSS refinement is only at a single freq.. There is no guarantee the mesh is good at another freq. As an example you just can use a filter and you can refine not in the pass-band but in the stop-band. As you will see, a freq. domain solver does not refine the mesh inside the filter because at the stop band freq. there will be no energy within the filter. If you then run a sweep, the results will be pretty much wrong! You have to make sure that the adaptive refinement is done in the pass-band.


Second: The numerical error introduced by the FFT can be neglected compared to other error sources (like mesh errors or tolerances in the geometry)

Third: The time needed to do the FFT is very small compared to the overall simulation time.

Fourth: Tools like MWS also have a spice model extractor (which is actually directly based on the time domain simulation) There is no need to re-run the simulation each time the pulse shape changes.


Best regards,

F.
 
3D Simulator

Hi,

Thanks for the explaination, very clear.

Right now, I'm in a learning curve. So, the first step is playing around with antenna parameters and analyse what'a can these softwares can provide. My real problem is to simulate the whole antenna system. Maybe one software will not do all but I want to limite software and the cost too. What I want to do is :

1) I can change the pulse shape easily and see the different effects related to this change,

2) Output data easy to manipulate, or moreover, a user friendly interface which gives me all the parameters plots of an antenna. In Far field plot not only Gain and directivity but also, the electromagnetic field plot.

Is it interessting you have mentionne fields plot in time domain, what is the utility of this plot? Can you explain more?

As I mentionne in the previous message, I'm using a demo version of SuperNec. I have 3 softwares I mind which are SuperNec , HFSS and XFDTD , XGTD from Remcom. Can you tell me the pros and cons of these softwares ?

Thanks for your time,

Mica
 

Re: 3D Simulator

Hi Mica,

you also should have a look at CST MicrowaveStudio. It is extensively used for antenna design. Just get a test version and have a look yourself.


F.
 

3D Simulator

Thanks, this will help. Where can I find the test version? I have navigate the CST website and I didn't see where I can download a test version.

Regards,

Mica
 

Re: 3D Simulator

Hi Mica,

you have to contact the guys from CST or one of their distributors directly. Just send them an email. The email adresses can be found on their web page.

Hope this helps.

F.
 

Re: 3D Simulator

RFSimulator,

MWS subgriding is still imature especially in medium and large problems. I refer you to:


The mesh in the adaptation frequency is accurate for all frequencies, and the best indication for that is the perfect comparison to measurements for the past 15 years or more.

As for the filter problem you mentioned, the HFSS user gets an alert that there is a problem in convergence,and there is more than one way to fix it. Does MWS alerts when there are accuracy problems?
Second, filter design is a frequency domain application per-se. I doubt that MWS can get accurate results for a narrowband filter using time domain.

Spice export - have you tried to run a broadband spice model with more than 10 ports (it takes a long time).A frequency domain representation and optimization algorithms in the circuit solver allows you to get the job done in a timly manner.

I must admit that MWS gained some market share because it released earlier a slicker UI, but since HFSS v9 I believe MWS is inferior.



Itai
 

3D Simulator

when I simulate the dual-frequency antenna in HFSS, mesh must be twice at two points, why?
At this point, I agree with RFSimulator.

Thanks for your brilliant comment!!
 

Re: 3D Simulator

I wonder why CST or Ansoft does not support parallel computing. I heard that MWS does to some extent. But thats only for multiport simulation setup. Time domain solvers are ideal for parallelization. For instance you can run Rsoft's Fullwave on many machines. Each machine simulating only a part of the simulation domain. This way you can beat the 2 GB RAM limit of windows and you got a great speed enhancement. I've ran fullwave on 5 machines using a total of 10 GB s RAM. Why CST doesnt support such feature on MWS? Is it to difficult to implement such feature to FIT formalism?
 

Re: 3D Simulator

EdmundZheng,

What do you mean "twice at two points" ? Please elaborate your problem.
Have you contacted Ansoft's technical support ?

Itai
 

Re: 3D Simulator

Hi,

CST does support parallel Processing for CPU's on the same mainboard. You can select the numbers of Processors you would like to use in the solver special settings if you have the appropriate license. I personaly use a double processor XEON system.

It also does support parallel computing for several ports in a network

Howerver CST does not support parallel processing in a computer cluster. I would guess that this is more a question of the implementation and not a question if is is possible

Best regards,

F.
 

Re: 3D Simulator

it would have been nice if CST have implemented such a feature. I know many guys who bought Fullwave just because it supports paralled computing over a network. Structures are getting much more memory intense. And network computing seems to be the only viable solution to beat such large problems
 

Re: 3D Simulator

itaifrenkel said:
asdfaaa ,

I disagree with some of your statements.

HFSS is much faster when accuracy is needed. The FDTD solvers have hard time refining their mesh and CONTROLLING the accuracy of the solution. So, if you intend to use (fast) time domain as a less accurate solver, you should add to the simulation time the time you intend to spend in the lab tuning your device. Accuracy control is a time saver. In HFSS you can choose to start solve a fast but less accurate solution, and then choose to solve it accuratly.

Second, HFSS has interpolating and fast sweeps which eliminate the need of solving each frequency point. Both broadband and shortband sweeps are done in an efficient manner.

Now - as for the time domain conversion being slower ....
Let us take a signal integrity problem such as a pulse entering a via. When you use a time domain solver and you need to co-simulate between the spice and the FDTD solver. Each time you cahnge the pulse shape you need to re-run the FDTD simulation !!!. That is unless you convert the time domain results into s-parameters (freq. domain) and then convert them back to time domain in spice.
HFSS creates directly the S-parameters, and it comes with a parametric model export into the circuit simulator. In addition, the circuit simulator can ask HFSS to solve a new variation if you do not want to interpolate between results. Furthermore, this circuit simulator is optimized for large s-matrices. There is no competition here.

I know of only one application that one might choose FDTD over HFSS, and that is if one must plot the fields in time domain.

Just another comment on Q3D. As long as the model is much smaller than the wavelength of the highest frequency it will be accurate. In the current version though, L is not frequency dependent and it does not calculate g (dielectric losses).

Mica188, what is the application you are looking a simulator for ?

Regards,
Itai

HI, sry for my delayed responsed. lolz, just back from weekend holiday.

You mentioned that HFSS is much faster than time solver. I admit that in single frequency calculation, it is faster than time-domain solver, but what will it be for wide frequency band calculation? Yes, HFSS have options for interpolating and fast sweeps, but for fast sweep type, the accurate solution is just at near the center frequency, at the frequency far from this center frequency, it is less accurate. of course, based on different application, you can accept this inaccurate solution, it is another issue. About the interpolating, I dont have much comments, I seldom use it, but for fast sweep type, sometimes it will give much ridiculous solution, for passive device, it give me the S-parameter great than 1 (in terms of dB), though it is faster than discrete solution type, so I doubt whether it is accurate enough.

Another issue is the convegence triteria of HFSS is based on port S-parameter (except eigen solution), if you are interested in the field intensity, you will find the convegence is much much slower than port S-parameter, so you have to increase your mesh, and use more strigent convegence criteria to achieve it.

Both HFSS and CST can provide SPICE model extraction. In this application, we need consider the DC component, it is the biggest problem for HFSS, since the lowest frequency limits it can provide.

For your mentioned example, we change the width of the input pulse, no matter HFSS and FDTD, we need re-do all the simulation, but for time domain solver, one round simulation we can get all information, freuency solver, i need to calucate enough frequency point information to re-construct the time-domain solution.

There is a long period of debate between HFSS and CST about the simulation time and accuracy problems, I have seen a report about the slotted-metal shield EMC calculation, it claims that CST cant compete with HFSS, and then CST provide another report to defend themselves, if you are interested in this, I can pass it to you.

Best Regards,
 
Re: 3D Simulator

RFSimulator said:
Hi,

CST does support parallel Processing for CPU's on the same mainboard. You can select the numbers of Processors you would like to use in the solver special settings if you have the appropriate license. I personaly use a double processor XEON system.

It also does support parallel computing for several ports in a network

Howerver CST does not support parallel processing in a computer cluster. I would guess that this is more a question of the implementation and not a question if is is possible

Best regards,

F.

My understanding is that CST don not use pure FDTD, for FDTD parallel computaion is quite normal and matured technique. In oder to improve simulation speed, CST empolyed integration in their formulation, exactly speaking, it has some similarity with xFDTD, I think that maybe the main reason for not easy implementing parallel simulation, and the same reason for HFSS.
 

Re: 3D Simulator

itaifrenkel said:
RFSimulator,

I must admit that MWS gained some market share because it released earlier a slicker UI, but since HFSS v9 I believe MWS is inferior.

Itai

The basic difference:

FDTD (of XFDTD or FIT of MWS) is better than FEM (of HFSS) and MoM (of SuperNec) in handling electrically large problem.

FEM (of HFSS) is better than FDTD and MoM in handling irregular 3D geometry.

MoM (of SuperNec) is better than the other two in terms of better accuracy and faster simulation for some limited types of problems (e.g. SuperNec treats metal wire and patch only, no dielectric; other MoM codes such as IE3D from Zeland can simulate more general structures.)
 

Re: 3D Simulator

Looks like Mica188 is interested in Antenna simulation. It is better to start learning some MoM code such as SuperNec, IE3D, and Feko. The code that comes with a book (e.g. Antenna and EM modeling with Matlab ?) might be sufficient. MoM is better than FDTD and FEM as a tool for learning the physics of the antennas. After getting familiar with MoM (and MoM codes), it would be easy to switch to FDTD or FEM code. Initially you might need to build some skill in solid modeling (for building the geometic model for the antennas), but eventually what counts is your understanding of the physics.

Just my opinion.






Mica188 said:
Hi,

Thanks for the explaination, very clear.

Right now, I'm in a learning curve. So, the first step is playing around with antenna parameters and analyse what'a can these softwares can provide. My real problem is to simulate the whole antenna system. Maybe one software will not do all but I want to limite software and the cost too. What I want to do is :

1) I can change the pulse shape easily and see the different effects related to this change,

2) Output data easy to manipulate, or moreover, a user friendly interface which gives me all the parameters plots of an antenna. In Far field plot not only Gain and directivity but also, the electromagnetic field plot.

Is it interessting you have mentionne fields plot in time domain, what is the utility of this plot? Can you explain more?

As I mentionne in the previous message, I'm using a demo version of SuperNec. I have 3 softwares I mind which are SuperNec , HFSS and XFDTD , XGTD from Remcom. Can you tell me the pros and cons of these softwares ?

Thanks for your time,

Mica
 

Re: 3D Simulator

it seems that there is no best code but best implementation
 

Status
Not open for further replies.

Part and Inventory Search

Welcome to EDABoard.com

Sponsor

Back
Top