Simulate UWB antennas in HFSS

Hi,

I have to simulate an antenna over a very large bandwidth : 70MHz - 20GHz. In HFSS, the high frequency (here 20 GHz) must be taken for the solution frequency. So the mesh is really thin compare to the big size of the antenna. Therefore I can't simulate the system because of the extremely large number of tetrahedra.

So if anyone have a idea to simulate UWB antennas with a good accuracy, I'm interested :grin:

I would suspect that you are studying a relatively low gain antenna. With that assumption I would imagine that with about 20 Gb of RAM you could make a simulation run at 20 GHz. The problem would probably run close to 1M tets.

HI,
Can you use another software like FEKO ?

Azulykit said:

I would suspect that you are studying a relatively low gain antenna. With that assumption I would imagine that with about 20 Gb of RAM you could make a simulation run at 20 GHz. The problem would probably run close to 1M tets.

Click to expand...

This antenna has a gain of 4dB in the 2 directions of propagation. Why did you say that it's possible to simulate a low gain antenna? and how is it possible?

NEMO25 said:

HI,
Can you use another software like FEKO ?

Click to expand...

I've tried to simulate this antenna on Ansoft Designer (2.5D) but I had an error "Out of memory" at the beggining of the simulation. I have 32 Go of RAM.
Do you think that FEKO would be better to simulate that kind of antennas?

Thank you for yours answers.

Hello,
If your antenna is planar, feko has infinite dielectric and ground plane that can help you to simulate the antenna, because only the metal part will be meshed. In HFSS the hole structure is meshed, so it needs more RAM to simulate the antenna.

hi all dear
u don't need simulation u antenna with other software when u have enough system for HFSS . for UWB simulation u need 4G RAM . but i should emphasize that hardly u can find and antenna that work in this range (70MHZ to 2GHz) and it return loss in all frequency under -10dB . and u should consider that if u simulated by HFSS u should simulated every 2 or 3 octave in one simulation !!! please up lode u simulation for more help.

I see 3 possible solutions:
1. Use HFSS freq solver and split your wide band on some narrow bands. With the ratio hi freq/low freq =2 it should be OK.
2. Use fdtd, CST MWS, XFDTD or something.
3. Use hfss 13 it has implemented transient solver inside.

There is another possibility:
- Use the standard HFSS frequency domain solver, size the airbox for about lambda/10 for the middle of your band, and use a PML for termination. Then enable the iterative solver in the solution setup options. If will compress the matrix solution to use significantly less RAM.

---------- Post added at 02:59 ---------- Previous post was at 02:55 ----------

Vins35 said:

I've tried to simulate this antenna on Ansoft Designer (2.5D) but I had an error "Out of memory" at the beggining of the simulation. I have 32 Go of RAM.
Thank you for yours answers.

Click to expand...

This doesn't make a lot of sense. Assuming that your antenna is planar, then Designer PlanarEM solver should work just fine, and shouldn't run out of memory. I've solved 256 element arrays in Designer using only about 2 GB of RAM.

sweetjimmy said:

Assuming that your antenna is planar, then Designer PlanarEM solver should work just fine, and shouldn't run out of memory.

Click to expand...

If the antenna is electrically large (many wavelengths at highest analysis frequency) and there is a lot of metal to be discretized, then the Method of Moments simulators might need much memory. Matrix size grows with N^2, where N is the number of mesh cells.

For these cases, time domain solvers (FDTD) might be useful because their simulation time and memory scales linear with N. This makes the FDTD method very efficient for electrically large antenna models.

you have an important problem with memory used in the HFSS. to reduced it you can used some symmetric plane in your simulation. with symmetric plane you can reduced memory and computation time.

volker_muehlhaus said:

Matrix size grows with N^2, where N is the number of mesh cells.

Click to expand...

This is only partially true. The PlanarEM solver in ANSYS Designer employs several matrix compression techniques. The most aggressive of which grows at only N log N. As I also mentioned I've solved a 16 x 16 element array with the planar solver on my laptop with less than 2 GB of RAM, so it is extremely efficient.

sweetjimmy said:

The PlanarEM solv0er in ANSYS Designer employs several matrix compression techniques. The most aggressive of which grows at only N log N.

Click to expand...

... which is still worse than ~N in FDTD solvers.

sweetjimmy said:

As I also mentioned I've solved a 16 x 16 element array with the planar solver on my laptop with less than 2 GB of RAM, so it is extremely efficient.

Click to expand...

I do not get your point. If the individual antennas are not too complex, and you go for a relatively coarse mesh, you can analyze arrays.
However, in this thread the antenna is ultra broadband and the dimensions vs. wavelength at the highest frequency are somewhat extreme. Your array example is very different in terms of requirements.

volker_muehlhaus said:

... which is still worse than ~N in FDTD solvers.

Click to expand...

Yes. That is a correct statement. However, comparing a 3D FDTD solver to a 2.5 D MoM solver is comparing apples to oranges. A FDTD solver will require the mesh for the volume which the Planar solver will not. Both methods will allow the user to use a minimally sufficient mesh to get a fast answer with the least amount of memory, but the best solver will give the most accurate answer for the minimum amount of compute resources.

volker_muehlhaus said:

I do not get your point. If the individual antennas are not too complex, and you go for a relatively coarse mesh, you can analyze arrays.

Click to expand...

In my example of a 256 element array in the PlanarEM solver, I was not using a coarse mesh. I was utilizing matrix compression techniques to bring the memory requirements down to about NlogN.

volker_muehlhaus said:

However, in this thread the antenna is ultra broadband and the dimensions vs. wavelength at the highest frequency are somewhat extreme. Your array example is very different in terms of requirements.

Click to expand...

A UWB antenna is going to have an electrically large mesh, just as an array antenna will have. I was simply comparing the electrical size of the mesh generated with a 256 element array in the PlanarEM solver, and the amount of memory it takes. I've also modeled a UWB spiral, cavity-backed antenna in HFSS requiring only about 2 GB of RAM. I was responding to the original poster's comment about the UWB antenna taking more than 32 GB of RAM for a planar solver. That doesn't make sense to me, and I suspect that there there is a setup problem. I would recommend that the user contact their local ANSYS support engineer to help diagnose this issue.

sweetjimmy said:

A UWB antenna is going to have an electrically large mesh, just as an array antenna will have. I was simply comparing the electrical size of the mesh generated with a 256 element array in the PlanarEM solver, and the amount of memory it takes.

Click to expand...

Sure, but look at the dimensions. Your patch array is only 16x larger than a single antenna. The frequency range given above is 285:1, so that the antenna will be very large (in wavelength) at 20GHz.

Instead of using a big software like HFSS, I modeled a first approximation with a wire MoM code. For example, I used **broken link removed**.