Antenna Array Issues

Asifrizwan · 2025-06-26T17:22:34+00:00

There was an error rendering this rich post.

Hello,

I am trying to verify different sizes (infinity x infinity, 16 x infinity, 16 x 16) of the referenced Dual-Polarized Planar Ultrawideband Modular Antenna Array [1]. I have created a simplified parametric model of the Unit Cell (attached herewith) in FEKO.

Instead of modeling the connectors and energizing using waveguide port, i have used simple wire ports with the required radius during meshing. Small cylinders, instead of wires, were created between the middle thin dielectric layer to avoid small segment errors (Error 116: The ratio of the segment radius to length is too large).

After creating a uniform array, say 16x16, the model is taking awfully long time to even open, leave aside simulating. Spacing of 6.6 mm (Model parameters SpacingX and SpacingY) is used in both X and Y directions. The 16x16 model is too big to be attached here. Can the unit cell be simplified any further? Tried using MLFMM as well (ERROR 55723: It is highly recommended to use the conventional MoM, since the MLFMM near field matrix fill exceeds 20% of the full MoM matrix).
Machine details. Windows 10 PRO OS, i7, 32 GB RAM.
3x3 array with discrete frequencies is taking prohibitively long amount of time. Sum of the CPU-times of all processes: 250336.552 seconds ( 69.538 hours)
. On average per process: 31292.069 seconds ( 8.692 hours)
How to simulate an infinity x infinity array using the same unit cell?

DPelementForArray.cfx

[1] Holland, Steven S., Daniel H. Schaubert, and Marinos N. Vouvakis. “A 7–21 GHz Dual-Polarized Planar Ultrawideband Modular Antenna (PUMA) Array.” IEEE Transactions on Antennas and Propagation 60, no. 10 (October 2012): 4589–4600. https://doi.org/10.1109/TAP.2012.2207321 .

Warm regards,

Asif Rizwan

Find more posts tagged with

Feko

v2024.1.0

Accepted answers

All comments

Asifrizwan

Hello,

Tried using MLFMM on a 3x3 array. Got the following error:

ERROR 55723: It is highly recommended to use the conventional MoM, since the MLFMM near field matrix fill exceeds 20% of the full MoM matrix

Using MOM for a point frequency is taking huge resources:

Not enough memory for an in-core solution, trying out-of-core ...

Out-of-core solution possible (using block size 64)

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++

++ Temporary files are used on the hard disk. ++

++ Memory requirement on hard disk: 34.972 GByte

++ i.e. 4.372 GByte per process.

++ Environment parameter FEKO_TMPDIR can be used in ++

++ order to specify the directory for these files. ++

++ FEKO_TMPDIR is set to the value: ++

++ C:\Users\User\AppData\Local\Temp\feko\

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++

For 6 discrete frequencies, a 3x3 array is taking following resources in terms of time:

Sum of the CPU-times of all processes: 250336.552 seconds ( 69.538 hours)
On average per process: 31292.069 seconds ( 8.692 hours)

Am i wrong in assuming that the finite array tool leverages Domain Green's Function Method for quick and accurate simulation of large and complex antenna arrays?

How to make the simulations quick?

Regards,

Asif

Torben Voigt

Hi @Asifrizwan,

Could you upload the 3x3 array model here: https://securefiletransfer.altair.de/filedrop/voigt@altair.com

I would like to have a detailed look at it. Yes, the DGFM (Domain Green's Function Method) might help improve runtime & memory here, but maybe there are other things one can improve.

Note, saving the model without mesh will reduce the file size. Even more important: Save under new name will also reduce the file size. In version 2025 the file size keeps increasing with each saving, which is going to be imrpoved soon.

Best regards,
Torben

Asifrizwan

Hi @Torben Voigt,

Thank you for your reply.

I have uploaded the 4x4 model (without mesh) to the link provided by you.

This array took entire weekend to simulate:

Specified CPU-times are referring to the master process only

Sum of the CPU-times of all processes: 1532730.897 seconds ( 425.759 hours)

On average per process: 191591.362 seconds ( 53.220 hours)

Please have a look and suggest.

Regards,

Asif Rizwan

https://community.altair.com/discussion/63500/antenna-array-issues

Torben Voigt

Thanks Asif, I will have a look asap.

Torben Voigt

Hi Asif,

Are the positions of the sources correct? It looks to me a little as if they were distributed randomly. Did something go wrong when creating the array?

I would like to try DGFM, but for that the sources would always have to be in the same place.

Best regards,
Torben

Torben Voigt

I just remembered that the cells must not touch each other in DGFM. Unfortunately, this is not an alternative...

Asifrizwan

Hi @Torben Voigt,

The positions of the sources are correct.

The encircled elements in black model two arms of the HP dipole on top surface of the middle dielectric.
The encircled elements in green model two arms of the VP dipole on bottom surface of the middle dielectric.
Two middle wires of each element are the feed-lines, and the other two wires model the shorting Vias.
I have manually removed sources from the edge elements (because they only have one dipole arm). We can keep them. Exciting them will not contribute significantly in a large array.

Asifrizwan

If the cells must not touch each other, then how should we model this array?

Is it possible to model this array with PBC?
Is it possible to model the substrates (3 layers) as infinite multilayer substrate and use DGFM for just polygons and lines (which are not touching)?

Torben Voigt

Thanks for confirming the sources' positions.

For PBC the unit cells must be identical (= same source position)
Planar Greens functions would be an option (need to check if this will result in significant reduction of memory/runtime), but still for DGFM the array elements must be identical (= same source position)

Will do some testing and get back to you at a later stage.

Best regards,
Torben

Asifrizwan

@Torben Voigt, I am a bit confused about what you mean by same source position.

The unit cell (model attached in the first question as DPelementForArray.cfx) can be used to model an array of any size using the macro. Each element of the generated array will be identical and have the same sources' position.

Asifrizwan

Correct me if I'm wrong. In the following Altair demo video, the cells do touch each each-other https://www.youtube.com/watch?v=I5rU8VYGuR0

Torben Voigt

Hi Asif,

Neither PBC nor DGFM is used in the video. There, an array is created with the help of the Genarate Array macro (like you did) and then simulated with MLFMM.

Best regards,
Torben

Torben Voigt

Hi Asif,

That's why I asked about the sources positions. In your model DualPol4X4.cfx some array elements have two sources, some have only one and one element doesn't have a source at all.

Best regards,
Torben

Torben Voigt

Hi Asif,

Your model DualPol4X4.cfx requires 86 GByte of RAM. That's why the solver goes "out-of-core" (uses hard disk) and the simulation gets very slow.

With planar Greens funcrion the simulaton gets much quicker: 1.145 h and 1.48 GByte with 16 parallel processes. Model is attached.

DualPol4X4_pGF.zip

I'm presently testing using FEM instead of MoM. Sould be much quicker as well, but a bit of manual process to build the model. Wires are not allowed with FEM (only for FEM Line Ports), and since the sources are not identical for all elements it took some time to build it.

Will report how it goes.

Best regards,
Torben

Torben Voigt

I would stay away from FEM for now. I already have the results after just 20 minutes, but I'm afraid that I may have made a mistake because I set the ports manually. In any case, the results differ significantly from MoM_pGF.

Maybe you want to check whether the ports and sources are correct.

DualPol4X4_FEM_decoupled.zip

Best regards,
Torben

Asifrizwan

Thank you @Torben Voigt,

I am not able to open the models. I have FEKO 2024.1 and models have been created with version 2025. Nonetheless I have modified my model with PGF and it has significantly reduced the resources required. Will revert to you once the simulation is over.

Thank you,

Asif

Asifrizwan

Can you please send me the 2024.1 version? I am not getting an option to upgrade FEKO to 2025 yet.

Asifrizwan

I am able to open the models now. I have installed the 2025 version.

Asifrizwan

Hi @Torben Voigt,

If the elements are identical and not touching as in the attached model, how can we speed up the simulation?

Regards,

Asif

DPelementForArrayWithGap.cfx

Torben Voigt

Hi Asif,

Please take a look at UserExample A.17. The use of DGFM is explained there. Sorry, I don't have time until next Tuesday.

Best regards,
Torben