Feko PO element Bistatic Far Field RCS not continuous/symmetrical

Hi @Jizheng He, 

Could you attach your model (.cfx)? I would recommend using geometrical symmetry so that the mesh is really symmetrical. Also, with PO (asymptotic solver!) there will always be small deviations, especially if the model is not electrically large. I would only recommend PO from a model size of at least 20 lambda.
Have you ever compared with full wave results (MoM, MoM with HOBF or MoM with CBFM)?

Best regards,
Torben

Hi @Jizheng He, 

Could you attach your model (.cfx)? I would recommend using geometrical symmetry so that the mesh is really symmetrical. Also, with PO (asymptotic solver!) there will always be small deviations, especially if the model is not electrically large. I would only recommend PO from a model size of at least 20 lambda.
Have you ever compared with full wave results (MoM, MoM with HOBF or MoM with CBFM)?

Best regards,
Torben

Hi Torben, thank you for your reply! I attached my .cfx file to this post.

My surface is constructed by rotating an arc and I enabled "use symmetry in ray-tracing when possible" in solver settings. My reflector is 10 inch diameter = 254 millimeters, much larger than the sub-mm-level wavelength for 64GHz frequencies.

Unfortunately I cannot run full-wave results with MoM: using curvilinear meshing calculating A1x submatrices take forever to run, and not using curvilinear meshing gives me 1M triangles with coarse meshing since the reflector is large. I don't know if MoM with HOBF or CBFM will make the calculation quicker?

Thanks,

Jizheng

Hi @Jizheng He,

I meant this with symmetry:

I'll see what the results look like with it. I'll also try to see how much memory and runtime is needed for MLFMM. Since there is only one incident plane wave angle, MLFMM is expected to be more efficient than MoM CBFM. Maybe MLFMM CBFM could also be tried.

Best regards,
Torben

Hi @Jizheng He,

In the model you have attached, the plane wave comes from the negative z-direction, which makes PO completely unsuitable here. PO does not take into account reflections between PO faces, which greatly distorts the result. If the plane wave came from above, as in your pictures, PO would definitely be an option.

By the way, I can simulate the model with MLFMM (preconditioner SPAI to save memory) within 1 h, with 16 cores and 170 GByte memory requirement. I attach the model here.

Best regards,
Torben

Hi @Torben Voigt Thank you so much! I meant to have the wave coming from the top (-z propagation direction, coming from +z) and the reflector as a convex reflector (which should be good with PO, as you pointed out).

However, my PO model (although symmetrical in the phi direction now) still makes no sense and are extremely rugged. 

The sides of this RCS is not making much sense to me - at oblique angles I am expecting something like

Why are there so many rugged edges and larger bistatic RCS at the sides?

Thank you for the tip on MLFMM with preconditioner SPAI. I am currently running one with course meshing and stabilization, and will post if it works!

Hi @Torben Voigt,

This is the MLFMM result with +z coming wave and course meshing. I kept everything else the same except for only calculating scattered fields in the .fek file you sent. I am still confused by the constant RCS between theta = 30 to 90: why is there even a scattered field at 90 degrees? With this large of a reflector (10 inch or 25.4cm diameter, compared to 4mm wavelength) there shouldn't be any diffraction as well...

I am trying to run it with standard meshing with decreased parallel processes due to my 128 GB RAM limit. Thank you so much for the help!!

Best Regards,

Jizheng He

Hi @Jizheng He,

Let me try on my side as well. What I can say is that I have no doubts about the MLFMM results. To me they don't look strange at all, RCS non-zero at theta 90° is absolutely expected (e.g. due to creeping waves on the surface).

Note: It is better not to use Stabilized MLFMM, as this can take a very long time. Stabilized MLFMM is only recommended for rare cases where the solver would otherwise have problems with convergence.

Best regards,
Torben