UWB antenna simulation

[AdamStastny]

Hello,
I am making my first simulation using gerber2ems package and UWB mini-hoe murata antenna. The antenna is created for channel 5 uwb (roughly 6,5 GHz). Now, I want to simulate this antenna with openems and make some possible changes to it in the future. My problem now is that whatever I do, I have positive S11 parameter. I uploaded some images and added my config files for gerber2openems.
Thank you for any ideas, that could help me!

[simulation.json](https://github.com/user-attachments/files/28277326/simulation.json) [stackup.json](https://github.com/user-attachments/files/28277329/stackup.json)

[LubomirJagos42]

I would say your mesh is totaly wrong, there is parameter --debug-PEC when openEMS is triggered and it will export PEC grid into .vtk file so you can see how PEC is represented in openEMS in Paraview and you will see that your antenna shape is poorly meshed, it's too corase like just few lines, you have fine mesh around input port but antenna shape itself is just too few lines I would personally add there at least 10x more gridlines to have real fine mesh for antenna shape

[LubomirJagos42]

more and more I am looking on that mesh it's sooo bad, even in Z coords your vias are not meshed and soo many things are there wrong, this would be better to define mesh manualy even faster that --debug-PEC is really useful specially for this

[AdamStastny]

Yes, thank you, I did not think about that! I will fix it right now.

[LubomirJagos42]

sure it's typical error, first step to get model into openEMS is cool and then when things not matching expectation is mostly there is mesh wrong, I used to switch that AppCSXCAD into 2D and observe if at least some lines are crossing input port what is important and then I am looking if my shape is covered by enough lines, more lines is better than few :D that mesh lines is fundamental things where whole simulation calculation magic is happening therefore if you have poorly meshed shape it behaves just like thin wires therefore your S11 is totaly wrong (like it's right but your model doesn't represent what you want :D)

[LubomirJagos42]

most of these automatic tools are trying to do some magic but your shape is exact example where I see that no automatic tool can make that mesh right since this geometry is hard to be recognized by some algorithm, it mades lines around your port right but don't know about your other connection or shape since it's closed loop, visually you can spot it right away therefore this case is book example that manualy you are able to make mesh right in 20min but some automatic algorithm cannot solve this, still better to spend 20min manualy define mesh and then 1hour for result

[LubomirJagos42]

there are another errors why your simulation will be wrong:

• no airbox around your antenna add there at least 1 lambda around antenna airbox
• airbox should be absorbing and there should be 8 cells (gridlines) in X,Y,Z to properly absorb energy from antenna

so even if you mesh your model right now it still will not work since energy wouldn't be absorbed by boundaries

specially for these cases I made my FreeCAD addon https://github.com/LubomirJagos42/FreeCAD-OpenEMS-Export

[LubomirJagos42]

if you post here your gerber model or whatever it comes I can try to play with it, it's public model probably and it's quite easy to make this work when you have some experience with openEMS

[AdamStastny]

Single-antenna.zip
Yes, that would be great! Since it is public model, I am sending here a zip with gerber export.

[AdamStastny]

Btw, the public model comes without GND layer. I added it based on DWM1000, that uses this antenna.

[LubomirJagos42]

maaaaaan, this design is main reason why I even made my addon for FreeCAD, this is so damn task that without FreeCAD I would never found why there are so many pitfalls specially because that shape is kind hard to understand PC how to close that shape, briefly this is my result

I would write it here how I made it, have to correct some things which I screwed since I was extending my addon and now it missing some 6 lines to generate correctly, long story shor I have to:

1. import gerbers into KiCAD
2. export them as KiCAD board
3. import them in FreeCAD
4. correct many thing, almost redraw that shape of top antenna layer in FreeCAD manualy what is fast as it's CAD and drawing is fast
5. make sure that top layer is solid to be right exported as solid stl file
6. run simulation

main reason why probably that gerber to openems tool is failing is that when I was looking what FreeCAD imported from KiCAD it was far far from solid, it can visualy looks right but it's not fastest was used FreeCAD drawing skill and redraw it and run simulation

[LubomirJagos42]

@VolkerMuehlhaus just briefly looking in openEMS forum for "STL hollow" problem like we were both solving 2 years ago #250
and I remember it was older problem and that thread is still somewhere here when we discover that problem about hollow STL objects coming from other tools, it's not that I don't trust semi-automatic tools but when they create some STL files you should check that STL files if there are ok, there should be somewhere some python script or whatever just to check if STL is valid, FreeCAD has this checker inside, internet is nowadays flooded with invalid STL files, it's not disaster as there STL healers, but when I see STL file I don't trust it until it's exported from FreeCAD as I know for 100% FreeCAD heals them for sure because it's solid modeller

[VolkerMuehlhaus]

@LubomirJagos42 You are describing issues with STL, but from what I can see here, gerber2ems uses openEMS polygons with extrusion.

That's what I also use in my gds2openems workflow: good old solid polygons, AddLinPoly() plus extrusion. No trouble with that f*** STL data.

[LubomirJagos42]

@VolkerMuehlhaus ok I understand, can you simply make this simulation works? I saw this thread week ago and wanted to help and did it my way since I feel comfortable in my workflow and if I have to try new things which I don't fully understand makes this harder for me, if you have your workflow it's for you like 2hrs work, can you try to simulate this model with input data to see if you got similar result? I am asking because I am really interested in your workflow

[LubomirJagos42]

@VolkerMuehlhaus I don't mean it in bad way but what is reason to just give partial push to solution when you are able make whole solution, to make openEMS script is hard as f*** and just pushing people to solution is kinda no help, that's main reason why I created my GUI addon since I was pissed it's hard as f*** and wanted to just provide simpler way how to create simulation so they will not struggle for half year as me
You have your solution so if it's so easy show me your simulation, for me with my slow workflow it took just like 2hours, but I know that you are faster and I can learn something.

[VolkerMuehlhaus]

Hi @LubomirJagos42,

this gerber2openems tool is not my workflow, I never used this. I just mentioned my own RFIC workflow because both use native openEMS polygons, and avoid all the trouble related to STL files.

From what I read in the description, there is a mesh configuration in one of the JSON config files, so I think @AdamStastny can simply refine the mesh in the curved section.

Best regards
Volker

[AdamStastny]

First of all, thank you for your time. I struggled today with implementing it on my PC with your data, but finally, it works. I find out that for some reason, the macro cannot export Excitation parameters into Octave so I have to put it in there manually (not sure, how to fix that).

In Material PEC, there are all copper layers - vias, top and bottom copper, which should be always in this state as three parts. If I have for example 6-layer PCB, I should add those layers here as well, so in case of through-PCB-vias, I will have 7 "PEC parts". The same goes for DIEL_FR4, where in case of 6-layer, I will have 4 "parts" here.

The grid contains airbox, fine, Z 1 line and Z 3 line. Airbox is self-explanatory. Z 1 and Z 3 slices the Z "dimensions" into 4 separated parts. Z1 contains only copper, while Z3 contains substrate.
As you can see, I do not how to set-up the grid... I will figure it out after some time, when I will get used to it, now only some more question.
Boundary offset is offset from (0;0;0), so if it is set to Z 100,0um, it means that this grid will start from 100,0um below/above Z plane. Fixed count then says, that theere will be 3 grids - how big would these be? Fixed distance then probably says that each cell will have defined dimensions. And what about smooth mesh?

Thank you for clarifying that and thank you for your time!
Beer sounds good, hope some different day, i am not in the prague now for 2 weeks.

[LubomirJagos42]

I fixed that generator now, it was my error since I was adding new solver support there and haven't corrected way how to grab data from GUI, now it's working :D :D :D I screwed that stuff in beginning of March so at least it wasn't screwed for long time :(

[LubomirJagos42]

I am not sure how many gridlines you should put to got real world measurement results there but easy to say that if you are not sure more lines and close to each other makes your simulation slower but at least you will got right result :)
I just wanted to run it quick so I put one line layer for copper (top and gnd) as it's enough since it's PEC and 3 lines for substrate to simulate it's thickness.

According to meshing (setting gridline positions) in openEMS this is done manualy, in scripted examples since designs are specified variables you can use like:
mesh.x = [mesh.x linspace(point_x1, point_x2, 10)]
to put 10 lines in x direction (these lines will be parallel with Y axis) but when you import STL files you have just thei boundaries, that's exactly what I do in my addon and from FreeCAD I have boundary coordinates and as you wrote there are options:

• fixed distance: starts from lower coordinate and put lines in fixed distance, therefore last line don't have to match end of object at that coordinate
• fixed count: it takes lower and higher coordinate in some ax and split that distance to specified count, if there is just 1 then that line is placed into middle of that distance
• that smooth mesh: I did this years ago and at that time I tried that but never used it in models so now I am not sure how it's running exactly but it's doing some adaptive spacing so distance between lines is gradually changing

All in all if you know boundary point coordinates you can write your own mesh method or whatever whole thing is that all these methods they just fills variable mesh.x, mesh.y and mesh.z and that's whole magic.

When you run simulation you can check if lines cross your model in AppCSXCAD that probably most important step and there you can easily spot if all is meshed ie. if gridlines are crossing input and output ports, if they are crossing PEC material and so I use to use here lot of view in 2D since it's exactly showing you precise gridlines position, 3D view is often lying specially when you are looking if gridline is crossing 35um thick copper trace.
And finally when you run simulation it's good to use --debug-PEC to check in Paraview how metal thing generated there it really shows you reality what is used as input in input matrix and if you don't see nice structure as I send at that point you can copy any file into simulation dir with name ABORT or abort and it will stop simulation in next timestep or just kill whole process in process manager.

If you try some examples all these things become natural since there is no other way how to do them.

[LubomirJagos42]

and that boundary offset it means that gridlines starts and ends with some offset from boundary coordinates from FreeCAD, I put it there since I often make 1 line layer for copper traces and 3 lines for substrate but when substrate lines starts without offset there is just (35/2)um what will force solver use small timestamp, I like idea to have 3 lines in Z for substrate since it has some thickness but also don't want to spend ages waiting for test simulation if model is running so I put distance 100um or something

there we added also some feature to delete some gridlines when they are too close to each other since when using step models or in general it can happen that your gridlines are 2um close what is hard spot in AppCSXCAD but little visible as that gridline there is more darker but you can definitely see it in openEMS solver output that it writes you that your timestep is too small and probably your mesh is too fine and you can see that timesteps are going too slow

when you have right simulation you simply know that you have right simulation, when you screw your simulation it's also clear that something is wrong :D

[LubomirJagos42]

have you solved that NF2FF probe? I can see it in mail but no in this discussion so I am not really sure

[AdamStastny]

Yes, I solved it right after I posted that comment - so that is why i deleted it.