How to correctly create copper traces for PCB-Antenna

[Langer-Jan]

Hi!
I am doing my very first steps with OpenEMS.
I saw the tutorial from Georgy Moshkin and the FOSDEM video from Thomas.
I am totally amazed what openEMS can do!

Now I am doing my first steps. I am trying to simulate my first meandering BLE antenna by modifying the sample code from Simple_Patch_Antenna.m.

I have added a return statement before RunOpenEMS is called, so that I can run my script and see the Output of my modeling with WriteOpenEMS without starting the simulation.

I want to create my antenna by adding multiple boxes like

CSX = AddBox(CSX,'patch',10,[1.4 4.7 substrate.thickness],[1.9 6.0 substrate.thickness]);

to create the zig-zag-pattern.

Every time I hit the run button to check my geometry I get the question

remove entire contents of tmp Patch_Ant? (yes or no)

in the Command window, where I have to enter yes.

After adding the fourth segment for the patch (line 71 in attached code) the Command window clears when pressing Start, but I do not get the above question anymore and Octave freezes infinitely. I have to terminate the program with the task manager.

I have no idea how to overcome this problem.
Any suggestions welcome!

I am using Octave 11.3.0 (2026-06-01) and openEMS_v0.0.36 on Windows 11.

Jan
Nochmal_Antenna.zip

[VolkerMuehlhaus]

Hi there!

The solution to the "remove entire contents" question is easy. You just need an additional command in your *.m file.

close all
clear
clc
confirm_recursive_rmdir(0);   % delete old data without asking

Regarding the crash, I was lazy and asked my dear friend Claude AI, answer is below.

---

Box 4 shares exact coordinates with box 3 at x=1.4 and y=4.7, and its region x:1.4–1.9, y:4.7–5.2 overlaps box 3 entirely instead of being adjacent to it.

That overlap is the problem. Later in the script:

matlab
mesh = DetectEdges(CSX, mesh,'SetProperty','patch','2D_Metal_Edge_Res', c0/(f0+fc)/unit/50);
mesh = SmoothMesh(mesh, c0/(f0+fc)/unit/20);

DetectEdges inserts refined mesh lines at every edge of every box making up 'patch'. Because box 4 overlaps box 3, it introduces duplicate or near-duplicate edge coordinates at 1.4 and 4.7 that are already present from the other boxes, but does so in a way that creates degenerate/extremely thin intervals. SmoothMesh then tries to iteratively enforce a maximum cell-size-growth ratio between neighboring lines — with near-coincident lines this constraint can't converge cleanly, and in Octave's implementation this manifests as an effectively infinite loop (it's a known pain point with SmoothMesh/DetectEdges on self-overlapping metal geometry, worse in Octave than MATLAB).

The overlap, precisely:

Box 3: x:[1.4, 4.3], y:[4.7, 5.2]
Box 4 (commented out): x:[1.4, 1.9], y:[4.7, 6.0]

Their intersection — x:[1.4, 1.9], y:[4.7, 5.2] — is non-empty. Both boxes are zero-thickness (same z on start/stop), sitting on the exact same plane, so you get two coincident PEC surfaces covering the same area with shared boundary coordinates at x=1.4 and y=4.7.

Why this hangs specifically in SmoothMesh:
DetectEdges pulls in edge lines from both boxes at x=1.4, x=1.9, x=4.3, y=4.7, y=5.2, y=6.0. Because box 3 and box 4 share the line x=1.4 and y=4.7 exactly, but diverge elsewhere, SmoothMesh ends up with mesh lines that are pathologically close together in some places and normal elsewhere. Its algorithm grows/shrinks cell sizes iteratively to satisfy a smoothness ratio between adjacent lines — when it hits nearly-coincident lines from the overlap, in Octave this loop doesn't terminate (this is a known rough edge in Octave's version of SmoothMesh, less forgiving than MATLAB's).

The actual fix — make box 4 non-overlapping:

matlab
start = [1.4 5.2 substrate.thickness]; % start at y=5.2, not y=4.7
stop = [1.9 6.0 substrate.thickness];
CSX = AddBox(CSX,'patch',10,start,stop);

This only adds the new area you actually want (the little tab above box 3) instead of re-covering box 3's region. With no interior overlap, DetectEdges/SmoothMesh gets a clean, non-degenerate line set and won't hang.

[VolkerMuehlhaus]

One more comment: if you are new to openEMS and willing to create your models in Python, you could use the wonderful automatic meshing add-on that Mustafa presented here:
#421