SWR/Impedance calculation way out

[TerryBarnaby]

Hi, I am new to Openems and am trying to use Openems to simulate a simple 868 MHz PCB antenna design that I have a PCB for so I can experiment with antenna designs. It certainly isn't a good 868 MHz antenna! After playing for a bit I have a Python program generating a simplified PCB arrangement, meshing it and running the Openems simulation with that.

In general it appears to be working an yielding return loss and SWR graphs that are about right in terms of frequencies, but with very high SWR/Return loss that does not match those provided by a NanoVNA on the real PCB. I have played around with a lot of aspects and am currently using the easyMesh4openEMS meshing system. No mater what I try I don't seem to be able to get my simulation to reasonably match the real world.

My Python code simply creates a PCB substrate, places a square ground plane on this and then creates a quaterwave PCB track antenna to the right of this roughly matching the PCB design.

Simulating a large flat plate with a quaterwave antenna perpendicular to this appears to work reasonably well (python sim-antenna-example.py 1).
Simulating my real PCB ((python sim-antenna-example.py 2) which is a PCB antenna with quite a small ground plane, is way off.
The nanoVNA reports a SWR of about 5, while Openems reports about 23 with a very low antenna impedance.

I wonder if anyone can provide me with any pointers on what I am doing wrong ?
The results look like the attached image vs the NanoVNA and the Python code is attached.

This is using the GIT OpenEms with Git easyMesh4openEMS on Linux Fedora43.

[sim-antenna-example.py](https://github.com/user-attachments/files/29157355/sim-antenna-example.py)

[VolkerMuehlhaus]

The mesh looks good, nice to see easyMesh4openEMS in action here!

From your screenshot, it seems that your boundary condition is too close. It should be at least lambda/4 away from the device.

If you are using PML boundary, that needs some additional margin: PML is implemented inside the simulation volume, with one cell per PML order, e.g. 8 cells on each side are occupied by PML8 material.

That said, there might also be a mistake in your measurement. In openEMS, the lumped port is truely floating. In measurement, if the coax feedline was attached to ground plane with no ferrite to suppress common mode, the coax shield will then become part of your radiating structure. This would detune the antenna, especially if the PCB ground itself is small.

[TerryBarnaby]

Many thanks for the reply.

I had tried increasing the boundaries and just double checked with a lambda/4 boundary, but that didn't make much difference (I assume that easyMesh4openEMS had used a useful size by default ?).

I hadn't considered the VNA coax feedline aspect, I will get a ferrite from work next week and try with that.

However I tried increasing the openems design's ground plane downwards by about 1.5lambda/4 and also 3lambda/4 (to match where my VNA lead goes) and that didn't change things much (as per design pic). I thought that might approach the real world better as per your suggestion.

I also tried extending the ground plane to the left. That did show the SWR approaching what the VNA was providing (for some reason) although its calculated impedances J was +ve rather than -ve as per the VNA ?

[VolkerMuehlhaus]

Hi Terry!

looking at your latest models, and the size of the 868 MHz radiator vs. margin to simulation boundary, that still looks too close? Wouldn't that margin be same distance as the radiator length?

I assume that easyMesh4openEMS had used a useful size by default ?

This rule applies to antennas only, and my best guess is that easyMesh4openEMS didn't know your design is an antenna.

I also tried extending the ground plane to the left. That did show the SWR approaching what the VNA was providing (for some reason) although its calculated impedances J was +ve rather than -ve as per the VNA ?

This comparison requires that the phase reference for measured S11 is at the same location where the lumped port in your model is. Otherwise, your measurement turns around in Smith chart, depending on the extra length between VNA calibration plane (SMA I guess) and the end of the cable that solders to your PCB. You could verify by creating a SHORT at the end of that cable, and check if you get a wideband short in measurement, or some phase offset that turns you around in Smith chart.

Somehow the link to your model in the initial post is broken, so I can't look at your model details to double check for other issues.

Best regards
Volker

[TerryBarnaby]

Any reasons you can think of as to why the larger vertical ground plane design did not affect the SWR while the horizontal one did ? I would have thought the vertical one (the same axis as the antenna section), that more closely matches the real world would have been the one that brought the simulation in line ?

On the PCB I have an UFL with a 0402 0R to the antenna drive track (there is another 0402 0R to the actual drive circuit removed). I think reasonably closely matches the Openems design (well the PCB has an 8mm long 50 Ohm track (over an internal ground plane (4 layer PCB)) to the antenna feed point) and the VNA is calibrated to the UFL connector on its lead. So I think the openems design reasonably closely matches. However I am only simulating a single copper plane rather than the actual 4. I did try making that single plane the thickness of the PCB in case that helped. Mind you I'm not sure the material priority settings I was using were actually working and maybe the substrate material was being used. I will play a bit more with that.

Attached is the Python code again as a *.py and *.txt (in case this system has issues with *.py):

sim-antenna-example.py
sim-antenna-example.txt

Thanks for you input :)

[VolkerMuehlhaus]

Thanks for describing some more details, that sounds ok to me, including the single layer reduction.

No, I have no intuitive answer why extra vertical ground has little effect. But as said, from many years of experience in EM simulation of such PCB-integrated antennas, I would not trust your model yet, due to the small margin to the boundary. This loads the antenna different from free space radiation, if the absorber is too close. I will look into your attachment later today, and see what I get.

[VolkerMuehlhaus]

@TerryBarnaby Which design in your model code should I look at?

[TerryBarnaby]

Design2 "python sim-antenna-example.py 2" is the main one.
Design1 is a quaterwave antenna perpendicular to a largish ground plan (that looks ok).
The other designs are with the larger ground plane in different directions and other plays (I have other code with multiple designs/simulation changes and adding antenna matching LC's etc).

[VolkerMuehlhaus]

@TerryBarnaby Ok, I started looking at your code, and see that you use PML_8 boundary, so you need to have this boundary margin:
lambda/4 at 868 MHz plus an extra 8 cells to make room for the PML material.

Due to the very modular structure of your code (great!), it's a bit difficult for me to apply that fix in a clean way. Maybe you as the code owner can implement that on your side and double check? The small margin that I see in present models is definitely wrong.

[VolkerMuehlhaus]

Let's try to call @MustafaAlchalabi for help, he might know better how to make sure we get the extra +8 cells for PML8 when using easyMesh4openEMS.

[MustafaAlchalabi]

Hi Volker, hi Terry,

thanks for tagging me.

You are right: when using PML_8, the mesh needs additional cells for the PML region. In the previous easyMesh4openEMS implementation, boundary_distance did not explicitly account for these extra PML cells.

I have addressed this in the development branch. The intended behavior is now that boundary_distance represents only the air margin, for example λ/4 at the antenna design frequency between the device/antenna and the start of the absorbing boundary region, while the additional PML_N cells are added automatically by the mesher.

The important requirement is that enhance_FDTD_for_auto_mesh(...) is called before FDTD.SetBoundaryCond(...), so that easyMesh4openEMS can detect boundaries such as PML_8.

For the current released version, users still need to add the PML margin manually. But with the current development branch, this should be now handled automatically.

I added more details and an example workflow here:
MustafaAlchalabi/easyMesh4openEMS#8 (comment)

[TerryBarnaby]

Doesn't the "'boundary_distance': [81, 81, 81, 81, 81, 81]," in createMesh() do what is wanted (Well that is about lambda/4 at 868 MHz) ?
I tried that and it certainly added lots of cells, but had little effect on the results. I will try with a larger distance.

[VolkerMuehlhaus]

Doesn't the "'boundary_distance': [81, 81, 81, 81, 81, 81]," in createMesh() do what is wanted (Well that is about lambda/4 at 868 MHz) ?

Yes, except for the extra +8 cells that are required for inserting the PML material. The lambda/4 is air spacing between the DUT and the absorbing boundary.

That said, I also simulated your original design with a lot more air margin, and indeed this has no effect on 868 MHZ radiation: very weak resonance, very weak radiation.

Whatever happens in your hardware measurement (most likely coax feeed becoming part of radiating structure), that is an artefact inserted by the measurement, and not an intrinsic good match from the PCB antenna.

That's a bold claim, you might say, but I double checked by inserting the exact geometry from your openEMS model (STL export from AppCSXCAD) into my commercial FDTD solver, which confirms that there's only very weak resonance at 868 MHz for this geometry. Like your original model, this simulation included no metal loss and no dielectric loss.

[VolkerMuehlhaus]

Once I had the model in my commercial FDTD, I also verified your openEMS findings that placing some extra metal to the copper layer has much more effect on horizontal direction. This is confgirmed, this effect seen in openEMS is also seen using my commercial FDTD. There must be some real physical reason related to fields & radiation.

TL;DR:

1. the findings from openEMS @ 868 MHz look correct
2. The lambda/4 air margin is still important once the antenna radiates, it didn't have much effect here due to poor radiation at 868

[TerryBarnaby]

A bit of feedback for anyone interested. (repeated sent to wrong discussion)
Just tested my NanoVNA-F-V2 measurement with different Ferrites on the lead near the PCB, with different leads, leads at horizontal and vertical orientations, different calibrations, a different but similar PCB etc. The NanoVNA-F-V2 continues to report the SWR as around 5 at 868 MHz no matter what I do. I have tested it with different impedances at the UFL end, and all looks fine.
It could be an issue in the NanoVNA-F-V2 I have, but it reports sensible SWR's for things like 0R, InfinteR, 50R, 33R and 75R etc. when calibrated to the UFL end of its cable.

I have played with many more Openems designs, mainly trying to produce a model that is closer to the reality, but still simple.
One design, Design7, models an inner PCB ground plane using a different PCB stacking geometry to what I think it has along with modelling the 50 Ohm track from the UFL to the antenna feed. This model doesn't have the topside ground plane as yet. However I have noticed:

1. Not having the top ground plane does move the Openems simulation closer the the VNA measurement.
2. Using a different dielectric thickness to the ground plane (0.1mm vs 0.22) makes the Openems simulation fairly closely match the VNA.

Now I know with simulations, it can be "easy" to create modules and simulations that get results that look Ok, but are really quite wrong. However this is a prototype PCB, produced by JLCPCB, and looking at the order I'm not 100% sure if it has the expected 0.22mm dielectric stackup, it could have the 0.1mm. I will try and work out a way to determine this.
With Design7 I get:

Design

Mesh:

Results:

Simulation model:
sim-antenna-example-a3.py

Hope this is of interest, I will continue playing and also change my code to get the actual radiated power which is really what I wanted to do in the first place in order to try different antenna designs.

[VolkerMuehlhaus]

Hi Terry,

one comment on these results: we see quite a strong ripple in data at low frequency, which is unexpected when looking at the layout with rather short lines -> must be simulation artefact.

Usually this ripple is caused by terminating the FDTD run too early, when energy in the model is not yet converced to sufficiently low levels. Looking at your model, I see that you limited the number of timesteps indeed, to only 50k timesteps. This might give poor quality results, if energy convergence is not reached within that timestep limit. I would rather set to to 5M steps or so, to make sure convergence is always reached.

BR, Volker

[TerryBarnaby]

Yes, I did introduce the step limit so I could play with designs in a shorter time and I understand it would compromise the details of the simulation.

With unlimited steps I get the below:
This still shows the SWR being closer to the VNA, but as stated, not sure if this simulation model more accurately reflects the real world measurement sensibly or I am just changing the model to get a result that matches. I am surprised that changing the track impedance a bit of a fairly short track (relative to 868 MHz) would have such an effect on SWR, but that may be reasonable. I haven't had time to consider yet or check what the actual PCB stack dimensions are.