Simple microstrip line impedance mismatch

[GMellar]

I made a simple test PCB with a substrate material. There are two lumped ports and the walls are PML so there is no reflection from the wall where the PCB ends. I experience a line impedance of 41Ohm at center frequency which is too low. I already used that substrate in real world application and measured an impedance for a trace like this and the impedance was not off that much.

I suspect there is some issue with the simulation settings and I need your help. I attached a working python file to elaborate the geometry and run the simulation.

At the moment there seems to be no way to consider loss tangent in openEMS so kappa is zero in that case. In my opinion that should not do much at that frequency.

Simulation.zip

[VolkerMuehlhaus]

@GMellar
From what I can see, you have added 8 mesh cells between the outer simulation boundary and the port, and these are filled with PML_8 absorber. I believe your port then touches the first absorber of the PML, so try adding a few more air cells.

Mesh density: I would add two more dense mesh lines outside the signal trace.

Air around: You don't need that much empty space around the line/substrate volume, something like 2 * ground plane width plus PML margins should be enough.

[VolkerMuehlhaus]

Thanks. I was under the impression that PML_8 fills 8 grid cells with PML.

Yes, that's correct.

The idea was that PML touches the substrate and metal lines so there are no reflections.

This is where you go wrong. If the line is terminated by a lumped port with built-in termination resistor, there must be no additional absorbing wall touching the line.

If you want to run a line into the PML, which is another possible approach, the port must not include a termination resistor.

[GMellar]

This is where you go wrong. If the line is terminated by a lumped port with built-in termination resistor, there must be no additional absorbing wall touching the line.

If you want to run a line into the PML, which is another possible approach, the port must not include a termination resistor.

The first approach was to have a single port and the substrate and line ends are connected to the PML. In the first approach the lumped R on the port had 50ohm.

[VolkerMuehlhaus]

Yes, there are many ways to model this. For my work, I use lumped ports a lot because port parasitics are not a big issue there.

In the tutorials, you will find the MSL_NotchFilter.py example with microstrip ports that end in a PML boundary (instead of port resistor), maybe you want to have a look at the details there. There, the measurement plane is at some distance from the excitation, so that fringing fields can decay and don't distort the measurement. Not sure if you need that level of precision, but the example shows how it can be done.

[GMellar]

Yes, there are many ways to model this. For my work, I use lumped ports a lot because port parasitics are not a big issue there.

Maybe it is better to not connect the substrate and metals to the PML boundary and have two lumped ports. What do you think. In the example the microstrip port adds a lumped resistor box too.

[VolkerMuehlhaus]

Maybe it is better to not connect the substrate and metals to the PML boundary and have two lumped ports. What do you think.

Yes, this is what I do all the time.

[GMellar]

Now I tried to put the excitation port away from the wall and set every wall except the y end wall to PEC. The port is still a port with a feed resistance of 50 ohm. The results do not make sense when driving the line into a PML wall.

Simulation.zip

[VolkerMuehlhaus]

[GMellar]

I forked the automated script generator and made some improvements.

That explains why there is so much reflection. I experimented a bit and made an example with all PEC walls (The simulations runs much faster) and two ports. I noticed the following: When put the ports inside the substrate and let the ports end at the metal lines on both sides the Zin does not match at all. When the ports are higher in z direction (substrate thickness + 2*metal thickness) this gives me reasonable results at center frequency. Impedance is off by 3ohm but I don't expect that much precision. I want to know why it makes such a difference. I attached the example which gives good results.

Simulation.zip

[GMellar]

You are using automation by the scripts created by Lubomir Jagos, so I think he should have a look.

This generator tool does not consider that and that is intentional. You have to know what you are doing (Obviously I didn't). This tool is for mesh and geometry generation and I think it is of great help because the entire model is a parametric model in FreeCAD.

[VolkerMuehlhaus]

I don't use Lubomirs script and I don't use STL. All my models are built from code using Python API, so that I have full control over all details. That's why I better stop here, not knowing what things your automation does internally to boundaries or mesh.

I have notified Lubomir that you are using his script and have a question here in the forum, hopefully he wil show up.

[GMellar]

I don't use Lubomirs script and I don't use STL. All my models are built from code using Python API, so that I have full control over all details. That's why I better stop here, not knowing what things your automation does internally to boundaries or mesh.

For some more complex models like connectors it is easier to see that in a CAD tool but thats my opinion. Anyway you helped me a lot.

[VolkerMuehlhaus]

@GMellar Sure, I just wanted to say that some effects that you see, like lines not running into the PML or ports that must "overshoot" the target position, are related to that workflow and not caused by openEMS API itself. I remember a discussion a year ago with Lubomir where we discussed that "port must touch" vs "port must intrude" and it turned out to be an artefact of his code.