Wave port#53
Conversation
|
|
@dot-cross why do you think S21 for waveport is 4dB more lossy? this simple cpw line should be very close to no loss. can you look at what mode is being excite in the waveport? maybe it's the wrong mode |
Not sure why is 4dB more lossy. The mode is set to one, which by Palace documentation correspond to the biggest wave number. |
This is for mode 2 @vvahidd
|
|
Maybe the waveport region should be made larger? I mean, perhaps the 2D eigenvalue solver is converging to the correct numerical solution, but the domain where the problem is defined (which is a truncation of the real unbounded domain) is too small. Maybe even the airbox is not large enough. For example, I know that in antenna problems we need an air region that is at least |
|
We could also add the visualization of the mode for verification purposes. For an example reference, here I have a CPW simulation where the airbox seems to be too small, as the eigenmode is non-zero at the top: 
I think that for this PR just making the domain (and the waveport) large enough should be sufficient, but eventually we need to bring the viz server back to life :-D |
dot-cross
force-pushed
the
wave_port
branch
2 times, most recently
from
4c8f3bd to
5eb90f6
Compare
Lumped port displaced inside the board. Margin set to zero for meshing in both cases.
|
|
I've been making minor changes in the waveport geometry. I added a flag to give it a "maximum size", and re run the simulations with the following settings:
The results are still unsatisfactory, with the waveport S-parameters ranging from -2 to -6. However, looking more closely at the waveport mesh I think the problem is that we do need an airbuffer zone at the sides 
|
|
In the last commit I added the option to have an airbuffer zone only to one side of the simulation domain. The mesh now looks like this: 
and the S-parameters are a lot closer to convergence: 
|
…=0 to waveport vs lumped port comparison
|
I refactored some visualization functions from the viz notebook to be able to use them here, in order to compare the fields we're getting from lumped vs waveports. 
Both plots look like there is room for improvement ;-) I think we could check the boundary conditions on the floor, as the fields are clearly non-zero there, so maybe we are getting some sort of spurious reflections. |
|
Ok, it seems this was a meshing issue. Using graded meshes with very fine 
I'll run some more tests to find a good compromise and come up with numbers. |
|
I ran some more tests, and the full effect is obtained by the combination of adding air below the substrate and using a finer mesh, which is specially important in the waveport example (it requires a finer mesh). The air below the substrate also helps reduce the lumped port simulation losses. |
|
I made some further improvement to graded meshing. Both simulation types now run in about ~3 minutes (local run) and show no loss throughout the frequency range. 
|
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## main #53 +/- ##
=======================================
Coverage ? 46.74%
=======================================
Files ? 57
Lines ? 6197
Branches ? 1079
=======================================
Hits ? 2897
Misses ? 3003
Partials ? 297 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
| - substrate_thickness: Thickness below z=0 in um (default: 2.0) | ||
| - air_above: Air box height above top metal in um (default: 200) | ||
| - air_below: Air box height below substrate/oxide in um | ||
| (default: 0) |
There was a problem hiding this comment.
| (default: 0) | |
| - air_below: Air box height below substrate/oxide in um (default: 0) |
4 participants
Added support for waveports. Modified palace_demo_cpw notebook to simulate and plot both lumped and waveports.