Question on plotting current density and low frequency simulations

[AJ528]

Hello,

I have been trying to simulate the ground-plane current density vs frequency of a "U"-shaped trace. I expect to see the return current take the shortest path at low frequencies and flow under the trace at higher frequencies. I am encountering a couple issues that I'm hoping others could help me with.

First, I'd like to plot the current density in an intuitive way, but my graphs are not coming out as expected. I would like to plot a 2D heatmap of the current density with a logarithmic color scale in dB. The highest current density would be 0dB at the port, and then the colors would show the various return current paths and the current density along each path. (similar to the images shown in this article: https://medium.com/@andre.kuehne_67586/pcb-return-paths-visualized-622bb952e1ac)

I have capture the current density in a field dump and scale it by the incidental current from the excited port (which should be the maximum value), but I am somehow still getting values above 1.0. I was expecting the density to be 1.0 or lower everywhere. Any ideas what I should be scaling current density by in order to get values between 0 and 1.0? Or am I thinking about this incorrectly?
Here's my current plot, you can see the density is almost 200:

Secondly, in order to see current take the shortest path, I believe the input signal needs to be a low frequency--ideally 10 Hz or so. Because my mesh grid isn't a couple kilometers, the simulation time is cartoonishly long. The best I'm able to get is a 10kHz gaussian pulse, but that still takes like 30 mins to simulate. I know that when wavelength and mesh grid are nowhere close to each other simulation times get long, but I was wondering if there was some approximation/shortcut that can be done when you're essentially simulating at DC frequency. I'm willing to sacrifice accuracy for speed, and I don't care about errors if they're above my impulse frequency. Does such a shortcut exist? Is there a way to configure openEMS to do DC analysis?

If relevant, here is my sim: current_density_sim.zip

If anyone can offer advice, it would be much appreciated!

[VolkerMuehlhaus]

The best I'm able to get is a 10kHz gaussian pulse

It is a bit difficult to spot the relevant settings in your lengthy modular code - did you create a modulated pulse (fc at half maximum frequency) or a regular gaussian pulse with fc = 0 Hz? The fc=0 excitation will have less timesteps for the excitation signal.

That said, of course we need to simulate a sufficient number of timesteps to get accurate convergence of time domain data. Is that where your 30 minutes come from? How did you estimate convergence of field data?

Unfortunately it's quite time consuming to understand the structure of your code, so I didn't look into the other details.

[VolkerMuehlhaus]

I can't help on the low frequency question, I am afraid that FDTD just can't do that very well due to the obvious issue with timestep vs. 1/f

Regarding your current density: I see that you used dump type 13, whereas I successfully used type 12 for my test cases where I show current distribution.

12 : for electric current frequency-domain dump
13 : for total current density (rot(H)) frequency-domain dump

And I'm not sure if you possibly look at the wrong cutting plane and see the vertical port currents, so maybe it's best to check again in Paraview. Your mesh is sooooo very coarse that a small offset in evaluated mesh location will matter.

~~

Side note and rant: why do most users throw the simulation data (which is so important for debugging) into a temp folder, where others need to search for results and model file? I really don't understand this.

[AJ528]

Yeah, it was my understanding too that FDTD struggles when timestep and lambda are far apart. I just wasn't sure if there were any shortcuts you could program in to sacrifice accuracy for speed (e.g. "If you force the timestep to be X it will reduce sim time 10x but accuracy is reduced 2x").

Do you know why it's significantly slower to calculate a step excite vs gauss excite? In my simulation file I only change "SetGaussExcite(f0, fc)" to "SetStepExcite(f0)", but the calculation speed drops from 180 MC/s to 30 MC/s.

Regarding current density: it's interesting to hear you've successfully used type 12. Every time I try to use dump type 12 every data point is "0.0". Is it because my dump box is 2D? I wonder what you're doing differently.
Update: I tried making my dump box 3D (start at TopMetal_zmin and stop at TopMetal_zmax), but the data for dump type 12 is still all 0's.

It's certainly possible that I'm getting vertical port currents; my H5_Viewer class simply plots the overall magnitude of every data point. After making the mesh finer, I think I'm getting more reasonable results.

As for why I put the simulation data in a temporary folder, I was just following what the examples showed. But you make a good point: it would make more sense to put that data in a Simulation_data folder next to the source files. I'll update my framework code to do so!

[VolkerMuehlhaus]

Is it because my dump box is 2D? I wonder what you're doing differently.

Yes, I had created a 3D dump box, and looked at results in Paraview.

With the FDTD field values for E and H offset in space by 1/2 cell, I would not know what is the exact 2D plane to extract, so I never tried to specify 2D dump boxes.

Do you know why it's significantly slower to calculate a step excite vs gauss excite? In my simulation file I only change "SetGaussExcite(f0, fc)" to "SetStepExcite(f0)", but the calculation speed drops from 180 MC/s to 30 MC/s.

I don't know what the effect is. I had done some TDR modelling with StepExcite, where I controlled the time resolution in output data by an additional SetOverSampling(), but I didn't look at cells/s speed.

[AJ528]

Are you able to get dumptype 12 working with the current version of openEMS? I modified my example slightly to use 3D dump boxes and output in paraview format, and it still just reports "0" at every data point for all time (dumptype 13 works fine). If you have an example that shows dumptype 12 working, I would love to see it so I can figure out what I'm doing wrong. Here's my updated sim files:
sim_files.zip

[VolkerMuehlhaus]

If you have an example that shows dumptype 12 working, I would love to see it

I tried this using my RFIC simulation framework, so there's a lot of non-standard code involved. The complete files are attached, the model to run is run_inductor_diffport_dumpfield.py. You will need to install gdspy module because the geometry is read from GDSII layout file.

But anyway, here is the relevant code section:

    # create dump box in frequency domain, type 12 for electric current frequency-domain dump
    CSX = FDTD.GetCSX()

    dump_frequency = 30e9
    Jf = CSX.AddDump('Jf', file_type=0, dump_type=12, frequency=dump_frequency, sub_sampling=[1,1,1])

    dumpMetal = metals_list.getbylayername('TopMetal2')
    start = [allpolygons.xmin, allpolygons.ymin, dumpMetal.zmin]
    stop  = [allpolygons.xmax, allpolygons.ymax, dumpMetal.zmax]
    Jf.AddBox(start, stop)

This places the 3D dump box around the metals on layer TopMetal2, with the parameters shown in the code. The frequency is set as 30 GHz.

Plotting this in ParaView shows the expected currents on the conductor edges, and agrees with current density that I saw in other EM tools. So I think dump type 12 works as expected.

And there is another example in that directory with a microstrip line that has lumped ports on both ends. This also shows a reasonable current density.

Data is stored in the "output" directory next to the model file. The dump file used for plot is the *abs.vtr file.

Hope this helps!

edit: Not sure if that makes any difference, but my lossy metals are implemented as 3D material volumes (not flat sheet, not PEC).

workflow.zip