Directional excitation + modal absorption

[gadiLhv]

Although the PR is not yet implemented, I am already starting to work on the next ones.

1. Uni-directional excitation

To avoid the need to absorb (most) of the excited wave, close to the source, I explored the concept of exciting the mode in a single direction.

For that extent, I also excited the H field, half a step opposite of the propagation direction. In the silly example I wrote, it looks like this

    Ey(:,srcLocX,n) = Ey(:,srcLocX,n) + modeDist*EySrc(n);
    Hz(:,srcLocX - 1,n) = Hz(:,srcLocX - 1,n) + HzSrc(n)*modeDist;

So it looks kind of like this:

Single_Direction_Excite_Snip.mp4

2. Absorb specific modes

At least to the best of my knowledge, in most commercial software, the specific modes or any set of modal decomposition is absorbed. Here is one article, for example

I also had some success in that. It seems that the best performance is obtained with 2 consequtive modal absorbers. Maybe @thliebig can shed a bit of light on that topic.
Surprisingly, it didn't improve the performance using a Mur B.C., in addition! I'm guessing it's computationally easier than a comprehensive modal decomposition.

[thliebig]

Well the commercial solver that I know just combine the excitation with a local PML. That always works and does not require any additional knowledge. What is the issue that you try to solve with this?

[gadiLhv]

What is the issue that you try to solve with this?
In which of the two? Each one is for a different issue.

The single sided excitation is pretty obvious. Be it PML or be it any type of absorber, you are basically blasting it with power from a close proximity source. Apart for a sarcastic remark of it being pretty wasteful, the absorber is limited in it's ability. So why not nip it in the bud? The simulation should also converge faster due to this, I think (the immediate reflected wave is removed).

The modal absorber, that's a whole different story. I noticed that when the excitation is rather wide band ($2f_c \ge f_0$) , the Mur B.C. (and the SA/SIBC expansion) leaves a DC ($f = 0$) component in the simulation. See the video below of the E-field in a coaxial line, with $f_0 = 3.5_{GHz}$ and $f_c = 2.5_{GHz}$.

This could be because of the imperfection of the mode or some other numerical issue, but the Mur BC in it's core is not built to filter out DC. To the best of my knowledge, PML won't solve this issue as well. So I started looking for other strategies. Modal absorption is one of them

DC_res_example.mp4

[gadiLhv]

But since you mentioned it

Well the commercial solver that I know just combine the excitation with a local PML

A common practice is to terminate the T-line or waveguide with a PEC block. How do you do that with a PML?

[thliebig]

Also I think the plane wave excitation does this already, but I would need to check again how. That's like many decades ago I last looked at it.