missing python interface DumpFF2VTK

[LubomirJagos]

Hi, in current python interface there is missing DumpFF2VTK function to be able generate 3D field vizualization from NF2FF function so I have reworte matlab/octave one and test it, here it is:

import math

def DumpFF2VTK(farfield, t, a, filename):
    """
       Create .vtk file
       params:
           farfield: 2D array of values of field
           t:        theta angles in radians
           a:        phi angles in radians
           filename: output file name
    """

    with (open(filename, 'w') as outFile):
         outFile.write(f"# vtk DataFile Version 3.0\n")
         outFile.write(f"Structured Grid by python-interface of openEMS\n")
         outFile.write(f"ASCII\n")
         outFile.write(f"DATASET STRUCTURED_GRID\n")

         outFile.write(f"DIMENSIONS 1 {len(t)} {len(a)}\n")
         outFile.write(f"POINTS {len(t)*len(a)} double\n")

         for na in range(len(a)):
             for nt in range(len(t)):
                 val1 = farfield[nt][na]*math.sin(t[nt])*math.cos(a[na])
                 val2 = farfield[nt][na]*math.sin(t[nt])*math.sin(a[na])
                 val3 = farfield[nt][na]*math.cos(t[nt])
                 outFile.write(f"{val1} {val2} {val3}\n")

         outFile.write(f'\n\n');
         outFile.write(f'POINT_DATA {len(t)*len(a)}\n');
         outFile.write(f'SCALARS gain double 1\n');
         outFile.write(f'LOOKUP_TABLE default\n');
         for na in range(len(a)):
             for nt in range(len(t)):
                 outFile.write(f"{farfield[nt][na]}\n")

[oberstet]

this is great, love it!

I want to try that using a most simple model - a center-fed half wave dipole - which is well understood and for which I can already compute a number of interesting data/plots with openEMS, and I could (basically) verify values expected from theory - which is what creates "trust" for me.

I have now added some some directivity measures, however I do not fully understand those pieces yet though.

Hence, I don't have a valid/trusted basis yet to add even more, like a 3D radiation plot! Which I do want;)

In any case, I will do some experimentation, and I also post complete code and results below.

---

Code: complete working and self-contained code is here https://gist.github.com/oberstet/ce57c9579591d6d1b1f9222afa00bccc

Results:

---

Open questions (well, "open" for me;):

1. One piece in particular that I do not understand is the selection of frequencies (pls see https://gist.github.com/oberstet/ce57c9579591d6d1b1f9222afa00bccc#file-dipole-py-L417 .. also, the results in the comments there ... how can "Efficiency: nu_rad" be >100%?), the mesh resolution and the radius for NF2FF recording box / computation
2. Another piece that I have trouble with is "theta" and "phi" .. my dipole is oriented along the X-axis. I don't know if that is "common", and in any case, I haven't found a definition of theta/phi in the docs:

# CalcNF2FF(sim_path, freq, theta, phi, radius=1, center=[0, 0, 0], outfile=None, read_cached=False, verbose=0)
# theta/phi – array like – Theta/Phi angles to calculate the far-field

Well yes, and that's what it is used for, but what is the definition of Theta/Phi? Obviously, it could mean "theta = angle between FF and Z axis" .. or it could mean sth else.

Ultimately, I want to render a picture like the following (which aligns the dipole along Z axis, not X like I do):

[VolkerMuehlhaus]

To explain the graded mesh strategy that I would use, I have attached an example. This is essentially your model, with the changes that I mentioned above, and different meshing in x direction.

One other minor change: in my opinion the port gap is part of the total dipole length, so I changed the dipole arm length code.

Model is attached. Note that my code ends after showing the model in AppCSXCAD, I removed the calculation part because I only wanted to visualize the model and mesh.

dipole3_reduced_volker.zip

[oberstet]

This is great feedback, and your explanations help me a lot!! And the fact that you even provided demo code is fantastic=) Thank you so much! I will need to digest that, and try your code.

---

your mesh is very coarse in the port region, and doesn't align with the ends of the metal or port. This is why it didn't work without cheating by the oversize value.

aaahhh, right! got it.

---

I am really missing some basics. Which isn't a surprise, as I am completely new to RF. I am coming from a different world: digital / software. I've got quite a lot of know how and experience there, also in open source, and also in Python specifically, which is one reason I was naturally drawn to openEMS, with the Python interface.

The way and reasons of why I ended here (creating some handheld radio etc) are strange-crazy, but too long to lay out here, and they don't matter (if you want to know, drop me a PM, I can share).

Now, being a developer, I am also kinda "lazy" ... doing things manually isn't my thing ... why not write a program that writes a program or searches through parameters? Why should I fiddle around manually for hours? But you are of course right with your hint that I might get results, but no explanation or insight. True.

I learned that already the hard way .. e.g. I couldn't get any dipole working initially .. there is no example .. and I was trying to figure out how to position, connect and drive two lumped ports - because a dipole has two antenna elements, right? What I was missing is: even a monopole also has two elements, one of which just turns out to be something acting as a ground plane, a reflector, and the goal is to make the reflected antenna image constructively overlay the primary image. Well, ok, then of course 1 lumped port in both cases makes a lot more sense;) Basics, I know.

Your model is somewhat confusing to me - you have defined 1µm wire diameter, is that what you wanted?

No, this whole dipole thing is just to establish the most simple antenna as a reference, and fully understand that, before moving to my actual goal of auto-optimizing an antenna (during which the reference will come in handy).

If you can explain your modelling task (resolve some finite dipole diameter or just simulate a thin wire?), I can then try to help with meshing.

my actual goal is

A PCB antenna (4-layer FR4) for the whole range ~400-950 MHz, omnidirectional, for a handheld radio, and with antenna size < ~ 70x10mm (thus, ~ lambda/8).

I initially thought: can't be too hard, should be done in 2 weeks;) Yes, hyperbole, I now realize it isn't that simple .. for various reasons. But I'm making progress, and if so only incrementally, I will reach the goal, I'm sure;)

So here a pic drawn in KiCad where I started from 2 weeks ago, and this layout is just me taking a wild guess;) By now, I'm not sure a planar helical will cut it, mostly due to bandwidth, anyways, once I got the auto-optimizer controlling a suitable geometry generator, I should be able to get more sleep;)

[oberstet]

alright, I was interrupted by sth else, but now I got back:

• tested your code verbatim: works!
• I've analyzed and integrated all your changes: works!
• I've re-optimized the opt/correction factor (because the changed dipole wire radius and gap) - the new value is a tiny bit below the old value (and also the impedance is 0.2 ohm lower):
• new code is here: https://gist.github.com/oberstet/8877417387c622f59700612818a1484d

take away: failing on proper mesh setup can evoke strange artifacts, like with my port overlap stuff. I learned something;) thanks a lot!

# use with dipole_wire_radius = 0.001
# opt_factor = 0.8651

# use with dipole_wire_radius = 2.0
# Found resonance frequency at 446.1 MHz with -91.6 dB at 70.8 Ohm
# Dipole (lambda/2) length is 285.8 mm
opt_factor = 0.8506

the optimizer aims to hit a fixed point of the S11 function, that is, it hits an opt_factor and feed impedance such that the simulated/computed resonance frequency and impedance for the excitation frequency (which is set to the center of the desired band) is identical to the former values.

I have no clue if this is "usually" done, or what, and what it means, but for me, it was natural (from a pure math angle) to aim for a fixed point, and practically, it achieves quite good results (see below)

================================================================================

Found resonance frequency at 446.1 MHz with -91.6 dB at 70.8 Ohm
Dipole (lambda/2) length is 285.8 mm

================================================================================

S11 at frequency 446.0 MHz is -64.4 dB at 70.8 Ohm [index 999]
S11 at frequency 446.1 MHz is -91.6 dB at 70.8 Ohm [index 1000]
S11 at frequency 446.2 MHz is -65.1 dB at 70.8 Ohm [index 1001]


S11 at frequency 412.6 MHz is -10.0 dB at 54.3 Ohm
S11 at frequency 492.9 MHz is -10.0 dB at 102.1 Ohm

S11 at frequency 426.9 MHz is -15.0 dB at 60.7 Ohm
S11 at frequency 469.0 MHz is -15.0 dB at 85.0 Ohm

S11 at frequency 435.0 MHz is -19.9 dB at 64.8 Ohm
S11 at frequency 458.3 MHz is -20.0 dB at 78.0 Ohm

S11 at frequency 439.7 MHz is -24.9 dB at 67.3 Ohm
S11 at frequency 452.8 MHz is -25.0 dB at 74.7 Ohm

S11 at frequency 442.5 MHz is -29.9 dB at 68.8 Ohm
S11 at frequency 449.8 MHz is -29.9 dB at 73.0 Ohm

S11 at frequency 444.9 MHz is -39.5 dB at 70.1 Ohm
S11 at frequency 447.3 MHz is -39.7 dB at 71.5 Ohm

S11 at frequency 445.7 MHz is -49.1 dB at 70.6 Ohm
S11 at frequency 446.5 MHz is -49.2 dB at 71.0 Ohm
================================================================================

---

finally I also oriented the dipole along Z-axis, and I am now figuring out this directivity stuff and hopefully will also be able to render a 3D radiation pattern soon

what does this plot tell me?

[VolkerMuehlhaus]

Hi Tobias,

maybe you should open a new thread, as we are off-topic and there are many items to discuss.

For me, your code approach looks somewhat unusual. When I design PCB antennas, I use wideband excitation for FDTD simulation, that works fine and is much easier than your approach.

Extracting the resonance frequency by code is fine, but overall your model code is somewhat overloaded for the basic steps that we need to cover now. I was unable to get your latest code up and running here, so not sure what goes wrong.

One important detail is mesh density around the radiator/wire, in your last model that was much too coarse except for meshing in longitudinal wire direction. The fields at the end of the wire was modelled with big rectangles (large mesh size perpendicular to the wire) that are much too large to resolve the actual field distribution at the open end (field gradients are very strong there!)

Regarding the pattern that you show above: that is a 2D cut of the antenna pattern, not sure what direction (what cutting plane) we see here. Dipole directivity should peak at ~ 2.15 dBi in the maximum of the "donut" antenna pattern, so the -40dBi look unexpected.

But let's open another thread, and maybe someone else can jump in to support you.

Best regards
Volker

[oberstet]

+1! sorry for OT noise. #152

[oberstet]

alright, I've added above function and the VTK file dump at the end to above example. full updated source is here:

https://gist.github.com/oberstet/8e771e2466395d6f164d93fe8f7a7d32

this runs into the following error:

Traceback (most recent call last):
  File "/home/oberstet/scm/typedefint/rfminer/_work/poc_test_dipole2.py", line 577, in <module>
    directivity = nf2ff.Prad[0] / nf2ff.Prad * 4 * math.pi
                  ^^^^^^^^^^
AttributeError: 'nf2ff' object has no attribute 'Prad'

as the nf2ff object attribute dump https://gist.github.com/oberstet/8e771e2466395d6f164d93fe8f7a7d32#file-dipole2-py-L571 above shows, the nf2ff object neither has a Prad nor a P_rad attribute, but only these attributes:

['CSX',
 'CalcNF2FF',
 '__class__',
 '__delattr__',
 '__dict__',
 '__dir__',
 '__doc__',
 '__eq__',
 '__format__',
 '__ge__',
 '__getattribute__',
 '__getstate__',
 '__gt__',
 '__hash__',
 '__init__',
 '__init_subclass__',
 '__le__',
 '__lt__',
 '__module__',
 '__ne__',
 '__new__',
 '__reduce__',
 '__reduce_ex__',
 '__repr__',
 '__setattr__',
 '__sizeof__',
 '__str__',
 '__subclasshook__',
 '__weakref__',
 'center',
 'directions',
 'dump_mode',
 'dump_type',
 'e_dump',
 'e_file',
 'freq',
 'h_dump',
 'h_file',
 'mirror',
 'name',
 'phi',
 'start',
 'stop',
 'theta']

[oberstet]

I just recognize that you are using nf2ffBoxList to create the NF2FF recording box, as in

#
#	nf2ffBoxList[<name>] - NF2FF box structure which should be calculated
#       INPUT ANGLES ARE IN DEGREES for python interface!
#
nf2ff = nf2ffBoxList['nf2ffbox_1_-_farfield_1'].CalcNF2FF(Sim_Path, plotFrequency, thetaRange, phiRange, outfile='3D_Pattern.h5', verbose=True, read_cached=False)

whereas I am using

nf2ff = fdtd.CreateNF2FFBox("nf2ff-box", start=start, stop=stop, opt_resolution=[mesh_res_farfield] * 3)

Where does nf2ffBoxList come from?

[oberstet]

alright, is seems I've got that fixed https://gist.github.com/oberstet/3ec68a8348494452c2347764f1920bdd

...
================================================================================


Calculating the 3D far field...
Analyzing far-field at radius 126
Analyzing far-field for 1205 frequencies:
array([4.12843245e+08, 4.12910160e+08, 4.12977075e+08, ...,
       4.93275075e+08, 4.93341990e+08, 4.93408905e+08])
nf2ff: Reading planes: /tmp/dipole/nf2ff-box_E_0.h5 & /tmp/dipole/nf2ff-box_H_0.h5
nf2ff: Data-Size: 1x7x7
nf2ff: Analysing far-field for 1205 frequencies.  
nf2ff: Reading planes: /tmp/dipole/nf2ff-box_E_1.h5 & /tmp/dipole/nf2ff-box_H_1.h5
nf2ff: Data-Size: 1x7x7
nf2ff: Analysing far-field for 1205 frequencies.  
nf2ff: Reading planes: /tmp/dipole/nf2ff-box_E_2.h5 & /tmp/dipole/nf2ff-box_H_2.h5
nf2ff: Data-Size: 8x1x7
nf2ff: Analysing far-field for 1205 frequencies.  
nf2ff: Reading planes: /tmp/dipole/nf2ff-box_E_3.h5 & /tmp/dipole/nf2ff-box_H_3.h5
nf2ff: Data-Size: 8x1x7
nf2ff: Analysing far-field for 1205 frequencies.  
nf2ff: Reading planes: /tmp/dipole/nf2ff-box_E_4.h5 & /tmp/dipole/nf2ff-box_H_4.h5
nf2ff: Data-Size: 8x7x1
nf2ff: Analysing far-field for 1205 frequencies.  
nf2ff: Reading planes: /tmp/dipole/nf2ff-box_E_5.h5 & /tmp/dipole/nf2ff-box_H_5.h5
nf2ff: Data-Size: 8x7x1
nf2ff: Analysing far-field for 1205 frequencies.  
Radiated power: P_rad = 3.3366863684313314e-21 W
Directivity: D_max = -42.45806364612437 dBi
Efficiency: nu_rad = 24.669637065384926 %
Theta_HPBW = 43.0 °
['CSX',
 'CalcNF2FF',
 '__class__',
 '__delattr__',
 '__dict__',
 '__dir__',
 '__doc__',
 '__eq__',
 '__format__',
 '__ge__',
 '__getattribute__',
 '__getstate__',
 '__gt__',
 '__hash__',
 '__init__',
 '__init_subclass__',
 '__le__',
 '__lt__',
 '__module__',
 '__ne__',
 '__new__',
 '__reduce__',
 '__reduce_ex__',
 '__repr__',
 '__setattr__',
 '__sizeof__',
 '__str__',
 '__subclasshook__',
 '__weakref__',
 'center',
 'directions',
 'dump_mode',
 'dump_type',
 'e_dump',
 'e_file',
 'freq',
 'h_dump',
 'h_file',
 'mirror',
 'name',
 'phi',
 'start',
 'stop',
 'theta']
(cpy311_7) oberstet@intel-nuci7:~/scm/typedefint/rfminer$ tree /tmp/dipole/
/tmp/dipole/
├── 3D_Pattern_CPLH.vtk
├── 3D_Pattern_CPRH.vtk
├── 3D_Pattern_E_norm.vtk
├── 3D_Pattern.vtk
├── et
├── ht
├── nf2ff-box_E_0.h5
├── nf2ff-box_E_1.h5
├── nf2ff-box_E_2.h5
├── nf2ff-box_E_3.h5
├── nf2ff-box_E_4.h5
├── nf2ff-box_E_5.h5
├── nf2ff-box_H_0.h5
├── nf2ff-box_H_1.h5
├── nf2ff-box_H_2.h5
├── nf2ff-box_H_3.h5
├── nf2ff-box_H_4.h5
├── nf2ff-box.h5
├── nf2ff-box_H_5.h5
├── port_it_1
└── port_ut_1

0 directories, 21 files
(cpy311_7) oberstet@intel-nuci7:~/scm/typedefint/rfminer$ 

---

let me now check paraview .. I guess ..

[LubomirJagos]

3D_pattern.vtk is radiation pattern
3D_Pattern_E_norm.vtk is electric field

[oberstet]

this is what I've got (the output generated by above script)

(cpy311_7) oberstet@intel-nuci7:~/scm/typedefint/rfminer$ du -hs /tmp/dipole/3D_Pattern.vtk
2,4M	/tmp/dipole/3D_Pattern.vtk
(cpy311_7) oberstet@intel-nuci7:~/scm/typedefint/rfminer$ du -hs /tmp/dipole/*.vtk
2,4M	/tmp/dipole/3D_Pattern_CPLH.vtk
2,4M	/tmp/dipole/3D_Pattern_CPRH.vtk
2,9M	/tmp/dipole/3D_Pattern_E_norm.vtk
2,4M	/tmp/dipole/3D_Pattern.vtk

how do I make paraview digest that and show something that makes sense?

---

[oberstet]

mmmh .. I mean, it does look interesting;) what do I need to do in paraview, which of the 4 files to open? what rendering/filter settings to set? sorry, I'm pretty much a paraview newbie ...