Triple-feed patch antenna

[mcflatface]

I'm trying out openEMS to simulate a triple feed patch antenna similar to https://www.maartenbaert.be/quadcopters/antennas/triple-feed-patch-antenna/

This is simulating the 24mm patch raised 4mm above a circular ground plane. The patch is fed by three pins through holes in the ground plane. Normally there would be a microstrip feed network on the back-side FR4 that drives these pins but that's not included here, just LumpedPorts on the 3 pins, one of them being excited.

My goal is to extract the 3x3 S-matrix that relates the three feeds. Anyway, I'm finding the results depend significantly on the dimensions of the LumpedPort. Is it possible I'm doing something wrong here. Is the LumpedPort supposed to touch ground as well as the driven pin? I am ensuring there are mesh lines on the LumpedPort boundaries but I could be making other newbie mistakes.

Code: triple_feed_patch.zip
Any suggestions are much appreciated.

Other red flags:

• Radiated power: Prad = 1.3101e-25 Watt (?!)
• Due to the symmetry of the circular design, I would expect symmetric results when driving each of the pins, but I see very different results for each pin.

+Z radiating patch side

-Z Backside (feed ports, which could be microstrip once designed)

Results:

Run outputs:

Create FDTD operator (compressed SSE + multi-threading) 
FDTD simulation size: 67x61x77 --> 314699 FDTD cells  
FDTD timestep is: 3.33611e-13 s; Nyquist rate: 220 timesteps @6.81251e+09 Hz 
Excitation signal length is: 8588 timesteps (2.86505e-09s) 
Max. number of timesteps: 1000000 ( --> 116.442 * Excitation signal length) 
Create FDTD engine (compressed SSE + multi-threading) 
Running FDTD engine... this may take a while... grab a cup of coffee?!? 
[@        4s] Timestep:          385 || Speed:   26.1 MC/s (1.206e-02 s/TS) || Energy: ~1.82e-20 (- 0.00dB) 
[@        8s] Timestep:          880 || Speed:   36.9 MC/s (8.525e-03 s/TS) || Energy: ~7.08e-19 (- 0.00dB) 
[@       12s] Timestep:         1540 || Speed:   50.8 MC/s (6.195e-03 s/TS) || Energy: ~7.03e-17 (- 0.00dB) 
[@       17s] Timestep:         2310 || Speed:   56.9 MC/s (5.528e-03 s/TS) || Energy: ~3.45e-15 (- 0.00dB) 
[@       21s] Timestep:         3080 || Speed:   60.0 MC/s (5.241e-03 s/TS) || Energy: ~5.60e-14 (- 0.00dB) 
[@       25s] Timestep:         3850 || Speed:   56.8 MC/s (5.543e-03 s/TS) || Energy: ~3.06e-13 (- 0.00dB) 
Multithreaded Engine: Best performance found using 5 threads. 
[@       29s] Timestep:         4620 || Speed:   56.9 MC/s (5.527e-03 s/TS) || Energy: ~5.80e-13 (- 0.00dB) 
[@       34s] Timestep:         5390 || Speed:   56.6 MC/s (5.565e-03 s/TS) || Energy: ~3.98e-13 (- 1.63dB) 
[@       38s] Timestep:         6105 || Speed:   53.1 MC/s (5.927e-03 s/TS) || Energy: ~1.15e-13 (- 7.03dB) 
[@       42s] Timestep:         6875 || Speed:   57.4 MC/s (5.478e-03 s/TS) || Energy: ~8.81e-15 (-18.18dB) 
[@       46s] Timestep:         7645 || Speed:   58.1 MC/s (5.417e-03 s/TS) || Energy: ~8.38e-16 (-28.40dB) 
[@       50s] Timestep:         8415 || Speed:   58.0 MC/s (5.422e-03 s/TS) || Energy: ~1.15e-16 (-37.03dB) 
[@       54s] Timestep:         9185 || Speed:   58.8 MC/s (5.349e-03 s/TS) || Energy: ~5.13e-17 (-40.53dB) 
[@       59s] Timestep:         9955 || Speed:   57.5 MC/s (5.472e-03 s/TS) || Energy: ~1.32e-17 (-46.41dB) 
[@     1m03s] Timestep:        10725 || Speed:   59.3 MC/s (5.303e-03 s/TS) || Energy: ~4.68e-18 (-50.93dB) 
Time for 10725 iterations with 314699.00 cells : 63.27 sec 
Speed: 53.34 MCells/s  
calculating far field at phi=[0 90] deg...
 ----------------------------------------------------------------------  
 | nf2ff, near-field to far-field transformation for openEMS  
 | (C) 2012-2023 Thorsten Liebig <thorsten.liebig@gmx.de>  GPL license 
 ----------------------------------------------------------------------  
radiated power: Prad = 1.2015e-25 Watt
directivity: Dmax = 13.0305 (11.1496 dBi)
calculating 3D far field pattern and dumping to vtk (use Paraview to visualize)... 

[VolkerMuehlhaus]

Is the LumpedPort supposed to touch ground as well as the driven pin?

Yes, the lumped port must have one terminal on the backside ground and the other terminal on the patch.

You must remove the ground plane cutout that you created there, and you can also remove the substrate cutout. The port for patch antennas can be placed directly in the dielectric, with no need to create an air cutout.

[mcflatface]

Ah, so lumped port, having a nonzero width in z direction has two terminals, one +z and the other -z face?
I guess if I must eliminate the ground cutout then I must put the lumped port entirely above the ground plane and not behind it like where a microstrip feed would come from (normally the feed passes through the PCB with vias, and there is a ground plane cutout around those vias).

Given the feed pins are 1mm diameter, should I make the lumped port a 1x1x1mm cube sitting on top of the ground plane (between the patch floating 4mm above and the gnd plane)?

[VolkerMuehlhaus]

@mcflatface The lumped port is your source, which excites a voltage between the two terminals. This must be connected so that the excitation voltage is between ground plane and the patch. The port length must fill the entire distance between them, so it must be 4mm. Both ends of the port (= both terminals) must connect to the metal on each end.
Does that make sense?

[mcflatface]

Yes that does, I'm just worried it wouldn't give me accurate S parameters for what a microstrip feed network would see. I suppose I could instead add a small microstrip stub on the back side and then drive that.

If I include the microstrip, I'll follow the example at https://wiki.openems.de/index.php/Tutorial:_Patch_Antenna_Phased_Array.html

Is it necessary to extend the microstrip to the ground edge and put a lumped port on the edge like that? I suppose I can deembed a long microstrip's contribution to the s parameter results.

[VolkerMuehlhaus]

@mcflatface Sorry if I misunderstood your idea, modelling the physical via (with ground cutout) followed by some length of backside microstrip trace is possible for sure. The lumped port doesn't need to be on the edge, just give it some distance from the via so that vertical current in the port and opposite direction vertical current in the via don't couple heavily.

Talking about accuracy: from what I can see, you use cartesian coordinate system, so the circular patch will have staircase boundary, depending on mesh resolution. You can run some convergence test with different mesh resolution at the patch boundary, to see how much resonance frequency is shifted.

[mcflatface]

OK, I'll give the microstrip approach a try and report back. Perhaps the documentation on lumped ports could clarify with a sentence about how the excitation direction is defined and how two opposite faces should be connected to different metals and have mesh lines. It makes more sense now, thanks for the guidance.

[mcflatface]

Adding the microstrip seemed to work well, getting 96.3% radiated efficiency, with very similar results for each of the three feeds, as expected by rotational symmetry. Seeing good consistency even when messing with the mesh.



The pattern is an oblong shape here because we are only exciting one of the three feeds. The idea is to excite all of them with 0,120,240deg differences to get LHCP or RHCP polarization (depending on the order). I'll probably combine the far-fields with phase differences to get that pattern.

Btw, my earlier question about why power is so small, I see makes obvious sense with the time-domain Gaussian pulse of wide bandwidth--the 1 volt default voltage being the peak amplitude of that pulse but only lasting nanoseconds. The power will also vary with the openEMS's choice of timestep, so radiated power jumps around if I mess with the mesh in a way that affects timestep.

I do get some warnings from my mesh, but seem to be getting away with it. I don't yet have a great sense of what the implications are,

found resolution decrease smaller than ratio: 0.2499 < 1/1.625=0.61538
found resolution increase larger than ratio: 1.7768 > 1.625 
found resolution increase larger than ratio: 2.6 > 1.625 
found resolution increase larger than ratio: 4.0017 > 1.625 
found resolution increase larger than ratio: 4.7189 > 1.625 
found resolution decrease smaller than ratio: 0.33333 < 1/1.625=0.61538 

[VolkerMuehlhaus]

The power is not related to pulse length. Excitation is 1V per meter which means 1mV at 1mm port length.

[mcflatface]

Hmm, then I wonder why I see such large variations with mesh choice and timestep.

[VolkerMuehlhaus]

No idea. But you can evaluate the port for frequency domain voltage and current, to check the incident power !?