28 GHz case study¶
examples/06_full_pipeline_case_study.py runs the whole chain for a
5G mmWave scenario: unit-cell full-wave simulation, coupling-aware 8x8
array patterns, a 200 m link budget at 400 MHz bandwidth, and a
40-point trade study over array size and transmit power. EdgeFEM is
required for this one:
pip install "apab[edgefem]" # needs CMake + Eigen3 (see README)
python examples/06_full_pipeline_case_study.py
What the pipeline computes¶
- Unit cell (EdgeFEM) — a grounded patch on a Rogers-class substrate, swept 26–30 GHz with Floquet ports. EdgeFEM excites the cell with a plane wave, so for a grounded patch the reflection magnitude stays near one across the band and resonance appears as a reflection-phase transition, not a magnitude dip. The example detects resonance from the maximum phase derivative.
- Feed-port model (analytical) — return loss and impedance bandwidth come from a cavity model of the probe-fed patch. The two models cross-validate: FEM phase resonance at 28.50 GHz vs the analytical 28.40 GHz, under 0.4 % apart.
- Scan behavior — Floquet reflection versus scan angle characterizes the element in its array environment, feeding the coupling-aware pattern computation.
- Array pattern — 8x8 with taper, steered cuts, directivity and sidelobe level.
- System trade study — Latin-hypercube sampling over array size (4–16 per axis) and per-element power (10–500 mW) against a 5G NR comms scenario, with Pareto extraction.
The run writes six figures to examples/output/ (reflection phase,
feed-port S11, scan behavior, link budget, trade study, array layout).
Why the hybrid modeling approach¶
Floquet-port simulation answers array-environment questions (scan impedance, phase response, blindness onset) that a feed-port model cannot; the analytical feed model answers match-bandwidth questions that a plane-wave excitation cannot. The case study keeps both and checks one against the other — the same split you would use with a commercial solver before committing to a fabricated design.
Companion materials¶
The repository includes a LaTeX write-up of this study
(examples/case_study_paper.tex) and the agent-driven variant of the
same workflow in the quickstart, where the LLM
sequences these steps from a natural-language request instead of a
script.