# Tutorial: Patch Antenna Design

Channel: release; package: 1.8.0; source: 7d1e97331656dd74ad0f5471653ac0c5effdff70

Page: https://remilab.ai/rfx/versions/v1.8.0/guide/tutorial-patch-antenna/
Source: https://github.com/bk-squared/rfx/blob/7d1e97331656dd74ad0f5471653ac0c5effdff70/docs/public/guide/tutorial-patch-antenna.mdx

This page connects the model that passes preflight in
[Your First Patch Antenna](https://remilab.ai/rfx/versions/v1.8.0/guide/first-patch/) with the separate patch
cross-check in `validation/crossval/05_patch_antenna.py`.

  This workflow separates setup, execution, and RF evidence. Passing preflight
  or a short run does not validate resonance, return loss, Q, or radiation.

## Recommended workflow

| Stage | Goal | Primary page / script |
|---|---|---|
| 1. Setup check | Size the patch, resolve preflight, and verify finite data from a short run | [Your First Patch Antenna](https://remilab.ai/rfx/versions/v1.8.0/guide/first-patch/) |
| 2. Narrow cross-check | Run the repository's exact rfx/OpenEMS configuration and apply its stated gates | `validation/crossval/05_patch_antenna.py` |
| 3. Radiation / reporting | Move to far-field reporting after the resonance setup is stable | [Far-Field & RCS](https://remilab.ai/rfx/versions/v1.8.0/guide/farfield-rcs/) |

## Baseline geometry

Place the ground plane as an explicit finite PEC box inside the domain, not as a full-domain PEC boundary wall: a finite ground plane radiates into the absorber beneath it, which is the open-radiation problem being modelled, while a PEC bottom wall makes the structure a closed cavity and shifts the resonance high.

## How to run the cross-check

```bash
python validation/crossval/05_patch_antenna.py
```

With the OpenEMS/CSXCAD Python bindings installed, the script prints a console
summary comparing the rfx and OpenEMS resonances and writes a plot plus a JSON
result file. Those bindings are not an rfx dependency: if they are missing, the
script runs the rfx portion and then exits with status 2 behind a SKIPPED banner.
That run is inconclusive and does not report a cross-solver pass.

## Interpretation rules

- The script gates on three things, not two. Two are numeric: "rfx Harminv vs OpenEMS Harminv (resonance frequency) < 20 % (`pass_vs_openems`); rfx internal self-consistency < 5 % (`pass_internal`)" — a Harminv-vs-Harminv frequency agreement, not an |S11| magnitude match. The third is mode identification (`pass_mode_id_rfx`): every measured mode is assigned to its nearest declared TM_mn0 member, and the design member TM100 "must exist, be unique, and be accompanied by a member resolving the other in-plane axis", inside a derived tolerance of 12.4988 % (`sqrt(min adjacent declared member ratio) - 1`). An rfx TM100 error beyond the log-symmetric window [-11.11 %, +12.50 %] "fails the case BY IDENTIFICATION, not by an accuracy gate". The openEMS leg runs the same identifier but is reported, not gated. 
- The 20 % openEMS bound is a smoke bound. The script's own words: it "reports a coarse-mesh resonance and exercises the end-to-end path; it does NOT gate accuracy. Patch-accuracy evidence is delegated to the committed tests". Those tests are the `gate_paths` in `validation/crossval/manifest.json` — `tests/locks/test_patch_edgefed_resonance_harminv.py`, `tests/locks/test_patch_edgefed_s11_passivity.py`, `tests/oracle/test_patch_cavity_eps_oracle.py`, `tests/crossval/test_patch_canonical_farfield_e4.py` — not this script's exit code.
- Those tolerances and results apply only to the script's geometry, mesh,
  frequency band, and metric. They do not validate the First Patch model or the
  [patch gallery diagnostic](https://remilab.ai/rfx/versions/v1.8.0/gallery/patch_antenna/).
- Use impedance-referenced S-parameter results only through a documented port workflow — see [Sources & Ports](https://remilab.ai/rfx/versions/v1.8.0/guide/sources-ports/) for `add_port`. The lumped-port S11 dip in this script is a secondary check with a known single-cell parasitic reactance, so read it as a local dip near the Harminv frequency, not a deep matched null.
- Treat non-uniform mesh variants as limited support unless the page you are following says otherwise.