matlab-create-ai-antenna

# AI Antenna Workflows Skill

You are an expert RF and antenna engineer assisting with AI-accelerated antenna design and pattern reconstruction. Use MATLAB Antenna Toolbox `AIAntenna` for rapid design-space exploration and `patternFromAI` for 3D pattern reconstruction from 2D slices.

## When to Use

- User wants to quickly explore antenna design space without running full EM simulations
- User wants to sweep antenna parameters and instantly see performance (resonant frequency, bandwidth, beamwidth)
- User wants to reconstruct a full 3D radiation pattern from two measured 2D pattern cuts
- User has datasheet or chamber-measured 2D pattern slices and wants a 3D model
- User asks about AI-based antenna design or surrogate models
- User wants to tune antenna dimensions interactively and export the result

## When NOT to Use

- User wants full-wave EM simulation accuracy — use `matlab-design-antenna`
- User wants to design from scratch with `design()` — use `matlab-design-antenna`
- User wants PCB stackup or fabrication — use `matlab-design-pcb-antenna`
- User wants to optimize antenna dimensions with SADEA — use `matlab-optimize-antenna`

## AIAntenna Overview

`em.ai.AIAntenna` creates a pretrained surrogate model of a catalog antenna. Once created, you can instantly:
- Tune geometric parameters and get predicted resonant frequency, bandwidth, beamwidth, and peak radiation
- Explore the full design space without running MoM simulations (milliseconds vs. minutes)
- Export the tuned parameters to a real catalog antenna for full-wave validation

**Key advantage:** 1000x faster than full-wave simulation for parametric sweeps. The AI model predicts antenna performance from geometry in milliseconds.

**Limitation:** Predictions are approximate (surrogate model). Always validate final designs with `exportAntenna` + full-wave analysis.

### Requirements

- Antenna Toolbox
- Statistics and Machine Learning Toolbox

### Supported Antenna Types

| Type | Description |
|------|-------------|
| `"dipole"` | Half-wave dipole |
| `"patchMicrostrip"` | Rectangular microstrip patch |
| `"patchMicrostripCircular"` | Circular microstrip patch |
| `"patchMicrostripElliptical"` | Elliptical microstrip patch |
| `"patchMicrostripInsetfed"` | Inset-fed microstrip patch |
| `"patchMicrostripEnotch"` | E-notch microstrip patch |
| `"patchMicrostripHnotch"` | H-notch microstrip patch |
| `"patchMicrostripTriangular"` | Triangular microstrip patch |
| `"pifa"` | Planar inverted-F antenna |
| `"dipoleHelix"` | Helical dipole |
| `"waveguide"` | Open-ended waveguide |
| `"horn"` | Pyramidal horn |

### Creation

Create an AIAntenna using `design` with `ForAI=true`. **Only the 12 antenna types listed above are supported** — other catalog antennas (helix, yagiUda, vivaldi, monopole, etc.) do not have pretrained AI models.

```matlab
freq = 2.4e9;

% Create AI model from a supported catalog antenna
ant = patchMicrostrip;
antAI = design(ant, freq, ForAI=true);

% Or directly with the antenna constructor
antAI = design(horn, 10e9, ForAI=true);
```

`design(..., ForAI=true)` initializes the AI model with appropriate default dimensions for the target frequency. The direct `em.ai.AIAntenna()` constructor is not supported — always use `design`. If the user requests an antenna type not in the supported list, recommend using `matlab-design-antenna` or `matlab-optimize-antenna` skills instead.

### Core Workflow

```matlab
freq = 1e9;
antAI = design(horn, freq, ForAI=true);

% View default tunable parameters
defaults = defaultTunableParameters(antAI);
disp(defaults)

% Check tunable ranges
ranges = tunableRanges(antAI);
disp(ranges)

% Visualize geometry
figure; show(antAI);

% Get performance predictions (instant)
fRes = resonantFrequency(antAI);
[bw, fL, fU, matching] = bandwidth(antAI);
fprintf("Resonant frequency: %.3f GHz\n", fRes/1e9);
fprintf("Bandwidth: %.1f MHz (%.3f - %.3f GHz)\n", bw/1e6, fL/1e9, fU/1e9);
fprintf("Matching: %s\n", matching);  % "Matched", "Almost", or "Not Matched"
```

### Tuning Parameters

After creation, the AIAntenna object exposes dynamic properties matching the catalog antenna's geometric dimensions. Set them directly:

```matlab
antAI = design(patchMicrostrip, 2.4e9, ForAI=true);

% Check what's tunable and its bounds
ranges = tunableRanges(antAI);
disp(ranges)

% Tune dimensions directly (property names match catalog antenna)
antAI.Length = 0.035;
antAI.Width = 0.045;
antAI.Height = 0.002;

% Instantly check new performance
fRes = resonantFrequency(antAI);
[bw, fL, fU, matching] = bandwidth(antAI);
fprintf("After tuning: fRes=%.3f GHz, BW=%.1f MHz, %s\n", fRes/1e9, bw/1e6, matching);

% Reset to defaults
reset(antAI);
```

### Tunable Ranges

`tunableRanges` returns a table with bounds for each tunable property:

```matlab
ranges = tunableRanges(antAI);           % default: "all" bounds
ranges = tunableRanges(antAI, "strict"); % tighter bounds (higher accuracy)
ranges = tunableRanges(antAI, "loose");  % wider bounds (may reduce accuracy)
```

Use `"strict"` bounds for best prediction accuracy. Parameters outside strict bounds may give unreliable results.

### Peak Radiation and Beamwidth

```matlab
freq = 5.8e9;
ai = design(horn, freq, ForAI=true);

% Peak radiation (gain and direction)
[peakGain, az, el] = peakRadiation(ai, freq);
fprintf("Peak gain: %.2f dBi at az=%.1f, el=%.1f deg\n", peakGain, az, el);

% Beamwidth
[hpbw, angles, plane] = beamwidth(ai, freq);
fprintf("HPBW: %.1f deg (E-plane), %.1f deg (H-plane)\n", hpbw(1), hpbw(2));
```

### Parametric Sweep (Design-Space Exploration)

The main use case — sweep a parameter and plot performance:

```matlab
freq = 2.4e9;
ai = design(patchMicrostrip, freq, ForAI=true);

ranges = tunableRanges(ai, "strict");

% Sweep patch length
lengthRange = linspace(ranges.Length(1), ranges.Length(2), 20);
fResVec = zeros(size(lengthRange));
bwVec = zeros(size(lengthRange));

for k = 1:numel(lengthRange)
    ai.Length = lengthRange(k);
    fResVec(k) = r