1. Introduction

This dataset is designed for training generative models for the purpose of physics-constrained antenna design. Each data sample consists of an antenna image (.png) and a corresponding metadata file (.json) that details its physical properties, performance metrics, and descriptive captions.

2. Directory Structure

The dataset is organized as follows:

antenna_dataset/
├── README.md             <-- This documentation file
├── images/
│   ├── sample_001.png
│   └── ...
└── metadata/
    ├── sample_001.json
    └── ...

Each image in the images/ directory has a corresponding JSON file with the same base filename in the metadata/ directory.

3. Metadata JSON File Structure

Each .json file contains a single object with the following top-level keys:

---

id

• Type: String
• Description: A unique identifier for the data sample. It matches the base filename of the corresponding image and metadata files.
• Example: "patch_antenna_001"

---

topology

• Type: String
• Description: The fundamental engineering classification of the antenna.
• Possible Values: "Microstrip Patch", "Monopole", "Vivaldi", "Horn", "Yagi-Uda", etc.

---

style

• Type: List of Strings
• Description: A list describing the visual or conceptual style, crucial for creative generation.
• Example: ["standard"], ["fractal", "bio-inspired:leaf"]

---

performance

• Type: Object

• Description: Contains precise, quantitative performance metrics obtained from electromagnetic simulations. These are the ground truth values for calculating physics-based loss.

  • center_frequency_ghz

    • Type: Number (Float)
    • Description: The primary resonant frequency in Gigahertz (GHz).

  • fbw_percent

    • Type: Number (Float)
    • Description: Fractional Bandwidth in percent (%), calculated as (Bandwidth / Center Frequency) * 100.

  • gain_dbi

    • Type: Number (Float)
    • Description: The antenna gain in dBi (decibels relative to an isotropic radiator).

---

classes

• Type: Object

• Description: Contains qualitative, categorical labels derived from the quantitative performance data. Useful for simplified text prompts and classification tasks.

  • bands

    • Type: List of Strings
    • Description: A list of the frequency bands of operation. Designed as a list to support multi-band antennas. UHF：0.3–1 GHz, L：1–2 GHz, S：2–4 GHz, C：4–8 GHz, X：8–12 GHz, Ku：12–18 GHz, K：18–26.5 GHz, Ka：26.5–40 GHz, V：40–75 GHz, W：75–110 GHz.
    • Possible Values: "L", "S", "C", "X", "Ku", "Ka", "mmW"

  • fbw_class

    • Type: String
    • Description: The categorical classification of the Fractional Bandwidth. UltraNarrow (<2%), Narrow (2–5%), Medium (5–10%), Wide (10–20%), UltraWide (>20%)
    • Possible Values: "UltraNarrow", "Narrow", "Medium", "Wide", "UltraWide"

  • pattern

    • Type: String
    • Description: The primary direction of radiation.
    • Possible Values: "Broadside", "Endfire", "Omnidirectional", "MultiBeam"

  • pol

    • Type: String
    • Description: The polarization of the radiated wave.
    • Possible Values: "Linear", "Circular", "Dual", "Elliptical"

  • gain_class

    • Type: String
    • Description: The categorical classification of the antenna's gain. Low (<2 dBi), Mid (2–8 dBi), High (8–15 dBi), Very-High (>15 dBi).
    • Possible Values: "Low", "Mid", "High"

---

physical

• Type: Object

• Description: Describes key physical properties and environmental constraints.

  • diameter_mm

    • Type: Number (Float)
    • Description: The largest bounding box dimensions of the antenna in millimeters (mm).

  • Height_mm

    • Type: Number (Float)
    • Description: The bounding box height dimensions of the antenna in millimeters (mm).

---

training_captions

• Type: List of Strings
• Description: Pre-generated, natural language sentences describing the antenna. These are fed to the text encoder of the generative model during training. Multiple variations enhance model robustness.
• Example: ["A photo of a Microstrip Patch antenna...", "High-performance antenna design..."]