📡 RadPat-50K: A Large-Scale Benchmark for Antenna Array Radiation Patterns

Description

📡 RadPat-50K: A Large-Scale Benchmark for Antenna Array Radiation Patterns RadPat-50K is a large-scale synthetic dataset of 50,000 radiation patterns generated for Uniform Linear Arrays (ULA). Each sample includes polar plots and rectangular plots of the array factor, along with metadata for design and performance parameters. 📊 Dataset Highlights • Size: 50,000 radiation patterns • Array Elements (N): 4, 8, 12, 16, 24, 32, 48, 64 • Element Spacing (λ): 0.25, 0.5, 0.75, 1.0 • Steering Angles (°): –60, –45, –30, –15, 0, 15, 30, 45, 60 • Weighting Schemes: o Uniform o Binomial o Cosine o Kaiser o Hamming o Hann o Blackman o Exponential • Variations (for diversity): o ⚡ Amplitude noise o 📐 Phase noise o 🎯 Steering jitter 🔬 Applications This dataset is well-suited for research in machine learning, deep learning, and signal processing, including: • 📑 Antenna pattern classification • 🎛️ Beamforming analysis • 🚨 Grating lobe detection • 🏗️ Data-driven array design ________________________________________ 🧾 Metadata per Record Each radiation pattern entry includes: • Antenna parameters: number of elements, element spacing, steering angle, weighting scheme • Noise parameters: amplitude noise, phase noise, steering jitter 🧾 Performance Metrics • 📡 Directivity • 📏 Half-Power Beamwidth (HPBW) • 📐 Main Lobe Angle 🤖 Benchmark Potential RadPat-50K provides a standardized benchmark for: • Training and evaluating classification & regression models • Developing robust antenna array designs under noise conditions • Advancing AI-driven RF and antenna research 1. Classification Benchmarks • Task A: Weighting Scheme Classification o Input: Radiation pattern image o Output: Weighting scheme label (uniform, cosine, blackman, etc.) • Task B: Number of Elements Classification o Input: Radiation pattern image o Output: Class label (N = 4, 8, 12, 16, 24, 32, 48, 64) • Task C: Spacing Classification o Input: Radiation pattern image o Output: Class label (d = 0.25λ, 0.5λ, 0.75λ, 1.0λ) • Task D: Joint Classification o Input: Radiation pattern image o Output: Multi-task prediction (N, spacing, weighting, steering angle category). ________________________________________ 2. Regression Benchmarks • Task E: Directivity Prediction o Input: Radiation pattern image o Output: Directivity (linear or dB). • Task F: HPBW Prediction o Input: Radiation pattern image o Output: Half Power Beamwidth in degrees. ________________________________________ 3. Multi-Label / Structured Prediction • Task G: Parameter Recovery o Input: Radiation pattern image o Output: A set of antenna parameters (N, spacing, weighting scheme, steering angle). ________________________________________ 4. Vision-Language Benchmarks (VQA-style) • Task H: Antenna Q&A o Input: (Image + Question) o Example Qs:  "What is the main lobe direction?"  "Which weighting scheme is applied?"  "How many array elements are used?" o Output: Answer (text).

Steps to reproduce

The RadPat-50K dataset can be regenerated from first principles using standard array factor formulations. The following procedure was used: Environment Setup Install Python (≥3.9) along with the required packages: numpy, scipy, matplotlib, and pandas. Array Configuration A uniform linear array (ULA) was considered with varying numbers of elements and inter-element spacings. The number of elements was selected from N = {4, 8, 12, 16, 24, 32, 48, 64}. Inter-element spacings were chosen from values such as 0.25λ, 0.5λ, 0.75λ, and 1λ. Steering angles were selected from −60°, −45°, −30°, −15°, 0°, 15°, 30°, 45°, and 60°. Weighting Schemes Multiple excitation tapers were applied to control sidelobe levels and beamwidth. The following weighting functions were used: uniform, binomial, cosine, Kaiser, Hamming, Hann, Blackman, and exponential. Pattern Computation For each configuration, the array factor was computed and radiation patterns were generated in both rectangular (dB vs. angle) and polar coordinates. Derived antenna parameters such as gain, directivity, and half-power beamwidth (HPBW) were extracted. Sidelobe levels (SLL) were not included. Variations for Diversity To simulate practical conditions, additional variations were introduced: ⚡ Amplitude noise 📐 Phase noise 🎯 Steering jitter For these variations, only CSV metadata files were generated (no images). Data Generation A complete sweep over all combinations of array elements × steering angle × element spacing × weighting scheme was performed. The resulting radiation patterns were stored as JPG images (750×750 resolution), while the corresponding numerical attributes were saved in CSV files. 📂 Dataset Description The RadPat-50K dataset is organized into three primary files for ease of experimentation: • 📁 Rect_50K.zip → Contains 50,000 rectangular plot images of antenna array radiation patterns. • 📁 Polar_50K.zip → Contains 50,000 polar plot images of antenna array radiation patterns. • 📑 Metadata_50K.csv → A structured metadata file providing detailed information for each sample. 📊 Metadata Contents Each record in Metadata_50K.csv includes: • id → Unique identifier for each sample • N → Number of array elements • spacing_wavelengths → Element spacing in wavelengths • weights → Applied weighting scheme • steering_deg_nominal → Nominal steering angle (in degrees) • steering_deg_noisy → Steering angle with noise (in degrees) • amp_noise_std → Standard deviation of amplitude noise • phase_noise_std → Standard deviation of phase noise • grating_lobe → Presence of grating lobe (True / False) • image_rect → File name of the corresponding rectangular plot • image_polar → File name of the corresponding polar plot • D_peak_dBi → Peak directivity (in dBi) • HPBW_deg → Half-Power Beamwidth (in degrees) • main_lobe_angle_deg → Main lobe angle (in degrees) ⚡ Quick Experimentation Subset (RadPat-10K) For rapid prototyping and experimentation, we provide a 10K subset of the full dataset.

Institutions

• Bapatla Engineering College

Categories

Licence