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

📡 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).