📡 Welcome to the RadioDiff Family

Radio map construction via generative diffusion models — UNIC Lab, Xidian University

---

🔷 Base Backbone

RadioDiff — The foundational diffusion model for radio map construction. 📄 Paper | 💻 Code |

---

🔬 Physics-Informed Extensions

RadioDiff-k² — PINN-enhanced diffusion guided by the Helmholtz equation. 📄 Paper | 💻 Code |

iRadioDiff — Indoor radio map construction with physical information integration. 📄 Paper | 💻 Code |

---

⚡ Efficiency & Dynamics

RadioDiff-Turbo — Efficiency-enhanced RadioDiff for accelerated inference. 📄 Paper |

RadioDiff-Flux — Adaptive reconstruction under dynamic environments and base station location changes. 📄 Paper |

---

🌐 Extended Scenarios

RadioDiff-3D — 3D radio map construction with the UrbanRadio3D dataset. 📄 Paper | 💻 Code |

RadioDiff-FS — Few-shot learning for radio map construction with limited measurements. 📄 Paper | 💻 Code |

---

📶 Sparse Measurement & Localization

RadioDiff-Inverse — Sparse measurement-based radio map recovery for ISAC applications. 📄 Paper | 💻 Code |

RadioDiff-Loc — Sparse measurement-based NLoS localization using diffusion models. 📄 Paper |

---

📚 For a comprehensive categorized overview of radio map research, visit Awesome-Radio-Map-Categorized.

🎉🎉🎉 The paper has been accepted by IEEE JSAC!

An intelligent radio-map reconstruction system based on diffusion models. 📶✨

RadioDiff-k² is an advanced radio-map reconstruction project that leverages conditional diffusion models to generate high-quality radio-coverage maps from sparse measurements. The project serves 5G and 6G network planning, propagation prediction, and network optimization. 🚀📡

✨ Key Features

🎯 Multiple Simulation Methods

• DPM — deterministic propagation modeling with high speed and accuracy
• IRT4 — iterative ray tracing with high-precision prediction
• DPMCARK — vehicle-aware enhancement for urban mobility scenes 🚗📡

🏗️ Advanced Architecture

• Conditional diffusion model built on Swin Transformer
• VAE encoder for compact and efficient representation
• Multi-scale processing for flexible resolution support 🧠🧩

📊 Rich Conditioning Features

• Building layouts for realistic urban environments
• Transmitter positions to capture source attributes
• Vehicle data for dynamic occlusions
• k² features to encode physical propagation traits 🏙️📍🚘📐

🚀 Quick Start

Environment Requirements

Python >= 3.8
CUDA >= 11.0
PyTorch >= 1.12

📁 Project Structure

RadioDiff-k2/
├── 📋 configs/ # Configuration files
│ ├── BSDS_sample_*.yaml # Inference configs
│ └── BSDS_train_*.yaml # Training configs
├── 🧠 denoising_diffusion_pytorch/ # Diffusion core
├── 🔧 lib/ # Utilities
│ ├── loaders.py # Data loaders
│ └── modules.py # Network modules
├── 💾 model/ # Pretrained models
├── 📊 inference/ # Inference results
│ ├── DPMCARK/ # DPMCARK outputs
│ ├── DPMK/ # DPMK outputs
│ └── IRT4K/ # IRT4K outputs
├── 📈 metrics/ # Evaluation metrics
├── 🚀 train_cond_ldm.py # Training script
├── 🔮 sample_cond_ldm.py # Inference script
├── 🏗️ train_vae.py # VAE training
├── 🧮 caculate_k.py # k² feature computation
├── 🎯 demo.py # Usage examples
├── 📦 requirements.txt # Dependencies
└── 📖 README.md # Project docs

Method 2: conda

🎯 Usage Guide

1️⃣ Data Preparation

Dataset Layout

RadioMapSeer/
├── 📁 png/
│ ├── buildings_complete/ # Building images 256x256
│ ├── antennas/ # Transmitter positions 256x256
│ └── cars/ # Vehicle information optional
├── 📁 gain/
│ ├── DPM/ # DPM simulation results
│ ├── IRT4/ # IRT4 simulation results
│ └── IRT4_k2_neg_norm/ # k² feature maps
└── 📁 metadata/ # Meta files

Generate k² Features

# Run the k² feature computation script
python caculate_k.py

2️⃣ Model Training

Step 1 — Train the conditional diffusion model

# Train the main model
python train_cond_ldm.py --cfg configs/BSDS_train_DPMK.yaml
python train_cond_ldm.py --cfg configs/BSDS_train_DPMCARK.yaml
python train_cond_ldm.py --cfg configs/BSDS_train_IRT4K.yaml

3️⃣ Inference

Basic Inference

# DPMCARK inference
python sample_cond_ldm.py --cfg configs/BSDS_sample_DPMCARK.yaml
# DPMK inference
python sample_cond_ldm.py --cfg configs/BSDS_sample_DPMK.yaml
# IRT4K inference
python sample_cond_ldm.py --cfg configs/BSDS_sample_IRT4K.yaml