Documentation
Welcome to the technical documentation for AeroVerse 6G. This dataset is generated using a high-fidelity digital twin pipeline integrating Unreal Engine 5, Sionna, and CADFEKO.
1. Sensor Configuration
Our data collection platform utilizes a multi-sensor setup designed for low-altitude perception.
| Sensor | Specification | Description |
|---|---|---|
| LiDAR | 128 Channels | 30Hz, 1024 horizontal resolution. Range: ~100m. |
| Radar | 77 GHz FMCW | 2GHz Bandwidth, 100Msps, 128 Chirps/Frame. Supports Doppler. |
| Camera | RGB & Depth | synchronized at 100Hz. Depth maps compressed in .npz. |
| CSI | 4.9 GHz | Generated via Sionna RT. Includes full multipath parameters (AoA, AoD, Delay). |
2. Coordinate Systems (Critical)
AeroVerse 6G uses aligned coordinate systems between visual and wireless domains. Please note the conversion when using raw data:
- Visual Domain (UE5): Left-handed, Z-axis up, Unit: cm.
- Wireless Domain (Sionna): Right-handed, Y-axis up, Unit: m.
Transformation Rule: To convert from UE5 to Sionna, divide coordinates by 100 and swap Y/Z axes.
3. Directory Structure
The dataset is organized by Scenario -> Frequency -> Weather.
Dataset_Root/
├── San Francisco (Urban) /
│ ├── Scene 1/ # Sub-scene
│ │ └── sunny/ # Weather conditions
│ │ │ ├── 1_uav_z_trace_1 # Agents number and trajectory
│ │ │ │ ├── multipath/
│ │ │ │ │ └──f4p9GHz_V
│ │ │ │ ├── lidar/
│ │ │ │ ├── rgb/
│ │ │ │ ├── depth/
│ │ │ │ ├── imu/
│ │ │ │ ├── poses/
│ │ │ │ └── ...
│ │ │ ├── 1_uav_z_trace_2
│ │ └── ...
│ └── ...
├── San Francisco Style City (Suburban) /
├── SJTU IEEE /
└── ...
4. Simulation Methodology
Wireless Channel Modeling
We utilize NVIDIA Sionna for ray tracing. To ensure physical realism across 1-100 GHz:
- Materials: Mapped using ITU-R P.2040 standards.
- Weather: Integrated ITU-R P.838 (Rain) and P.840 (Cloud/Fog) attenuation models.
- Atmospheric: Implemented ITU-R P.676 to account for oxygen and water vapor absorption peaks (e.g., at 60GHz and 22GHz).
Radar RCS Generation
Unlike simple point-scatterer models, our radar data is generated by:
- Exporting the drone mesh from AirSim.
- Calculating high-fidelity RCS (Radar Cross Section) using CADFEKO (RL-GO solver).
- Combining RCS with Sionna multipath data to synthesize accurate 4D radar signatures.