Documentation
A comprehensive guide to the LAMBDA dataset: Data Pipeline, System Configuration, and Usage.
🛠️ Data Collection Pipeline
The LAMBDA dataset utilizes a sophisticated Sim-to-Real pipeline. We leverage Unreal Engine 5 for photorealistic visuals and AirSim for sensor extraction, while combining Sionna's ray tracing with CADFEKO's electromagnetic simulations.
⚙️ System Configuration
We simulate a multi-sensor UAV platform equipped with visual, laser, and radio-frequency sensors.
- Camera
- LiDAR
- Wireless
- Radar
RGB & Depth Cameras are physically aligned to provide pixel-level correspondence via AirSim.
| Parameter | Value | Description |
|---|---|---|
| Category | RGB Camera | Standard pinhole camera model |
| Resolution | 1920 × 1080 | Full HD |
| FOV | 110° | Wide-angle field of view |
| Frame Rate | 100 Hz | High-speed capture |
| Position | X:-8.1, Y:-157, Z:-35.7 | Relative to Global Frame |
| Orientation | Pitch:40, Yaw:-180 | Downward-looking tilt |
A mechanical spinning LiDAR simulation for dense 3D mapping.
| Parameter | Value | Description |
|---|---|---|
| Channels | 128 lines | High vertical resolution |
| Range | 150 m | Max sensing distance |
| Horizontal FOV | 120° | [-60°, +60°] |
| Horizontal Res. | 1024 points | Approx 0.12° angular resolution |
| Vertical FOV | 90° | [0°, 90°] |
| Scan Rate | 30 Hz | Frame rate |
Wireless Channel State Information (4.9 GHz) based on Sionna Ray Tracing. The data preserves Sparse Multipath Components.
1. Channel Physics (Per Path)
- Complex Gain:
a_real,a_imag(Includes weather attenuation) - Amplitude:
- Phase:
- Delay:
tau(Propagation delay in seconds) - Doppler:
doppler(Doppler shift in Hz, derived from UAV velocity)
2. Geometric Angles
- AoD (Tx):
theta_t(Zenith),phi_t(Azimuth) - AoA (Rx):
theta_r(Zenith),phi_r(Azimuth)
3. Interaction History
- Interactions: List of interaction types (LoS, Reflection, Diffraction, Scattering).
- Valid: Boolean flag for path validation.
4. Receiver State & Metadata
- State:
t(Timestamp),uav_pos[x,y,z],uav_vel[vx,vy,vz]. - Config:
carrier_frequency(4.9GHz),weather_kind(rain/fog/clear).
FMCW Radar (77 GHz) simulation combining multipath propagation and RCS modeling.
| Parameter | Value | Note |
|---|---|---|
| Frequency | 77 GHz | Automotive / UAV radar band |
| Bandwidth | 2 GHz | Provides high range resolution |
| Sample Rate | 70 Msps | Fixed sampling rate |
| Chirp Duration | 40 us | Typical setting |
| Num Chirps | 64 | Determines velocity resolution |
| Tx Power | 12 dBm | Transmitter power |
| Tx/Rx Gain | 25 dB | High-gain antenna array |
| Noise Floor | -100 dBm | Thermal noise level |
📂 File Structure
The dataset follows a structured hierarchy organized by Scenario, Weather, and Trajectory.
Dataset_Root/
├── San Francisco (Urban)/ # Scenario Name
│ ├── Scene 1/ # Sub-scene ID
│ │ ├── sunny/ # Weather Condition
│ │ │ ├── 1_uav_z_trace_1/ # [Num_UAVs]_[Trace_Type]_[Trace_ID]
│ │ │ │ ├── multipath/
│ │ │ │ │ └── f4p9GHz_V/ # Carrier Freq & Polarization
│ │ │ │ │ └── ... (.npz files)
│ │ │ │ ├── lidar/ # .pcd files
│ │ │ │ ├── rgb/ # .png files
│ │ │ │ ├── depth/ # .npz files
│ │ │ │ ├── imu/ # .json files
│ │ │ │ └── poses/ # .json/.npz files
│ │ │ ├── 1_uav_z_trace_2/
│ │ │ └── ...
│ │ └── rainy/
│ └── Scene 2/
├── San Francisco Style City (Suburban)/
└── SJTU IEEE/
💻 Usage & Coordinates
The LAMBDA dataset strictly follows Sionna's Right-Handed Coordinate System to align with mathematical intuition in wireless communications.
- X: Forward
- Y: Left
- Z: Up
Loading CSI Data (Python)
import numpy as np
# Load the compressed CSI file
data = np.load("path/to/csi_sample.npz")
# 1. Access Multipath Components
a_real = data['a_real']
a_imag = data['a_imag']
delays = data['tau']
doppler = data['doppler']
# 2. Reconstruct Complex Gain
complex_gain = a_real + 1j * a_imag
# 3. Access Angles (AoD / AoA)
theta_t, phi_t = data['theta_t'], data['phi_t']
theta_r, phi_r = data['theta_r'], data['phi_r']
print(f"Detected {len(delays)} paths.")
print(f"Max Doppler Shift: {np.max(np.abs(doppler)):.2f} Hz")