📁 1. channel_data/ — Ray-Traced Channels (HDF5)

The ray-traced channel data is stored in HDF5 format using a four-dimensional structure:

Structure:
[number of scenes, number of TX–RX pairs, maximum number of rays, number of stored parameters]

Each ray contains:

1. Received power (dBm)
2. Time of arrival (seconds)
3. Elevation angle of departure (degrees)
4. Azimuth angle of departure (degrees)
5. Elevation angle of arrival (degrees)
6. Azimuth angle of arrival (degrees)
7. LOS flag: 1 = LOS, 0 = NLOS
8. Ray phase (degrees)

This data enables high-fidelity reconstruction of multipath behavior, reflections, diffractions, and critical LOS/NLOS transitions.

---

📁 2. lidar_data/ — LiDAR Point Clouds (NPZ)

LiDAR point-cloud data is provided in NPZ format as a 4D structure:

[max number of TX/RX pairs per episode, PCD(X, Y, Z)]

The point clouds use uniform quantized 3D coordinates and represent the surrounding street geometry, vehicles, and blocking structures. This modality supports research on sensing-aided beam prediction, blockage detection, and scene reconstruction.

---

📁 3. beam_data/ — Codebook Beam Indices (NPZ)

Beam labels are stored as NPZ files with a 3D structure:

[valid channels, Tx codebook index, Rx codebook index]

These indices correspond to the selected transmitter and receiver beams for each valid channel. They are directly compatible with beam selection, top-K accuracy evaluation, supervised training of beam predictors, and analysis of codebook-based mmWave systems.

---

📁 4. mimo_channel/ — MIMO Channel Matrices (NPZ)

MIMO channel matrices are stored in NPZ format as:

[number of channels, Nr, Nt, Nc]

Default configuration:

• Nr = 8 (receiver antennas)
• Nt = 8 (transmitter antennas)
• Nc = 1 (subcarrier)
• Spatial-domain channel aligned with a ULA model

This modality is used for beamforming, precoding, massive-MIMO learning models, and spatial correlation analysis.

---

📁 5. image_data/ — RGB Camera Frames (PNG)

The dataset includes RGB images in PNG format with resolution 1280×720.
Each scene contains images captured from three cameras mounted at the base station, each with a ~60° field of view.

These images provide contextual visual information about streets, buildings, traffic, and blockages, enabling tasks such as:

• Vision-aided beam prediction
• Scene classification
• LOS/NLOS visual inference
• Multimodal sensing and ISAC research

---

📁 6. gps_user_data/ — GPS and Metadata (CSV)

User metadata and GPS information are stored in CSV format with nine columns (below image with first 5 lines of the CSV file):

1. Val – Valid (V) or Invalid (I) channel
2. EpisodeID
3. SceneID
4. VehicleArrayID (receiver index inside the vehicle)
5. VehicleName (e.g., “flow1.2”)
6. X coordinate
7. Y coordinate
8. Z coordinate
9. LOS flag: 1 = LOS, 0 = NLOS

This folder aligns channel labels, vehicle trajectories, and LOS states with the visual, LiDAR, and radio modalities.

---

📁 7. scenario_metadata/ — Simulation & Scenario Configuration

This folder includes text or JSON files describing:

• 3D environment (Rosslyn urban canyon)
• Wireless InSite simulation parameters
• Materials used (ITU concrete for buildings/ground)
• Propagation model (X3D)
• Number of reflections, diffractions, and rays
• Diffuse scattering configuration
• Vehicle types used in SUMO (cars, buses, trucks)
• Receiver heights and mobility profiles

These files ensure reproducibility and consistent environment reconstruction.

---

📁 8. documentation/ — README and Citation Information

This folder contains:

• Dataset README
• License and usage rules
• Citation generator references
• Links to related academic papers
• Change logs or updates