---
license: odbl
---
# OpenPathNet Dataset

This README describes the **OpenPathNet** dataset (the release referred to as **Link 1** in the [OpenPathNet project](https://github.com/liu-lz/OpenPathNet) documentation). The dataset is generated by the OpenPathNet toolchain from real-world **Miami** and **Boston** urban areas based on OpenStreetMap (OSM), and then simulated with NVIDIA Sionna ray tracing for RF multipath propagation / channel modeling research and AI tasks.

The dataset is also **carefully cleaned** to ensure **good building coverage** in every scene.

- Project code / generator repository: <https://github.com/liu-lz/OpenPathNet>
- City subsets: **Miami (806 scenes)** and **Boston (971 scenes)** (same directory layout across cities)

## Directory Structure

```
.
├── Miami/
│   ├── scenes/
│   │   └── scene_<lat>_<lon>/
│   │       ├── scene.xml
│   │       └── mesh/
│   │           ├── building_*.ply
│   │           └── ground.ply
│   ├── raytracing_results/
│   │   └── scene_<lat>_<lon>/
│   │       ├── raytracing_results.csv
│   │       ├── raytracing_results.pkl
│   │       ├── deepmimo_format.npy
│   │       ├── channel_gain_distribution.png
│   │       ├── delay_distribution.png # Here “delay” refers to ToA (Time of Arrival); the latest OpenPathNet fixes this filename
│   │       ├── path_type_distribution.png
│   │       ├── outdoor_receivers.png
│   │       └── heatmaps/
│   │           ├── azimuth_heatmap.png
│   │           ├── channel_gain_heatmap.png
│   │           ├── delay_heatmap.png  # Here “delay” refers to ToA (Time of Arrival); the latest OpenPathNet fixes this filename
│   │           └── elevation_heatmap.png
│   ├── generated_scenes.txt
│   └── raytracing.log
├── Boston/
│   └── ... (same structure as Miami)
└── README.md
```

### Naming

- Each scene directory is named `scene_<lat>_<lon>`, where `<lat>` / `<lon>` are decimal latitude/longitude.
- Scene folders under `scenes/` and `raytracing_results/` correspond one-to-one.

## Files

### generated_scenes.txt

A generation manifest and metadata (tabular text).

- The header records the total count, center coordinate, sampling radius, scene size, generation mode, etc.
- Each (tab-separated) row includes:
  - Scene file path (e.g., `data\scenes\scene_...\scene.xml`)
  - Original / actual latitude & longitude
  - Generation type (e.g., `OSM`)
  - Attempts
  - Offset distance (km)

### raytracing.log

A summary log for batched ray tracing (typically one line per scene), including runtime, number of receivers, and number of paths.

### scenes/

Geometry assets for each scene.

- `scene.xml`: scene description file (digital-twin / renderer-compatible format).
- `mesh/`: geometry meshes (e.g., buildings and ground) in `.ply`.

### raytracing_results/

Ray-tracing outputs and visualizations for each scene.

For each receiver point in each scene, this dataset keeps and records the **top 5 paths with the highest channel gain**. The files below contain the full multipath attributes for those retained paths, including receiver location, carrier frequency, path type, channel gain, **ToA (Time of Arrival)**, and departure/arrival angles.

- `raytracing_results.csv`: tabular results (easy to analyze/import).
- `raytracing_results.pkl`: Python-serialized results (fast loading).
- `deepmimo_format.npy`: DeepMIMO-style structured output for downstream ML pipelines.
- `heatmaps/` and `*.png`: visualizations (e.g., channel gain / delay(ToA) / azimuth / elevation).

#### Data schema: raytracing_results.csv / raytracing_results.pkl

- Structure: tabular data; typically **one row = (receiver `rx_id`, one path)**, so each `rx_id` usually appears 5 times.
- `raytracing_results.pkl` is a `pandas.DataFrame` with the same columns as `raytracing_results.csv`.

Columns:

- `rx_id`: receiver index (integer).
- `type`: path type (e.g., `LoS` / `Reflected` / `Scattered`).
- `channel_gain`: channel gain-related numeric value (scientific notation).
- `tau`: **ToA (Time of Arrival)** in seconds.
- `freq`: carrier frequency in Hz.
- `rx_coord`: receiver coordinates, formatted as a string like `"[x, y, z]"`.
- `phi_r`, `theta_r`: AoA azimuth / elevation angles.
- `phi_t`, `theta_t`: AoD azimuth / elevation angles.

#### Data schema: deepmimo_format.npy

- File content: a scalar `numpy.ndarray` with `dtype=object`; `arr.item()` yields a `dict`.
- Top-level keys:
  - `user`: a list of length $N_{rx}$; each element corresponds to one receiver.
  - `location`: a list used to describe scene / coordinate system information (may vary slightly across versions/configs).

Each `user[i]` is a `dict` containing:

- `location`: `numpy.ndarray` of shape `(3,)`, receiver coordinates `[x, y, z]`.
- `paths`: a `dict` containing (arrays are length 5, i.e., Top-5 paths):
  - `channel_gain`: `float32`, shape `(5,)`
  - `ToA`: `float32`, shape `(5,)`
  - `DoA_theta`: `float64`, shape `(5,)`
  - `DoA_phi`: `float64`, shape `(5,)`
  - `num_paths`: `int` (5 in this dataset)

## Reproducibility / Regeneration

This dataset is generated by the OpenPathNet toolchain. For generation scripts, ray-tracing entry points, and system requirements, please refer to:

- <https://github.com/liu-lz/OpenPathNet>

## Citation

If you use OpenPathNet in your research, please refer to the citation information in the [OpenPathNet repository documentation](https://github.com/liu-lz/OpenPathNet).

## License & Notes

- This directory contains a dataset slice/subset. For licensing, the generator code license, and third-party data source statements (OSM, etc.), please follow the [OpenPathNet repository documentation](https://github.com/liu-lz/OpenPathNet).