1. Introduction

Wireless AI Research Dataset is a flexible and easy-to-use dataset with realistic environments designed for various wireless AI tasks. It supports sensing tasks such as localization and environment reconstruction, MIMO tasks such as reflection system and beam-forming, and PHY tasks such as CSI feedback and channel estimation…

• Realistic:10000 scenarios are randomly selected from real maps of more than 40 major cities around the world. The building layout is more suitable for real scenarios.
• Flexible: Provides rays generated by ray-tracing, allows users to customize communication parameters, and generates data required by wireless AI tasks.
• Easy-to-use: Data generation code, data preprocessing code, and sample task training code are provided for users to quickly use.

2. Data organization

In terms of different scenarios and device densities, the dataset consists of two parts：

• Part 1: Sparse Deployment (10000 maps, 5 BS locations and 30 UE locations in each map)
• Part 2: Dense Deployment (100 maps, 1 BS location and 10000 UE locations in each map)

Fig 1 Buildings and BS/UE positions in top view	Fig 2 Buildings in bird-eye view

For detailed dataset information, please refer to the description file "Wireless AI Research Dataset README. docx".