Skip to content

U6G XL-MIMO Radiomap Prediction: Multi-config Dataset & Beam Map Approach

A benchmark project for multi-configuration radiomap prediction in U6G / XL-MIMO systems, featuring 800 scenes, multi-frequency, multi-antenna, and multi-beam settings.

Public release: The complete dataset, pretrained models, and code are now publicly available.

  • Dataset & Pretrained Models (Hugging Face): https://huggingface.co/datasets/lxj321/Multi-config-Radiomap-Dataset
  • Code Repository (GitHub): https://github.com/Lxj321/MulticonfigRadiomapDataset
  • Project Website: https://lxj321.github.io/MulticonfigRadiomapDataset/

Dataset Quickstart Benchmark Pretrained

Overview

This project is designed for studying:

  • multi-configuration radiomap prediction
  • cross-configuration generalization
  • cross-environment generalization
  • beam-aware radiomap modeling
  • sparse radiomap reconstruction

A key feature of this project is the joint design of:

  • height maps
  • configuration-only beam maps
  • ray-tracing radiomap labels
  • optional mesh assets for ray-tracing reproduction
  • UNet / GAN baseline pipelines

The current website is intended to:

  • preview the dataset structure
  • preview benchmark task definitions
  • preview pretrained model organization
  • help finalize documentation before public release

What Is Included

The current public release includes:

  • Dataset (Dataset/)
  • height maps
  • radiomaps
  • configuration-only beam maps
  • optional mesh assets
  • Baselines
  • UNet training / evaluation scripts
  • GAN training / evaluation scripts
  • Pretrained models (Pretrained_Model/)
  • GAN checkpoints for benchmark tasks
  • UNet checkpoints and related evaluation resources
  • Dataset generation pipeline
  • OSM → Sionna meshes → height maps → ray-tracing radiomaps → beam maps

Quick Facts

  • Scenes: 800 (u1..u800)
  • Frequencies: 1.8 / 2.6 / 3.5 / 4.9 / 6.7 GHz
  • TX antennas: up to 1024 TR
  • Beam counts: 1 / 8 / 16 / 64
  • Beam pattern: 3GPP TR 38.901

Recommended Entry Points

If you are browsing this preview site, start with:

  • Dataset — dataset structure and naming rules
  • Benchmark — task definitions and benchmark settings
  • Pretrained — planned pretrained model organization
  • Quickstart — preview of the intended evaluation workflow

Please note that some pages currently describe the planned public structure and may be refined before release.

  • Project Website: https://lxj321.github.io/MulticonfigRadiomapDataset/
  • Code Repository: https://github.com/Lxj321/MulticonfigRadiomapDataset
  • Dataset & Pretrained Models: https://huggingface.co/datasets/lxj321/Multi-config-Radiomap-Dataset
  • Paper/Preprint: coming soon

Future Updates

Future updates may include:

  • paper / citation metadata
  • refined tensor shape and unit documentation
  • additional benchmark examples
  • more detailed checkpoint-to-task mapping

Citation

Citation information will be added after the paper metadata is finalized.

@article{to_be_added,
  title   = {U6G XL-MIMO Radiomap Prediction: Multi-config Dataset and Beam Map Approach},
  author  = {Xiaojie Li and collaborators},
  journal = {to be added},
  year    = {2026}
}

License

  • Code: MIT License
  • Dataset: CC-BY-4.0 License

Contributor

Xiaojie Li (李宵杰) Email: xiaojieli@seu.edu.cn/xiaojieli@nuaa.edu.cn