5G NR Positioning (POS)

CSI dataset for high-precision positioning with 5G NR signals.

Official links

• Raw data: https://ieee-dataport.org/open-access/csi-dataset-towards-5g-nr-high-precision-positioning
• DOI: https://dx.doi.org/10.21227/jsat-pb50

Tasks supported

• Position estimation (Mean Localization Error in meters)

Targets

• Continuous (x, y, z) coordinates (regression); no class labels.

Split

• 80/20 random split

Preprocessing

1. Point the script at a scenario directory that contains the three raw .mat files (different SNRs).
2. Each .mat file is HDF5 and includes features and labels/position.
3. Cast features to float32 and resize to 224x224.
4. Min-max normalize features using scene-specific bounds (indoor or outdoor).
5. Standardize with scene-specific mean/std.
6. Normalize positions to [-1, 1] using scene coordinate bounds.
7. Write normalized features and positions to HDF5.

Script: preprocess_nr_positioning.py

python preprocessing/preprocess_nr_positioning.py \
  --data-path <POS_DIR> \
  --output data/pos.h5 \
  --scene outdoor

Metric

• Mean Localization Error (meters) — lower is better.

Citation

@data{jsat-pb50-21,
  doi = {10.21227/jsat-pb50},
  url = {https://dx.doi.org/10.21227/jsat-pb50},
  author = {Kaixuan Gao and Huiqiang Wang and Hongwu Lv},
  publisher = {IEEE Dataport},
  title = {CSI Dataset towards 5G NR High-Precision Positioning},
  year = {2021}
}