Fraunhofer Institute for Integrated Circuits IIS
Fraunhofer Institute for Integrated Circuits IIS

RAIL-5G

Environment

  • Location: SNCF Technicentre in Bischheim, France
  • Area: approx. 38 m × 22 m, with 811 m²
  • Two distinct propagation regions:
    • An open area with line-of-sight (LOS) to several radio units (turquoise).
    • Two narrow corridors between large metallic train compartments and maintenance infrastructure with strong reflections, frequent NLOS, and dense multipath propagation (light green).

Setup

  • 5G Cell: Amarisoft 5G NR, band n77, 3.95 GHz carrier frequency
  • Reference signal: SRS with 100 MHz bandwidth, 100 Hz transmission rate
  • Radio units (RU): 8 × Benetel RAN550 n77u Radio Units, spatially distributed
  • Sampling Rate: 122.88 MHz
  • User Equiment (UE): Quectel RM502Q-AE 5G modem, mounted on mobile robot (0.42 m height)
  • Robot Speed: approx. 1m/s

Reference System

  • System: 7D Kinematic Metrology iGPS (infrared laser triangulation)
  • Number of laser transmitters: 8
  • Accuracy: sub-millimeter
  • Sampling rate: 40 Hz

 

Dataset Format

The datasets are stored in HDF5 files that contain a single dataset named ds with 52,588 entries. Each entry represents one snapshot of CIR measurements and contains the following fields:

Field Shape Type Description
antenna_mask (8) bool Indicates whether the CIR data for each base station is valid (True) or invalid (False).
CIR (2, 8, 96) float32 Complex CIR measurements. The first dimension contains the real and imaginary parts, the second dimension represents the 8 base stations, and the third dimension contains 96 delay taps.
interpolation_diff (2) float32 Time offsets in seconds to neighboring reference points: index 0 is the time since the previous reference point, and index 1 is the time until the next reference point.
reference (2) float32 Ground-truth 2D reference position in meter.
timestamp () float64 Timestamp of the snapshot in seconds.
toa (8) float32 Time-of-arrival measurements in seconds for each base station, relative to the earliest arrival.

Recordings

Take Description Number of Samples  
take01 Meander, lower corridor 52,588 Download
Take Description Number of Samples  
take02 Meander, free space 51,552 Download
Take Description Number of Samples  
take03 Meander, upper corridor 54,963 Download
Take Description Number of Samples  
take04 Serpentine, lower corridor 43,130 Download
Take Description Number of Samples  
take05 Serpentine, open space 25,809 Download
Take Description Number of Samples  
take06 Serpentine, upper corridor 31,682 Download
Take Description Number of Samples  
take07 Random, full area 57,585 Download
Take Description Number of Samples  
take08 Coarse meander, full area 68,045 Download

If you use our dataset for your research, please consider citing:

@inproceedings{pirkl2026rail5g,
  author       = {Pirkl, Jonas and Eidloth, Andreas and Kasparek, Maximilian and Feigl, Tobias and Mutschler, Christopher},
  title        = {RAIL-5G: A Challenging Real-World Dataset and Benchmark for AI-Based 5G Indoor Localization},
  booktitle    = {2026 16th International conference on Indoor Positioning and Indoor Navigation (IPIN)},
  year         = {2026},
}