A Novel Channel Measurement Dataset for Cellular Networks

Problem formulation

To efficiently develop new cellular network algorithms, researchers often require high quality channel models. This stands especially true for data-driven approaches, such as Machine Learning (ML) algorithms. To provide researchers with real-life data from a variety of scenarios, we develop a novel channel recording and measurement testbed, as well as gather and release the dataset obtained via the testbed.

Testbed description

We design our testbed using a LIME software-defined radio, as well as modified srsLTE framework for RX processing pipeline. The scheme of the testbed demonstration is presented. The testbed performs measurements on the cells deployed by cellular network operators, for the best real-world approximation. The reconfigurability of SDR testbed, as well as wide frequency range of the SDR hardware allows us to capture data in different cell deployment bands.

Dataset description

We release the data for multiple scenarios of interest, such as

  1. Pedestrian Scenario, average speed of 3 km/h;
  2. In-Vehicle Scenario, average speed of 60km/h;
  3. High-Speed Train Scenario, average speed of 80km/h.

Each scenario measurements are presented in a matrix form (time-frequency grid, i.e. 2d matrix of SINR values), extracted from Cell Reference Signals (CRS) transmitted by cellular operator’s deployed cells. Each time-step  in the dataset is 1 millisecond, and each frequency step is 1 RB. The file type in which the data is saved is .txt (CSV format), SINRs are saved as floating-point values in decibels.

Dataset can be downloaded via the following link and will be fully available to the public after acceptance notification.

Dataset can be downloaded via the following link