Data and code for "Same Signal, Different Story: Demystifying Receiver Effects in Wi-Fi Channel State Information"

The dataset(s) here were collected with a Wi-Fi testbed of eight different receivers:

2x Asus RT-AC86U2x ESP32-S31x Intel IWL53001x Intel AX2101x Qualcomm Atheros QCA AR94621x USRP N2954-R

The transmitter for all the datasets is another USRP N2954-R with Wi-Fi frames generated using the MATLAB WLan Toolbox.

All data was collected using our python scripts, published on GitHub: https://github.com/nzqo/sensession.

Relevant subprojects for CSI extraction are found in csi_tools in sensession.

You can find the experiment design (e.g. frame precoding) by inspecting the /experiments directory.

Evaluation scripts of the data are also shared in the /evaluation directory.

Format

All data is stored as .parquet files, which is tabular data. Each line contains a single CSI recording. The meta_id column marks a group (a coherent recording), with further information linked in the corresponding meta.parquet files. These files can be easily loaded into dataframes using e.g. pandas or polars.

Every CSI value is represented as a nested list corresponding to shape [num_antennas, num_streams, num_subcarrier].

HAR dataset

The HAR dataset comes with a dedicated README with more information on the contents.

Over-the-wire data

For the OTW data, all devices are connected to the transmitter using cables of equal length, feeding the signal through a waveguide powersplitter (ZN8PD1-63W-S+). Archive names describe the measurements we ran (e.g. single subcarrier modifications).