ESPARGOS: Phase-Coherent WiFi CSI Datasets for Wireless Sensing Research

TL;DR

The paper addresses the need for phase-coherent, richly labeled WiFi CSI data for wireless sensing by introducing ESPARGOS, a low-cost, WiFi-based, phase-synchronous multi-antenna channel sounder. It details the hardware/software design, demonstrates how publicly accessible CSI datasets with ground-truth metadata can be collected, and showcases a data-driven application—Channel Charting—on ESPARGOS data. Key contributions include the ESPARGOS platform, publicly released CSI datasets (with detailed metadata and 3D environment information), and a Channel Charting demonstration that validates the utility of the datasets for indoor sensing tasks. The work highlights practical benefits such as real-time capability, backward compatibility, and data-driven research potential, while inviting community contributions and future developments.

Abstract

The use of WiFi signals to sense the physical environment is gaining popularity, with some common applications being motion detection and transmitter localization. Standard-compliant WiFi provides a cost effective, easy and backward-compatible approach to Joint Communication and Sensing and enables a seamless transfer of results from experiments to practical applications. However, most WiFi sensing research is conducted on channel state information (CSI) data from current-generation devices, which are usually not meant for sensing applications and thus lack sufficient spatial diversity or phase synchronization. With ESPARGOS, we previously developed a phase-coherent, real-time capable many-antenna WiFi channel sounder specifically for wireless sensing. We describe how we use ESPARGOS to capture large CSI datasets that we make publicly available. The datasets are extensively documented and labeled, for example with information from reference positioning systems, enabling data-driven and machine learning-based research.

Figures (5)

• Figure 1: ESPARGOS antenna array with $2 \times 4$ patch antennas.
• Figure 2: With a common clock and phase reference signal, multiple ESPARGOS devices can be combined into a large phase-synchronous antenna array (schematic diagram and cropped photo of combined $4 \times 8$ array)
• Figure 3: Visualization of $\mathbf { \bar{H} } \in \mathbb C^{B \times 2 \times 4 \times N_\mathrm{sub}}$ with $B = 4$ and $N_\mathrm{sub} = 117$: Channel coefficient amplitude and phase over subcarrier index $n = 0,\ldots,N_\mathrm{sub}-1$, each color representing one of the $B \times 2 \times 4$ antennas.
• Figure 4: The exemplary dataset espargos-0002: The figure shows (a) a photograph of the environment with the antenna array in the background, (b) a rendering of the 3D pointcloud and (c) a scatter plot (top view) of colorized "ground truth" positions of datapoints in $\mathcal{S}$, including antenna array and metal wall.
• Figure 5: Learned channel chart (a) before and (b) after optimal affine transformation, datapoint colorization is preserved from Fig. \ref{['fig:topview']}