Introduction

Modern Wi-Fi access points constantly probe the radio channel they operate on, and the resulting Channel State Information (CSI) is sensitive to anyone moving through the room: bodies attenuate, scatter and Doppler-shift the propagating signal in ways that a learning algorithm can exploit to recognise gait, count people or estimate where they are. The ratification of the IEEE 802.11bf amendment in 2024 has turned this idea from a research curiosity into a standardised capability, and commercial vendors (Origin AI, Cognitive Systems) are already shipping CSI-based motion detection inside consumer routers, but always limited to presence and coarse motion. Per-person identification and group analysis remain at the research stage.

The aim of this project is to explore different approaches described on the SoTA on the topic of Wi-fi sensing and help collect a small dataset with Ettus USRP equipment that would allow us to evaluate these models and algorithms.

Project Goals

The exact scope is intentionally flexible and will be agreed at the interview based on the candidate’s background and the duration of the thesis/project. Ideally we would like to:

  • Do a small experimental campaign in one of the lab’s measurement rooms, with 1 – 5 volunteer subjects, single-person and small-group scenarios, and at least two recording days.
  • Develop a clean data schema to store CSI captures.
  • Evaluation of person-identification and group-counting models on this dataset.
  • A comparison of the SDR-based capture against a commodity Wi-Fi NIC (PicoScenes on an Intel AX210, or Nexmon CSI on a Broadcom-based device) capturing the same emitter at the same time.

If time permits, additional directions could be:

  • Implement additional model architectures (Doppler-spectrum descriptors, attension-based fusion)
  • Building a real-time demonstrator

Requirements

The student should be proficient or willing to learn:

  • Python and a modern deep-learning framework (PyTorch preferred)
  • Wireless experimentation with software-defined radios (Ettus USRP, GNU Radio or OpenCPI)
  • Basic OFDM / Wi-Fi PHY concepts (CSI, CFO, channel estimation)
  • Linux, RF and ML: CLI tools, common data storage formats (shell, git, HDF5 / SigMF)