Through-Wall Occupancy Sensing with WiFi 6

Motivation

Every WiFi access point is already a radar. As people move through a room, they reshape the radio channel, and the receiver measures that as channel state information (CSI) [1], [3]. No camera, no wearable, no app: the sensor is the router you already own. That makes WiFi one of the most promising privacy-preserving ways to sense presence and occupancy [4].

Most published sensing work still runs on a decade-old CSI tool built for 802.11n [1]. WiFi 6 (802.11ax) changes the game: many more subcarriers, wider bandwidth, and now the tooling to extract CSI from commercial hardware [2]. Almost nobody has measured what that richer signal buys you for the hardest case: sensing through a wall.

This thesis builds a real WiFi 6 sensing testbed and finds out.

Problem statement

Through-wall occupancy sensing is where WiFi sensing is most useful and least understood. Walls attenuate and scatter the signal, and results depend heavily on wall material, geometry, and hardware, yet most studies use one room, legacy NICs, and no material comparison. It is unclear how far the extra resolution of 802.11ax CSI [2] actually pushes through-wall occupancy counting, and whether it works in real time on cheap edge hardware.

That is the gap. This thesis builds a WiFi 6 testbed (Intel AX210 on a Raspberry Pi 5, with an ESP32 as a legacy baseline), and measures occupancy detection (0 to 3 people) across rooms and wall types, end to end from capture to a live prediction.

Research questions

How much does 802.11ax CSI improve through-wall occupancy detection over legacy (802.11n / ESP32) CSI [1], [2], on the same rooms and people?
How do wall material and thickness change what is recoverable, and can one calibration generalise across rooms?
Can the full pipeline (capture, calibration, detection) run in real time on a Raspberry Pi 5?

Suggested methodology

Stand up the testbed: AX210 on a Raspberry Pi 5 for 802.11ax CSI [2], an ESP32 node as the legacy baseline, in a repeatable room layout.
Collect a labelled dataset of 0 to 3 people across at least two wall types (e.g. drywall and concrete), with controlled ground truth.
Build a calibration and denoising step (our OpenCSI self-calibration layer is a natural starting point for heterogeneous nodes).
Train and compare occupancy models on 802.11ax vs legacy CSI, reporting accuracy per wall type and cross-room generalisation.
Deploy the best model on the Pi and measure real-time latency and stability.

What you need: curiosity for wireless systems and Python. Embedded experience helps but is not required, we bring the CSI tooling and the testbed.

Previous theses in this line

Feeling the Room: A Practical Study on WiFi-based Occupancy Detection, Simon Sigg, the group’s first WiFi occupancy testbed.
The paper OpenCSI: Self-Calibration Layer for Heterogeneous Mesh Wireless Sensor Networks (Khamaisi, Sigg, and Rodrigues, LCN 2026), which this topic builds on directly.

References

[1] D. Halperin, W. Hu, A. Sheth, and D. Wetherall. “Tool Release: Gathering 802.11n Traces with Channel State Information.” ACM SIGCOMM Computer Communication Review, 41(1):53, 2011.

[2] F. Gringoli, M. Cominelli, A. Blanco, and J. Widmer. “AX-CSI: Enabling CSI Extraction on Commercial 802.11ax Wi-Fi Platforms.” ACM Workshop on Wireless Network Testbeds, Experimental Evaluation & Characterization (WiNTECH), 2021.

[3] Y. Ma, G. Zhou, and S. Wang. “WiFi Sensing with Channel State Information: A Survey.” ACM Computing Surveys, 52(3):46, 2019.

[4] S. Depatla, A. Muralidharan, and Y. Mostofi. “Occupancy Estimation Using Only WiFi Power Measurements.” IEEE Journal on Selected Areas in Communications, 33(7):1381-1393, 2015.