OpenCSI demo accepted at IEEE LCN 2026

Karim Khamaisi’s demonstration OpenCSI: A Self-Calibrating, Cross-Hardware WiFi Sensing Mesh was accepted at the 51st IEEE Conference on Local Computer Networks (LCN 2026), taking place October 7 to 9, 2026 in Coimbra, Portugal. It brings the OpenCSI paper, accepted at the same conference, onto the demo floor as a live system attendees can walk into. The work builds on contributions from Simon Sigg’s Bachelor’s thesis at HSG.

What happens at the booth. Several ESP32 nodes are arranged around the sensing area, with a commodity router and a laptop running the OpenCSI backend and dashboard. Visitors move through four short stages: watching the per-link baselines form while the room is empty, stepping into the mesh and seeing it flip to present within about a second, the hardware swap, and a deliberate break where a node is unplugged or a chair moved so the dashboard flags the stale links, which re-mature in about 30 seconds.

The swap is the centerpiece. The mesh is physically rebuilt from ESP32-C6 to ESP32-S3 nodes, so the radios change chip generation, band, and PHY format (802.11ax HE20 to 802.11n HT20), and the same unchanged classifier keeps detecting occupancy. No retraining, no labelled data from the new hardware: the swapped-in nodes bring only their own quiet-period baseline, and the dashboard names both sides of the mismatch, model C6, mesh S3.

Why demonstrate a calibration layer. WiFi sensing demos usually show a downstream task, gesture or activity recognition, running on fixed hardware. This one shows the layer underneath, which is what decides whether any of those tasks survive a move to a different room or a different chip. Making the calibration observable while it runs is the point: the dashboard exposes the per-link evidence behind each decision, and a link whose baseline goes stale says so instead of failing quietly.

The source code and dataset are openly available under Apache 2.0.

Congratulations, Karim!