Motion Detection Over WiFi.

Motion detection with no motion sensor. A WiFi signal arriving at a receiver is the sum of many paths — direct, off the walls, off the furniture, off people — and the radio already measures how that multipath distorts each OFDM subcarrier, because it must equalize them to decode anything. That measurement is Channel State Information, and the ESP32 exposes it. A person walking through the room changes the paths, the per-subcarrier amplitudes ripple, and software can see it.

The firmware side is small by design: the ESP32 joins my network, pings the router 20 times a second to guarantee steady traffic, captures CSI from every reply (filtered to the router’s MAC), and streams each frame over USB serial as a CSV line. The interesting constraints are operational — WiFi power save has to be disabled because it gaps the capture, and the CSI callback runs inside the WiFi task, so frames cross to the output loop through a FreeRTOS queue rather than doing serial I/O in the callback.

The host side does the actual sensing: a Python viewer plots a live waterfall of subcarrier amplitudes next to a motion metric — the mean per-subcarrier standard deviation over a two-second sliding window — against an adaptive threshold that learns the room’s quiet level slowly and falls back quickly, with hysteresis so the presence flag doesn’t flicker. In synthetic-data tests the detector flags motion within about half a second of onset and recovers even when it starts up mid-motion.

It works in simulation, but it is not as accurate as it should be in reality. Fine-tuning the parameters is the current work.