Definitions for terms used across the Wavey documentation. For how they connect, start with how it works , the sensing pipeline , and the detection ladder .
Channel State Information (CSI)
A per-subcarrier measurement of how a WiFi signal traveled from transmitter to receiver — amplitude and phase at each OFDM frequency slice. The core sensing signal. See how it works .
RSSI (Received Signal Strength Indicator)
A single scalar: total received power. Too coarse for micro-motion or activity recognition. CSI preserves the per-subcarrier detail RSSI discards.
OFDM subcarrier
One narrow frequency component of an OFDM WiFi channel. CSI is measured independently at each subcarrier, giving dozens of parallel channel measurements per packet.
Multipath
Radio signals arrive via multiple paths (direct, reflected, scattered). The received channel is a superposition of these paths. Bodies add and modify paths, changing CSI.
Micro-Doppler
Small periodic motion (breathing, heartbeat) that modulates CSI phase at low frequencies (0.1–0.5 Hz for respiration). Far weaker than macro-motion signatures.
Phase sanitization
Preprocessing that removes hardware-induced phase artifacts (CFO, SFO, per-antenna offsets) before inference. Mandatory for micro-motion detection. See the sensing pipeline .
Spectrogram
A time-frequency representation — typically STFT of CSI amplitude or phase over a sliding window. The standard input representation for deep learning activity classifiers.
Baseline fingerprint
A statistical model of CSI in an empty room — mean amplitude profile, variance envelope, phase reference. Live CSI compared against the baseline powers occupancy detection.
Domain shift
When a model or threshold trained in one environment fails in another because multipath fingerprints differ. Cross-site generalization is a core engineering challenge. See baselines and generalization .
HAR (Human Activity Recognition)
Classifying what a person is doing (walk, sit, stand, gesture) from CSI patterns, typically via spectrogram classifiers. Harder than binary occupancy; needs labeled data per environment.
Beamforming feedback (CBR)
Compressed beamforming reports from standard 802.11ac/ax APs — channel quality feedback used for beam steering. BeamSense reconstructs multipath from CBR when raw CSI is unavailable. Nearly ubiquitous on deployed chipsets; raw CSI is not.
Signal tendency index (STI)
Shape similarity between adjacent CSI amplitude curves. Used in FreeDetector and WiFree for occupancy detection with greedy subcarrier selection. WiFree reported 99.1% occupancy accuracy on building-scale deployments.
Roaming model
An offline model (CrossSense) that synthesizes target-environment CSI training data from source-site calibration samples, enabling cross-site activity recognition without full relabeling.
FallNet
An autoencoder (SiFall) trained on normal daily activities. Falls detected as high reconstruction error — anomaly detection, not fall-template classification.
C3SL
Convex shapelet learning on 3-way CSI tensors for gait-based identification. 91% among 20 enrolled people in controlled conditions. Distinct from anonymous occupancy sensing.
Device-free sensing
The person being detected carries no device. Wavey reads the body’s effect on WiFi directly.
Passive sensing
Using signals already present in the environment rather than emitting a dedicated probe signal.
ESP32-CSI
CSI capture on low-cost ESP32 WiFi microcontrollers via Espressif’s ESP-CSI library. See commodity CSI limits .
IEEE 802.11bf
The WiFi standard amendment for native environment sensing — formalizing sensing alongside data delivery.
Occupancy vs presence
Occupancy — binary: is anyone here? Most robust. Presence — confirms a still person via micro-motion (breathing). See the detection ladder .