Elder-Care Monitoring with WiFi Sensing

Aging in place needs continuous awareness without cameras in bedrooms and without wearables that get left on the nightstand. WiFi CSI offers ambient monitoring — presence, movement patterns, and fall-like events — from nodes already in the home.

Why fall is not a classification problem

Most early RF fall detectors train a classifier: fall vs walk vs sit. That assumes falls produce reproducible CSI signatures you can label and match. In practice, falls are accidents — impact varies with surface, body angle, and what they hit. Labeling enough real elderly falls to train a classifier is ethically and practically hard; most training data is simulated.

SiFall (SenSys 2022) reframes the problem:

  1. Learn normal — an autoencoder (FallNet) models the distribution of repeatable daily activities (walking, sitting, standing) for this person in this room.
  2. Front-end signal chain — CSI amplitude (not phase) → conjugate multiplication denoising → channel dynamics via cosine similarity across subcarrier vectors → acceleration threshold (Θ = 2.5 m/s²) to segment candidate events → STFT → FallNet input.
  3. Detect abnormal — a fall produces a CSI segment with high reconstruction error because it does not match any learned normal pattern.
  4. Adapt online — self-supervised incremental learning updates the normal-activity model as routines evolve, without requiring fall labels.

Reported 98.3% in real-world tests, 94.1% over 3-day continuous monitoring with one false alarm. This is fundamentally different from spike-plus-immobility heuristics. It handles variable fall kinematics because anything outside the routine distribution triggers — not because it matched a fall template.

Limitations from the published system: largely single-room operation, simulated fall data, and unvalidated on subjects over 60. False positives from unusual but non-fall activities (dropping onto a couch) remain a tuning problem.

Inactivity monitoring

Less dramatic but equally useful: prolonged absence of expected motion. If someone who normally moves through the kitchen every morning has not triggered motion by noon, that is a signal worth surfacing — built on presence detection and motion baselines, not fall classification.

Routine patterns emerge from temporal aggregation over days: typical wake time, movement frequency, room transitions. Deviations from routine are the alert, not a single event classifier.

What Wavey is not

Wavey is an awareness layer, not a medical device:

  • No certified fall detection with regulatory clearance.
  • No clinical-grade respiration rate or heart rate.
  • No identity — the system does not know who fell, only that an anomaly occurred in a zone.

For the full feasibility picture, see fall detection at rung 6 on the detection ladder .

Deployment considerations

  • Node placement — links should cross high-traffic areas (hallway, bedroom doorway) where falls are most likely to perturb CSI.
  • Baseline stability — furniture changes and new occupants require re-baselining.
  • Multi-person homes — superposition makes per-person fall attribution unreliable without spatial diversity.

Further reading