Wi-Fi Sensing & Privacy Weekly Brief
Ambient packets and coarse RSSI lower the sensing barrier
· Edition 1 · Historical public summary · original edition dated September 18, 2026; not newly source-checked
Download this Wi-Fi Sensing & Privacy brief (PDF) ↓This week's judgment
September 11–18, 2026 · Two items met the relevance, date and evidence checks. New research: one through-wall sensing preprint that became discoverable after the prior edition. Meaningful publication update: one June conference paper received an open arXiv manuscript; its underlying result is not new this week. Standards and regulatory updates: none verified for inclusion. The common signal is broader access to useful sensing inputs, but neither study establishes product-grade performance in occupied homes.
Through-Wall Detection using Software-Defined Radio based on adaptive Principal Component Analysis
What changed
The paper introduces a Welch-spectrum score that selects the five motion-relevant PCA components from CSI captured by a customized Bluebottle software-defined radio, then applies wavelet analysis and CFAR detection. It uses ambient Wi-Fi packets without transmitter control and became publicly discoverable after the prior briefing.
Why it matters
Adaptive component selection attacks a real deployment problem: the useful CSI dimensions change with noise and multipath. Passive use of existing traffic could reduce installation friction for security, rescue and eventual gateway-based presence sensing.
Evidence limits
Validation is a small, single-volunteer office study built around direct-path crossings and a custom SDR. Results are mostly representative plots and component scores, not aggregate sensitivity, specificity or home-level false-alarm rates; pets, still occupants and multiple people are untested.
Read in full?
Read in full. The method is concrete and the paper's narrow experiment makes the commercialization gap visible rather than hiding it behind a headline accuracy number.
Original source: arxiv.org ↗Cross-Domain Inference for Human Localization: Applying Wi-Fi RSSI Data to CSI-Trained Models
What changed
An open manuscript is now available for a June conference paper. It feeds normalized RSSI from three ESP32-C3 access points and a Raspberry Pi into a CSI-trained pose model, reporting roughly 80% average model confidence while the room is occupied without retraining the model.
Why it matters
RSSI is exposed on far more consumer and IoT platforms than CSI. Even a weak positive result widens the privacy threat model from specialist PHY access to ordinary signal-strength telemetry and suggests a cheaper path to coarse occupancy inference.
Evidence limits
Confidence is not localization accuracy. The evaluation uses one short, controlled room sequence, a tailored three-AP layout, empty-room calibration and hand-tuned preprocessing; empty-room inputs sometimes produced high confidence, and no multi-home or multi-person test is reported.
Read in full?
Skim. Read the setup and threats-to-validity sections for the privacy signal, but do not treat the 80% confidence figure as an accuracy claim.
Original source: arxiv.org ↗Read this edition in its original context
This is a preserved historical public summary. Use the stated evidence limits and original sources when assessing a finding.
Archived from the original Wi-Fi Sensing Review public summary, source commit eda94f23aff38b4e8e6aec7a4f77e22cc76ce128. Historical claims were not independently reverified in this formatting pass. No private desk analysis is reproduced. The source supplies an edition date, not an exact publication time; the archive timestamp is normalized to the start of that date in Eastern time.