← Wi-Fi Sensing & Privacy weekly reports

AUDEAMUS INTELLIGENCE / Wi-Fi Sensing & Privacy

Wi-Fi Sensing & Privacy Weekly Brief

The inference moves closer to the radio.

Download this Wi-Fi Sensing & Privacy brief (PDF) ↓

This week's judgment

Two research reports put sensing inference closer to the radio and improve passive tracking methods. IEEE privacy work and proposed consumer IoT labeling rules also advance. None establishes broad household deployment or complete consumer control over derived presence and activity inferences.

01 / Preprint / FPGA hardware prototype ·

Activity classification runs inside the radio

What changed

A 132-neuron spiking classifier on an RFSoC reports 94.7% activity accuracy, 90.7% macro-F1 and 206-microsecond inference on IEEE 802.11ay micro-Doppler data. It outputs an activity label from programmable logic.

Why it matters

Local inference could reduce raw-signal transport and support an always-on sensing path. Ask what remains powered and what labels leave the device.

Evidence limits

Four activities and seven people do not establish household performance. Test-subject tuning makes accuracy optimistic by about 1.5 points. Dynamic-energy comparisons exclude the receiver and spectrogram front end; static FPGA power dominates.

Read in full?

Read the hardware measurements and explicit accuracy and energy omissions.

Original source: Keeping Inference in the Radio ↗

02 / Preprint / single-office experiment ·

Passive CSI tracking learns changing uncertainty

What changed

A physics-informed state-space model learns uncertainty in angle, delay and Doppler observations without position labels. Its WiDFS-based evaluation reports 0.67-meter mean absolute tracking error, 31.9% lower than raw WiDFS.

Why it matters

Label-free training could reduce calibration work if the method transfers to unfamiliar rooms and residents. That transfer needs its own test.

Evidence limits

The paper evaluates a walking target in one office on Intel 5300 hardware. Participant and independent-trajectory counts are not stated, and unfamiliar homes and quiet presence are untested.

Read in full?

Skim the uncertainty model and inspect the single-setting evaluation.

Original source: Physics-Informed Deep SSM for passive WiFi CSI tracking ↗

03 / Published IEEE standard / public status record ·

IEEE 802.11bi-2026 is an active privacy amendment

What changed

IEEE lists 802.11bi-2026 as an active standard, published October 23 after September 25 Board approval. It modifies the Wi-Fi MAC sublayer to enhance user privacy.

Why it matters

Ask vendors which privacy mechanisms they implement and how users control sensing-derived inferences.

Evidence limits

The public record is not the normative text, which requires purchase or subscription. Publication alone proves neither vendor adoption nor household sensing controls.

Read in full?

Have the product team review the exact specification before claiming compliance.

Original source: IEEE 802.11bi-2026 status and scope ↗

04 / FCC public notice / proposals, not final rules ·

Cyber Trust Mark operating proposals invite comment

What changed

The FCC's October 27 record identifies a request for comment on Cyber Trust Mark label design and surveillance. The public sensing summary reports proposals for renewal, post-market oversight and sensor-capability disclosures.

Why it matters

Track the consultation when planning consumer sensing products. Cybersecurity labeling and consent to derived RF inferences are separate questions.

Evidence limits

These are proposals, not adopted requirements. The full notice returned an access error in this review, so its detailed provisions remain reported rather than independently checked.

Read in full?

Read the original notice when accessible before relying on a proposed obligation.

Original source: FCC Public Notice DA 26-1162 ↗

ONE QUESTION TO TAKE TO A VENDOR

Ask what the product infers about people, where those inferences are computed, who can disable the sensing function and how derived records are retained or deleted. Treat a local activity label and a cybersecurity mark as evidence to examine, not as proof of household consent.

Read the original public Wi-Fi sensing summary ↗