Research
The environment as a sensor: what we learned studying WiFi sensing
Can a space be read through the reflections of ordinary WiFi — no cameras, no wearables? Channel physics, the open science, and the conclusions that became the Qvijin Aura product by March 2026.
In January 2026 we set ourselves a research question: is ordinary WiFi infrastructure enough to understand what is happening in a room — without installing a single camera or putting a single device on a person.
The physics: a channel as a measuring instrument
Every WiFi packet carries more than data. The receiver estimates the channel state (CSI) — a complex-valued matrix describing how the environment distorted the signal:
H(n, m, k) = A(n, m, k) · e^{ j·Φ(n, m, k) }
where A is amplitude and Φ is phase for each antenna pair and subcarrier. The signal reaches the receiver along many paths — direct and reflected:
h(t) = Σ aₗ · e^{ j·φₗ } · δ(t − τₗ)
A human body is roughly 70% water with high dielectric permittivity, so presence and movement measurably change the path set — aₗ, φₗ, τₗ shift. That is the signature of presence. Motion adds a Doppler shift:
f_d = 2·v·cos θ / λ
and breathing adds a periodic phase modulation ≈ 4π·d(t)/λ: at a 12.5 cm wavelength, a chest displacement of a few millimeters produces a measurable phase shift. Spatial resolution is bounded by bandwidth, Δr = c / (2·BW) — which is exactly why the industry's move to wide WiFi 6/7 channels changes what this class of problems can do.
The open science: from radars to routers
A decade of published work tells the story. WiTrack (MIT) needed a purpose-built radar. SpotFi (Stanford) localized a person on a commodity network card. Person-in-WiFi (CMU, CVPR 2019) and DensePose from WiFi (CMU, 2022) demonstrated human pose estimation on tens-of-dollars hardware. In September 2025, IEEE ratified 802.11bf — WLAN Sensing became a legitimate function of the network itself, and next-generation chipsets are announced with native sensing and on-board edge AI.
Privacy: the argument and the counter-argument
Radio-channel sensing is private by construction: no images ever exist. But published work in 2025 also showed the other side — channel telemetry can identify a person with high accuracy. The engineering conclusion is singular: processing must live at the network edge, aggregates must replace raw data, and privacy must be in the architecture, not in the privacy policy.
What came of it
The study convinced us the direction had matured: the physics is measurable, the science is open, the standard is ratified. In March 2026 the direction was presented as a product — Qvijin Aura.
Aura's methods, datasets, and engineering decisions remain outside this publication — this is an article about why the direction works, not a recipe for reproducing our stack.