WiFi-CSI (Channel State Information) turns any existing WiFi signal into a room sensor. No camera, no microphone, no wearable on the body. Here we explain the basics — and disclose the scientific sources our approach is built on.
Every WiFi packet effectively asks the receiver: "How did my signal change on its way to you?" That answer is called Channel State Information.
WiFi based on IEEE 802.11n/ac/ax transmits on many narrow carrier frequencies in parallel (OFDM subcarriers). The receiver measures amplitude and phase of the received signal for every individual subcarrier and compares it with the known transmit pattern. The result — the CSI matrix — describes the current radio channel in high detail.
When a person moves through the room, they change the reflections and multipath of the radio signal. These changes can be read out statistically from the CSI time series — as presence, motion, stillness, or sudden change of position (suspected fall).
Classic RSSI-based occupancy sensors only know a single signal-level value per packet. CSI provides dozens to hundreds of values per packet (one complex number per subcarrier × Tx/Rx antenna pair) and is therefore far more expressive.
These features have been reproduced in the literature for years and are practical on ESP32 hardware.
Shown scientifically, but under lab conditions — not yet reliable enough for everyday clinical use.
In theory yes, in practice not reliably yet. Here is the honest framing with the research we follow.
Water has a very high relative permittivity (≈ 80 at room temperature, far above fat or bone). Tissue water content therefore affects how radio waves propagate inside it — amplitude, phase and reflection. Since CSI measures exactly these parameters per WiFi subcarrier, there is a theoretical path: inferring water content from CSI patterns.
Related applications are already established: microwave-based stroke diagnosis (Persson et al.), UWB breast cancer screening, mmWave vital sensing. All use the same underlying effect: water dominates the electromagnetic response of biological tissue.
Lab studies have shown correlations between RF signals and hydration state — mostly with dedicated on-body sensors, not room-wide via WiFi-CSI.
| Authors, year | Contribution | Relevance to hydration |
|---|---|---|
| Gabriel, Lau, Gabriel — 1996 | "The dielectric properties of biological tissues" (Phys. Med. Biol.) | Foundational: measures permittivity and conductivity of tissues vs. water content. Basis for all later RF hydration work. |
| Topfer, Schoebel — 2015 | "Millimeter-Wave Tissue Diagnosis: The Most Promising Fields for Medical Applications" (IEEE MWM) | Review: which frequency bands express tissue water most strongly in the RF signal — important for sensor frequency choice. |
| Garrett, Fear — 2014 | "Stable and Flexible Microwave Imaging of Wrist for Hydration Monitoring" (IEEE Trans. Biomed. Eng.) | Direct microwave hydration study at the wrist: demonstrates feasibility of contact-near RF hydration sensing — we aim room-wide, not contact-near. |
| Persson, Fhager et al. — 2014 | "Microwave-Based Stroke Diagnosis Making Global Prehospital Thrombolytic Treatment Possible" (IEEE TBME) | Evidence that microwave measurement can distinguish tissue states in the head — related methodology, but dedicated sensor helmet rather than WiFi-CSI. |
| Yousefi et al. — 2017 already above | "A Survey on Behavior Recognition Using WiFi CSI" | Survey makes clear what CSI can do today and what not. Hydration is not listed as a reliable application. |
We track new publications, scrutinise methodology and only put hydration into the product when it is reproducible under real care conditions.
The ESP32-S3 in each room reads the subcarrier values directly from the radio hardware for every received WiFi packet.
Noise, phase jumps and amplitude offsets are filtered; only the relevant features (variances, spectra) remain.
Simple thresholds for presence/motion; ML models for fall suspicion — at the edge or on the local ward server.
Only suspicious patterns raise a hint in the dashboard. No continuous surveillance, no data overload.
For a long time CSI was only accessible on special hardware with Linux and Intel/Atheros cards. The ESP32 family has changed this: with the ESP-CSI toolkit, Espressif exposes CSI directly from the WiFi stack — on a cheap, dedicated module.
CSI uses only what is already in the room: WiFi radio waves. There is no image, no audio. Processing happens locally in the facility. Personal references are kept to the minimum needed.
More in our technology overview.
Our approach stands on the shoulders of an active research field. These works have shaped our architecture and our understanding of what is realistic.
| Authors, year | Title / contribution | What we use it for |
|---|---|---|
| Halperin, Hu, Sheth, Wetherall — 2011 | "Tool Release: Gathering 802.11n Traces with Channel State Information" (ACM SIGCOMM CCR) | Pioneering work on extracting CSI from standard WiFi hardware. Founds the entire research field. |
| Wang, Liu, Shahzad et al. — 2015 | "Understanding and Modeling of WiFi Signal Based Human Activity Recognition" (CARM) (MobiCom) | Modelling how human motion manifests in CSI patterns — the basis for motion and activity recognition. |
| Palipana, Rojas, Agrawal, Pesch — 2018 | "FallDeFi: Ubiquitous Fall Detection using Commodity Wi-Fi Devices" (IMWUT) | Fall detection with commodity WiFi hardware. Direct inspiration for our fall-hint module. |
| Yousefi, Narui, Dayal, Ermon, Valaee — 2017 | "A Survey on Behavior Recognition Using WiFi Channel State Information" (IEEE Communications Magazine) | Survey that frames possibilities and limits realistically. |
| Pu, Gupta, Gollakota, Patel — 2013 | "Whole-Home Gesture Recognition Using Wireless Signals" (WiSee) (MobiCom) | Early evidence that WiFi radio can resolve fine motions through walls. |
| Adib, Katabi — 2013 | "See Through Walls with Wi-Fi!" (Wi-Vi) (SIGCOMM) | Evidence that spatially resolved radio sensing with commodity WiFi is fundamentally possible. |
| Gringoli, Schulz, Link, Hollick — 2019 | "Free Your CSI: A Channel State Information Extraction Platform for Modern Wi-Fi Chipsets" (Nexmon CSI) (WiNTECH) | Makes CSI accessible on inexpensive embedded hardware (Raspberry Pi, BCM chips) — an important step toward ESP32-CSI. |
| Geng, Huang, De La Torre — 2023 | "DensePose From WiFi" (arXiv:2301.00250) | Pose estimation from WiFi. Research roadmap for us — not MVP. See above. |
DOIs and arXiv links point to the original publications. We deliberately link primary sources rather than secondary pages.
Espressif official
Official CSI toolkit for the ESP32 family — the basis for our radio bridge.
Steven M. Hernandez
Practical CSI capture tool and datasets for ESP32 — a valuable starting point for own pipelines.
SEEMOO Lab, TU Darmstadt
CSI platform for Broadcom/Cypress WiFi chips (e.g. Raspberry Pi) — reference implementation beyond ESP32.
Halperin et al., University of Washington
The original CSI tool for Intel 5300 cards — historical starting point of many follow-ups.
Xie et al., NTU Singapore
CSI extraction for Atheros chipsets — an important alternative for research setups.
Research roadmap
Example repository for pose estimation from WiFi reflections. For us roadmap material — not in today's MVP.
Original paper above (Geng et al., 2023).
A few central open-source building blocks — in case you want to develop, get started or contribute.
Espressif · official SDK
FreeRTOS-based SDK for ESP32-S3. The examples directory (`wifi`, `bluetooth/nimble`, `system/ota`, `storage/spiffs`) is mandatory reading.
Apache Mynewt · bundled in ESP-IDF
Lightweight BLE stack for ESP32. Central + notification-subscribe is our pattern for smartwatch integration.
1technophile · ESP32 BLE→MQTT gateway
Direct inspiration for our bridge architecture. BLE scanning, vendor decoder, MQTT publish — many smartwatch models out of the box.
theengs · BLE vendor-frame library
JSON-based decoders for hundreds of BLE sensors and low-cost smartwatches. Portable to ESP32 — inspiration for our plugin adapter layer.
open-source Android gateway
The largest open collection of smartwatch BLE protocols (Colmi, D-series, Mi-Band, Amazfit and many more). We cross-check frame layouts against its source.
Espressif · cloud reference
Patterns for provisioning, per-device keys, OTA and telemetry.
Architectural template for our es_cloud endpoint.
Full list with
concrete action items in the internal doc
docs/firmware-quellen.md in the source repository.
In half an hour we'll show you what our notification system can do — and what, on purpose, it doesn't. Honest, without marketing noise.
Book a pilot conversation