WiFi Signals Can Reveal Your Heartbeat — A $9 ESP32 Becomes a Through-Wall Radar
> Source: RuView, https://github.com/ruvnet/RuView
1. Background: Radio Waves Are Silent Witnesses
Your home Wi-Fi router transmits thousands of signals every minute. These signals reflect off walls, furniture, and people — being reflected, absorbed, and distorted along the way. We usually think of this as just a "network connection," but it is effectively a continuous spatial scan.
RuView's premise: since Wi-Fi is already scanning the space, why not use the reflected signals to sense the environment?
No cameras, no wearables, no app required. A single $9 ESP32 reading Channel State Information (CSI) can detect people through walls, measure heart rate, recognize motion, and even estimate posture.
2. Core Principle: CSI Is a Spatial Fingerprint, Not a Throughput Metric
Channel State Information is normally internal data Wi-Fi chips use for modulation — how the channel attenuates and how multipath effects are distributed. RuView turns it into sensor data for spatial sensing.
When a person moves in a room, their body acts as a reflector that changes the multipath characteristics of the wireless channel. Chest movement during breathing and blood flow during heartbeats produce measurable phase changes.
RuView's model is remarkably small: 8 KB after 4-bit quantization, with microsecond-level inference on a Raspberry Pi. A pretrained model is published on Hugging Face (ruvnet/wifi-densepose-pretrained), reporting 100% presence-detection accuracy on its validation set.
3. What It Can Sense: From Presence to Posture
| Capability | Method | Accuracy / Speed | |------|------|----------| | Breathing rate | 0.1–0.5 Hz band-pass filter, zero-crossing counting | 6–30 BPM, real-time | | Heart rate | 0.8–2.0 Hz band-pass filter, zero-crossing counting | 40–120 BPM, real-time | | Presence detection | Phase-variance fallback + pretrained head | <1 ms | | 17-keypoint pose estimation | Candle engine, cog-pose-estimation | 8.4 ms cold start (Pi 5) | | Fall detection | Phase-acceleration threshold + 3-frame debounce | <200 ms | | Multi-person counting | Adaptive P95 normalization + deduplication | Real-time, self-calibrating | | Through-wall sensing | Fresnel-zone geometry + multipath modeling | ~5 m |
The key point is not any single metric, but full-stack integration: from $9 hardware to edge inference to smart-home integration.
4. Tech Stack: Rust + ESP32 + Spiking Neural Networks
RuView runs at the edge. Core pipeline:
- ESP32 mesh: multi-node networking, frequency-hopping scans across 6 Wi-Fi bands — even neighboring routers become free radar illuminators
- Spiking neural networks (SNN): adapt to a new environment within 30 seconds, more power-efficient than conventional DNNs
- Cognitum Seed: optional persistent memory layer — vector storage + kNN + an Ed25519 witness chain that cryptographically attests every measurement
- RuVector: the underlying CSI processing engine
- Health: sleep monitoring, apnea screening, elderly activity anomalies
- Security: presence detection, intrusion alerts, fall detection
- Commercial: foot-traffic counting, queue length, meeting-room occupancy
- Industrial: cleanroom personnel tracking, equipment monitoring
- Home Assistant: MQTT publishing with 21 entities (11 raw signals + 10 semantic states), plus 3 starter Blueprints
- Apple Home / HomePod: HAP-1.1 bridge, Siri voice queries
- Google Home / Alexa: via the HA bridge or a Matter endpoint
- SmartThings: Matter Bridge
- No camera — cannot capture your face
- No wearables — nothing to wear
- No app — no phone required
- Fully local processing — no cloud needed
- Cryptographic attestation — every measurement carries an Ed25519 signature
- Through-wall range is about 5 m and depends on signal quality
- Multipath-rich environments (metal, mirrors) may cause interference
- Pose estimation needs camera-supervised fine-tuning for best accuracy
- Wi-Fi bands vary by region; some features are limited by available spectrum
- RuView, GitHub: https://github.com/ruvnet/RuView
- Pretrained model: https://huggingface.co/ruvnet/wifi-densepose-pretrained
- Underlying engine: https://github.com/ruvnet/ruvector
Total BOM (bill of materials) cost is roughly $140.
5. 105 Edge Modules: From Health to Security
RuView is not a single app but an edge intelligence platform. Its 105 pluggable modules cover:
Each module is an independent "Cog" running directly on the ESP32, with no cloud required.
6. Smart-Home Integration: All Four Major Ecosystems
Native support for:
Semantic states include someone-sleeping, possible-distress, room-active, elderly-inactivity-anomaly, and fall-risk-elevated — not raw data, but inferred scenes.
7. Privacy: No Camera, No Cloud
RuView's privacy design is one of its biggest selling points:
This is not a "privacy feature" but privacy at the architectural level — peeking is physically impossible because there is no visual sensor.
8. Limitations
9. Closing: From Connectivity to Sensing
The shift of Wi-Fi from a "connectivity tool" to a "sensing tool" is a long-overlooked dimension in IoT. RuView is not the first CSI sensing project, but it is the first to turn it into a complete product stack — from $9 hardware to edge AI to smart-home integration to privacy by design.
Imagine an elderly person living alone — no camera, no wristband — yet family members can see in Home Assistant: "Dad slept poorly tonight, 3 apnea episodes." This scenario previously required expensive medical-grade equipment; now it needs only an ESP32 plugged into a wall socket.
> "Your router is already scanning the space. RuView just teaches it to read the scan results."
References