Now a team of researchers from the University of California has demonstrated that ordinary Wi‑Fi hardware can also function as a contactless heart‑rate monitor, no wearables or hardware changes necessary. By siphoning off ephemeral patterns in the radio waves that constantly bounce around a room, their prototype, which they’ve dubbed Pulse‑Fi, can detect beats per minute with a precision that’s comparable to clinical hardware and do so using out-of-the-box components such as a Raspberry Pi 4B or a $5 ESP32 module.
How Wi‑Fi turns a stethoscope
Wi‑Fi waves don’t just travel in straight lines; they bounce all over walls, furniture and people to create a complex tapestry of reflections. Contemporary routers and/or radios now capture that complexity as Channel State Information (CSI) that expresses how each subcarrier of the communication signal (source to destination) behaves. The subcarriers are disturbed slightly by the action of human chest motions and the beating heart. Pulse‑Fi separates those from the background and converts them to a steady pulse count.
This falls under “device‑free” sensing—an increasingly studied domain in both academia and industry, where IEEE papers in the last decade demonstrate Wi‑Fi can track respiration, human presence, and even coarse gestures. Pulse‑Fi takes it a step further, bringing vital‑sign monitoring to clinical‑grade accuracy while maintaining the hardware design simply.
The build: Raspberry Pi, ESP32, and five seconds worth of data
We tested two inexpensive setups: two ESP32 modules, and one raspberry pi 4B. Both are capable of CSI exposure under suitable firmware and driver configuration, therefore allowing raw subcarrier data collection. And that chipset in the Pi gives it the ability to see more subcarrier channels than on those ESP32s (234 observed versus 64 in the experiments), providing a finer-grained snapshot of the radio environment—and, after all, more accuracy.
And crucially, Pulse‑Fi requires only about five seconds of CSI to generate an estimate of your heart rate, which means it’s able to update near real‑time so that you can see an instant response as opposed to having to wait a bit to get the right reading. In experiments, the patient would passively sit, and no sensor would be attached to the patient, the reference (eg, fingertip pulse oximeter) was used as ground truth for comparison.
From a dirty waveform to a clean pulse
Extracting a heartbeat from ambient Wi‑Fi is not trivial. The processing pipeline starts by cleaning up the CSI stream to eliminate environmental noise—sounds of the HVAC, keyboard taps, or light movements in the body. A bandpass filter zeros in on 0.8–2.17 Hz, the frequency where a flexersor signal will correspond to approximately 48–130 beats per minute, excising any aliasing noise.
The subsequent step is characteristic extraction through the subcarriers and a light Long Short‑Term Memory (LSTM) neural network. LSTMs are a good fit for time‑series data; the small model can capture the periodicity without requiring expensive compute. That’s important: the pipeline works well on edge hardware, such as the Raspberry Pi, so you don’t need a power‑hungry server.
Clinical accuracy
Compared with simultaneous measurements by conventional heart‑rate monitors, Pulse‑Fi performed remarkably well. The proposed ESP32 setting had 0.51 BPM of average error in BPM, and 99.38% in accuracy. The Raspberry Pi configuration performed even better: 99.81% accuracy and a 0.2 BPM error. Those numbers compare favorably with what you’d see on a bedside monitor if you were lying in a hospital under controlled conditions, and beat the performance of many consumer wearables at rest, which independent testing shows can be off by a few beats per minute depending on skin tone, motion and fit.
The Pi’s gain results from its more sophisticated subcarrier perspective. The more subcarriers it has, the more times a second it can check for these microscopic periodic movements caused by a beating heart, improving the signal‑to‑noise ratio and its resistance to an odd reflection or two around the room.
Why this matters
Contactless vital-sign monitoring could streamline patient observation in clinics, cut down the wiring mess in step-down units, and bring silent wellness checks to the home. No cumbersome battery‑powered wearables that require recharging, no adhesives, no calibration for each user, no huge bulky boom mics, and no fuss. Just a small unobtrusive device that quietly reads the room. With cardiovascular disease still the world’s No. 1 killer, according to the World Health Organization, cheap and continuous monitoring tools have genuine public‑health stakes.
The economics are also quite favorable: Raspberry Pi 4Bs goes for $30–$50, and ESP32 (and other similar) boards can often be obtained for less than $10. That price makes Pulse‑Fi’s bill of materials accessible to research labs, startups and even advanced hobbyists.
Caveats and what’s next
Like any lab‑grade result there are challenges to bringing this to the real‑world. Both the subject or environmental movement can contribute to signal contamination. Multi‑person scenarios require source separation. Furniture arrangements, spacing and position of the antenna, can influence the performance, while the availability of CSI can vary and it is subject to the provided chipset and driver. The system would require extensive validation across various populations and settings, and regulatory approval from bodies like the FDA, before being put into clinical use.
Privacy also deserves attention. One is that a network that can sense physiological signals must be trustworthy and open about what it’s sensing. That’s a good thing, because thanks to its edge‑friendly design, processing can occur on premises, lowering the frequency at which sensitive information must be sent to the cloud.
The bottom line
Pulse‑Fi demonstrates that using some clever signal processing and a small neural network, it is quite literally possible to pull a persons heartbeat out of mid‑air — using a Raspberry Pi and regular Wi‑Fi. It’s a dramatic symbol of how connectivity infrastructure is transforming into a flexible sensing platform — and a sign that the next era of health monitoring won’t come in the form of a new gadget on your wrist, but as intelligence laid over the air around you.