No wristbands, no patches, no gadgets—just Wi-Fi. Find out how your heartbeat can be tracked invisibly using signals already in your home.
In recent years, growing public interest in health has made ubiquitous monitoring a key focus for researchers and users alike. Traditional methods for tracking vital signs rely on wearables that make contact with the body. Although these devices offer reasonable accuracy, they can be inconvenient for long-term use and are not always practical in daily routines. It is now possible, however, to measure heart rate without any physical contact, not even a lightweight wearable.
Imagine sitting in your drawing room while your heart rate is monitored automatically. This is now feasible using Wi-Fi signals already present in homes or offices. When these signals pass through the body, they undergo small changes. By analysing subtle variations in Channel State Information (CSI), the system can detect shifts in heart rate without touch. This non-contact approach improves comfort and increases real-world usability by eliminating the need for continuous user input.
Its appeal lies in its low cost, convenience, and unobtrusiveness. Earlier studies could detect breathing using Wi-Fi but struggled to track heartbeats because the signal changes were extremely faint. By combining CSI with machine learning (ML), current methods filter out irrelevant noise, including reflections from stationary objects, and isolate patterns linked to different heart rates.
| Common ways to access CSI data |
| • ESP32 and Raspberry Pi are the most common open platforms that support CSI. Most projects focus on these. • Arduino boards do not provide CSI support. • Qualcomm and Broadcom chips may support CSI but require specific laptop NICs and a Linux setup (e.g., Ubuntu 16.04 with a WLAN library). • Intel 5300 NICs can extract CSI, but need compatible cards, custom firmware, and specialised tools. Setup is complex, and hardware is hard to source. • Software-defined radios (SDRs) such as HackRF can capture physical-layer signals but not standard Wi‑Fi CSI. • Some high-end routers may expose physical-layer data; users can check documentation or flash custom firmware (e.g., OpenWRT), but this requires care. • The Adafruit HUZZAH32 is an ESP32-based board, so it can support CSI extraction in theory. |
How signals reveal your pulse
As mentioned earlier, in Wi-Fi heart-rate monitoring, CSI is used to analyse how chest movements from the heartbeat affect the Wi-Fi signal. The system captures fine-grained CSI data from standard Wi-Fi devices and uses signal processing and machine learning to extract subtle changes caused by the pulse. Typically, the heartbeat frequency is estimated from peaks in the signal.
CSI is usually available to network vendors to track how data packets move and whether they are received correctly. CSI provides information about the signal’s amplitude and phase, showing how it evolves.
When a transmitter sends a Wi-Fi signal and a receiver captures it after it passes through a person, the signal’s changes can be analysed further. Raw signals are filtered to remove irrelevant information, such as stationary objects in the room. Statistical methods isolate the parts of the signal that correspond to vital signs.
The filtered signal is then passed through a long short-term memory (LSTM) model, which identifies patterns corresponding to different heart rates. For example, the model learns the signal pattern at 60 beats per minute versus 110 beats per minute. By training the system on multiple people with varied heart rates, the model generalises to estimate heart rates accurately for new users.
A training dataset can be sourced from third-party surveys or generated manually using an oximeter. However, if the ground truth is obtained from an oximeter, any inherent bias in its readings is propagated to the system. A bias of around 1 BPM and an error below 5 BPM are generally acceptable, as most applications track trends rather than exact values. A sudden change in heart rate, such as during dizziness or panic, is more significant than the precise numerical value. Even with sensor bias, the system can still detect functional patterns.
Turning Wi-Fi noise into data
Oops! This is an EFY++ article, which means it's our Premium Content. You need to be a Registered User of our website to read its complete content.
Good News: You can register to our website for FREE! CLICK HERE to register now.
Already a registered member? If YES, then simply login to you account below. (TIP: Use 'forgot password' feature and reset and save your new password in your browser, if you forgot the last one!)
