Researchers at the University of California, Santa Cruz have developed a new method to measure heart rate using household WiFi signals, eliminating the need for wearable devices. The technology, called “Pulse-Fi,” uses low-cost WiFi equipment and a machine learning algorithm to monitor heart rate with accuracy comparable to clinical methods.
The team, led by Professor of Computer Science and Engineering Katia Obraczka, Ph.D. student Nayan Bhatia, and high school visiting researcher Pranay Kocheta from UC Santa Cruz’s Baskin School of Engineering, designed a system that interprets subtle changes in WiFi radio frequency waves as they pass through a person’s body. These changes can be detected and analyzed to determine heart rate.
“WiFi devices push out radio frequency waves into physical space around them and toward a receiving device, typically a computer or phone. As the waves pass through objects in space, some of the wave is absorbed into those objects, causing mathematically detectable changes in the wave.”
The Pulse-Fi system operates by filtering out environmental noise from the signal data. “The signal is very sensitive to the environment, so we have to select the right filters to remove all the unnecessary noise,” Bhatia said.
Experiments were conducted with 118 participants using inexpensive ESP32 chips costing between $5 and $10 and Raspberry Pi chips costing about $30. The study found that after five seconds of processing data from these devices, Pulse-Fi could measure heart rate with an error margin of only half a beat per minute. The accuracy improved further with longer monitoring periods.
Testing showed consistent results regardless of participants’ positions—sitting, standing, lying down, or walking—and regardless of where equipment was placed in the room. The system also worked effectively when participants were up to three meters away from the hardware.
“What we found was that because of the machine learning model, that distance apart basically had no effect on performance, which was a very big struggle for past models,” Kocheta said. “The other thing was position—all the different things you encounter in day to day life, we wanted to make sure we were robust to however a person is living.”
To develop their machine learning model, researchers created their own dataset by collecting WiFi signal data along with reference measurements from an oximeter at UC Santa Cruz’s Science and Engineering library. They also used an external dataset produced by Brazilian researchers employing Raspberry Pi devices.
Beyond measuring heart rate, researchers are now working on extending this approach to detect breathing rates as well—a development that could aid in identifying conditions like sleep apnea. Early unpublished findings suggest promising results for these additional applications.
Those interested in commercial applications can contact Marc Oettinger at UC Santa Cruz for more information: marc.oettinger@ucsc.edu.
