Researchers have used a Raspberry Pi single-board computer (SBC) to monitor heart rate without having to wear it or implant any sensors in the patient. A Raspberry Pi is a single-board computer that runs desktop-like operating systems while offering GPIO capabilities. Its price range various from $30 to $120 depending on the model you choose ($50 and upwards if buying one of the new models, such as the Raspberry Pi 5). This is a powerful combination of qualities for developers that want to slap together small machine-learning projects, as these researchers did.
The researchers used the WiFi signals by passing them through the patient’s body, and then analyzing the changes that their pulse causes in the WiFi signals after they have passed through it. For a human, this would be impractical to read. However, machine learning makes this non-intrusive monitoring method possible. They trained a machine learning (a subset of AI) model to decipher the outgoing WiFi signals so that they could then distinguish each heartbeat. They said that the project offered clinical-level accuracy. However, it is important to note that it isn’t necessarily safe to rely on something like this if you built it at home. It would need to be tested thoroughly by a professional.
The concept, called Pulse-Fi uses WiFi Channel State Information to train a low-compute Long Short-Memory (LSTM) neural network model, using two datasets. One was from the ESP-HR-CSI Dataset, and the other was from the EHealth project that uses the Raspberry Pi 4B. The EHealth dataset contained recordings of 118 participants. The Pulse-Fi concept is capable of providing readings in 5 seconds, which puts it on par with popular heart rate monitoring devices such as smartwatches.
Conclusion
This work demonstrates that high‑accuracy, non‑intrusive heart‑rate monitoring can be achieved with inexpensive commodity Wi‑Fi hardware and a lightweight machine‑learning pipeline. By harnessing Channel State Information (CSI) from off‑the‑shelf devices—ESP32 modules for the ESP‑HR‑CSI dataset and a $35 Raspberry Pi 4B for the EHealth recordings — they constructed Pulse‑Fi, a system that processes raw CSI through a custom low‑compute Long Short‑Term Memory (LSTM) network. Evaluation on two complementary datasets, including 118 participants in diverse postures and activities shows that Pulse‑Fi can estimate heart rate within clinical‑level accuracy after just five seconds of observation, regardless of whether the subject is standing, sitting, moving or walking. Importantly, its performance matches or surpasses that of conventional multi‑antenna solutions, which are typically costly and complex to deploy.
The dual‑dataset validation establishes Pulse‑Fi’s robustness across different hardware platforms and real‑world scenarios, making it a strong candidate for scalable deployment in hospitals, eldercare facilities, and at‑home monitoring systems (only after thorough testing and FDA approval). Its low computational demands enable real‑time inference on modest hardware, further reducing barriers to adoption. Future work will explore extending the pipeline to additional vital signs (e.g., respiration rate, blood oxygen saturation) and integrating the system into existing IoT ecosystems for continuous health surveillance. By marrying affordable Wi‑Fi sensing with efficient deep learning, Pulse‑Fi paves the way toward ubiquitous, unobtrusive cardiovascular monitoring that can be widely accessed without compromising accuracy or privacy.
