Leveraging CSI of WiFi for Breath Estimation: An RNN-Based Rate Computation Using ESP32 Data

Wireless vital sensing faces challenges due to a lack of non-intrusive, cost-effective solutions for long-term health monitoring. Existing research on Channel State Information (CSI) for vital sign monitoring reveals gaps, particularly in standardized datasets and the use of widely available hardware. This paper addresses these gaps by proposing a novel method that utilizes CSI data for breathing rate estimation with the cost-effective ESP32 WiFi device. A comprehensive CSI dataset was collected for breathing rates from 6 to 30 Breaths Per Minute (BPM), and a specialized Recurrent Neural Network (RNN) model was developed to estimate these rates accurately. The model demonstrated high accuracy, with a Mean Absolute Error (MAE) of 0.04 and an Absolute Percentage Error (APE) of 0.027. Comparative analysis with existing datasets confirms the robustness of this approach. This study highlights the potential for a practical, cost-effective, and non-intrusive solution for long-term health monitoring using existing WiFi infrastructure.