RF Sensing to Detect Breathing Abnormality using Machine Learning

Abstract

Radio Frequency (RF) sensing offers a completely novel and non-contact approach by exploiting RF reflections from, and through, the body for detecting small respiratory motions. The current study has used RF-based Software Defined Radio Frequency (SDRF) sensing to detect various breathing rates, including fast, normal, and shortness of breath. A correlation matrix and standard deviation analysis of 146 OFDM (Orthogonal Frequency Division Multiplexing) subcarriers was performed to determine signal variability and consistency. Machine learning-based results subsequently show that cleaned data improve subcarrier correlation uniformity, with an enhanced focus on normal breathing, while still maintaining signal features related to all breathing modes. Subcarrier selection, filtering and normalization strengthen the accuracy of the obtained data as the preprocessing stages eliminate the noises and artifacts. This paper demonstrates the reliability of Radio Frequency (RF)-based systems for respiratory monitoring, as well as the possibility of extracting highly detailed features relevant to developing more complex, real-time healthcare solutions.