Non-Invasive RF Sensing for Detecting Breathing Abnormalities Using Software Defined Radios

• Published in: IEEE sensors journal Vol. 21; no. 4; pp. 5111 - 5118
• Main Authors: Ashleibta, Aboajeila Milad, Abbasi, Qammer H., Shah, Syed Aziz, Khalid, Muhammad Arslan, AbuAli, Najah Abed, Imran, Muhammad Ali
• Format: Journal Article
• Language: English
• Published: New York IEEE 15.02.2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Abnormalities; Algorithms; Biomarkers; channel state information; Decision analysis; Decision trees; Discriminant analysis; Frequency modulation; Ground truth; healthcare application; Heart rate; Machine learning; Model accuracy; Monitoring; OFDM; Orthogonal Frequency Division Multiplexing; Radio frequency; Radio signals; Sensors; Software; software defined radio; Software radio; universal software radio peripherals USRPs; Vital signs; Wearable sensors; Wearable technology; Wireless communication; Wireless fidelity; Wireless sensor networks
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2020.3035960

The non-contact continuous monitoring of biomarkers comprising breathing detection and heart rate are essential vital signs to evaluate the general physical health of a patient. As compared to existing methods that need dedicated equipment (such as wearable sensors), the radio frequency (RF) signals can be synthesised to continuously monitor breathing rate in a contact-less setting. In this paper, we proposed the contact less breathing rate detection using universal software radio peripheral (USRP) platform without any wearable sensor. Our system leverage on the channel state information (CSI) to record the minute movement caused by breathing over orthogonal frequency division multiplexing (OFDM) in multiple sub-carriers. We presented a comparison of our breathing rate detection with wearable sensor (ground truth) results for single human subject. In this paper, we used wireless data to train, validate and test different machine learning (ML) algorithms to classify USRP data into normal, shallow and elevated breathing depending on the breathing rate. Although different ML models were developed using the K-Nearest Neighbor (KNN), Discriminant Analysis (DA), Naive Bayes (NB) and Decision Tree (DT) algorithms, however results showed KNN based model provided the highest accuracy for our data (91%) each time the trial was made. DT (71.131%), DA (59.72%) and NB (48.99%). Results presented in this paper showed that USRP based breathing rate is comparable to the wearable sensor demonstrating the potential application of our method to accurately monitor breathing rate of patients in primary or acute setting.