Non-Contact Monitoring and Recognition of Varied Respiratory Patterns From Dual-Subject Across Sleep Postures Using CW Radar

• Published in: IEEE journal of electromagnetics, RF and microwaves in medicine and biology pp. 1 - 10
• Main Authors: Shahriar, Md, Ferdous, Kaisari, Pramanik, Sourav Kumar, Islam, Shekh M. M.
• Format: Journal Article
• Language: English
• Published: IEEE 2025
• Subjects: Accuracy; Biology; Breathing pattern recognition; continuous-wave radar; Discrete wavelet transforms; Ethics; independent component analysis; maximal overlap discrete wavelet transform; Monitoring; non-contact respiratory monitoring; Pattern recognition; Radar; Radar cross-sections; Sleep apnea; Surface waves
• ISSN: 2469-7249; 2469-7257
• DOI: 10.1109/JERM.2025.3610697

Objectives: Non-contact and continuous respiratory monitoring is vital for detecting health risks such as sleep apnea, cardiac events, and respiratory disorders. Non-contact methods using Radar offer an unobtrusive solution but face challenges when multiple subjects and varied sleep postures are involved. Technology or Method: This study presents a novel non-contact dual-subject respiratory monitoring framework using a 24-GHz continuous-wave (CW) radar combined with Independent Component Analysis-Joint Approximate Diagonalization of Eigenmatrices (ICA-JADE) for isolating individual respiratory patterns from the combined mixtures and maximal overlap discrete wavelet transform (MODWT) for subharmonics compression. To our knowledge, this is the first reported investigation to recognize concurrent respiratory patterns of dual subjects across sleep postures using CW radar. Results: Respiratory signals from five groups of two concurrent subjects (10 subjects total) were successfully separated using ICA-JADE and classified into normal, fast, and slow breathing patterns across supine, side, and prone sleep postures. The system achieved consistently high classification accuracies across postures for normal breathing, with an average accuracy exceeding 90%. Fast breathing patterns were also classified with high accuracy but showed slightly more variability across postures. Slow breathing patterns, particularly in the prone posture, were more challenging to classify due to reduced respiratory displacement and subharmonic interference, leading to an initial accuracy drop to 76.68%. Application of Maximal Overlap Discrete Wavelet Transform (MODWT) enhanced slow breathing signal quality, improving prone posture classification to 85.14%. Across all breathing patterns and postures, the proposed method achieved a maximum overall classification accuracy of 88.48%. Clinical or Biological Impact: This technology paves the way for non-contact, privacy-preserving, continuous respiratory monitoring in sleep studies, intensive care, and home healthcare, with the potential to detect early signs of respiratory and sleep disorders without reliance on wearables.