Noncontact Respiratory Abnormality Monitoring: A Hybrid Empirical Mode and Variational Mode Decomposition Approach With Software-Defined Radio and Deep Learning

Respiratory signals are essential vital signs for monitoring conditions, where interrupted respiration can greatly affect health. Software-defined radio (SDR) offers a nonintrusive method for detecting respiratory patterns by observing minor chest wall movements. However, impediments such as disruption, dc components, and respiratory harmonics impede the precise identification of abnormal respiratory patterns like sleep apnea events. The proposed respiratory monitoring system employs SDR technology while utilizing empirical mode decomposition (EMD) and variational mode decomposition (VMD) signal processing methods for effective signal separation and better mode decompositions with Kalman filtering for dc component management. Signal separation becomes better while mode mixing reduction and dc component handling improve efficiently through the integration of a Kalman filter. The proposed system performs real-time respiratory signal extraction while maintaining sub-second processing latency while it achieves high accuracy for recognizing normal, slow, and fast breathing patterns with sleep apnea event detection capabilities. Performance evaluation using Bland-Altman analysis indicates strong agreement with reference respiratory rates (RRs). The convolutional neural network (CNN)-bidirectional long short-term memory (BiLSTM) model—employed for identifying respiratory patterns—achieved an exceptional performance with an accuracy of 99.4%. This contactless approach demonstrates the feasibility of continuous respiratory pattern detection and presents a feasible application in clinical diagnosis and health monitoring systems.