AI-Enhanced CW Radar for Vital Sign Monitoring in ISAC Applications

This paper presents a lightweight, free-running Continuous Wave (CW) radar system for non-contact vital sign monitoring with enhanced sensing accuracy through deep learning-based frequency correction. Unlike conventional radar systems requiring complex Frequency-Modulated Continuous Wave (FMCW) architectures or hardware-level stabilization, the proposed system adopts a low-cost CW radar configuration with audio waveform acquisition, enabling compact deployment and scalable data processing. To address the inherent frequency instability of free-running systems, a neural network-based correction scheme is employed. Among several candidate models evaluated, Bidirectional Long Short-Term Memory (Bi-LSTM) model was selected for its superior accuracy and computational efficiency. Experimental results demonstrate that the proposed system achieves over 97% accuracy in respiration and heart rate frequency estimation, with low power consumption and compact size. These characteristics make the system suitable for integration as a sensing block within block-level Integrated Sensing and Communication (ISAC) architectures, particularly in healthcare applications where precise physiological monitoring is critical.