Mp-ECGNet: Dual-subject ECG reconstruction using millimeter-wave radar

Electrocardiogram (ECG) signals are widely regarded as reliable indicators that accurately reflect the physiological state of the human heart. However, traditional contact-based ECG measurement methods may pose risks such as disease transmission, allergic reactions, and inconvenience during long term wear. Consequently, non-contact ECG measurement using radar has attracted increasing attention. Most existing radar based ECG measurement approaches focus primarily on single target scenarios, thereby limiting the broader application potential of radar systems. In this study, we propose a millimeter-wave radar-based dual-target non-contact ECG measurement method. First, radar raw echoes are processed through moving target enhancement and phase unwrapping to achieve preliminary extraction of phase signals. Subsequently, fast independent component analysis (FastICA) is employed to separate phase signals superimposed from multiple targets. For radar phase signals, we design the Mp-ECGNet architecture to reconstruct the target ECG signals. Within Mp-ECGNet, the multiview component block (MVCB) module is responsible for extracting latent multi-periodic components and forming high-dimensional features to mitigate noise interference. The Gaussian-linear module (GLM) then performs weighted aggregation of the features obtained from MVCB, emphasizing the desired periodic components of the ECG signal and thereby enhancing reconstruction performance. Experiments conducted on our self-collected dataset demonstrate that Mp-ECGNet achieves an R-peak reconstruction error of 15.4 ms, outperforming existing radar based ECG reconstruction methods.