Abstract:Millimeter-wave radar is promising to provide robust and accurate vital sign monitoring in an unobtrusive manner. However, the radar signal might be distorted in propagation by ambient noise or random body movement, ruining the subtle cardiac activities and destroying the vital sign recovery. In particular, the recovery of electrocardiogram (ECG) signal heavily relies on the deep-learning model and is sensitive to noise. Therefore, this work creatively deconstructs the radar-based ECG recovery into three individual tasks and proposes a multi-task learning (MTL) framework, radarODE-MTL, to increase the robustness against consistent and abrupt noises. In addition, to alleviate the potential conflicts in optimizing individual tasks, a novel multi-task optimization strategy, eccentric gradient alignment (EGA), is proposed to dynamically trim the task-specific gradients based on task difficulties in orthogonal space. The proposed radarODE-MTL with EGA is evaluated on the public dataset with prominent improvements in accuracy, and the performance remains consistent under noises. The experimental results indicate that radarODE-MTL could reconstruct accurate ECG signals robustly from radar signals and imply the application prospect in real-life situations. The code is available at: <a class="link-external link-http" href="http://github.com/ZYY0844/radarODE-MTL" rel="external noopener nofollow">this http URL</a>.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: when using millimeter - wave radar for non - contact electrocardiogram (ECG) signal reconstruction, how to improve the robustness against environmental noise and random body movements. Specifically, the existing radar - based electrocardiogram recovery methods have deficiencies in dealing with noise, especially performing poorly in the face of inevitable random body movement (RBM) noise. In addition, existing methods usually regard electrocardiogram recovery as a single task and it is difficult to optimize multiple related tasks simultaneously, resulting in unstable performance in practical applications. To solve these problems, the authors propose a multi - task learning (MTL) framework "radarODE - MTL" and introduce a new optimization strategy - Eccentric Gradient Alignment (EGA). This framework decomposes the electrocardiogram recovery task into three sub - tasks and uses EGA to dynamically adjust the gradients of each task, thereby improving the overall robustness and accuracy of the model. ### Specific problem description 1. **Noise influence**: - Millimeter - wave radar signals may be interfered by environmental noise or random body movements during the propagation process, which will destroy the subtle cardiac activity information and lead to inaccurate electrocardiogram signal recovery. - Existing methods are not effective in dealing with these noises, especially for random body movement noise, whose amplitude is much larger than that of the cardiac activity itself. 2. **Task complexity**: - Electrocardiogram recovery not only needs to reconstruct morphological features (such as PQRST waves), but also needs to detect the R - wave peak and predict the length of a single cardiac cycle. - There are complex interdependencies among these tasks, but existing methods usually only focus on a single task and it is difficult to optimize multiple tasks simultaneously. 3. **Optimization challenges**: - In multi - task learning, different tasks have different optimization difficulties, and a simple gradient combination may lead to a performance decline (negative transfer) of some tasks. - An effective optimization strategy is required to balance the gradient conflicts and optimization difficulties among different tasks. ### Solution - **Multi - task learning framework**: - Decompose the electrocardiogram recovery task into three sub - tasks: morphological feature reconstruction, R - wave peak detection, and single cardiac cycle length prediction. - Use a shared backbone network to extract time - frequency features and process each sub - task through three specific task decoders respectively. - **Eccentric Gradient Alignment (EGA)**: - EGA eliminates the conflict and dominance problems in the original task gradients through orthogonal projection. - Introduce an eccentric vector to dynamically adjust the joint gradient, making the optimization process benefit all tasks more fairly, especially providing more optimization support for difficult tasks. Through the above methods, the "radarODE - MTL" framework can reconstruct electrocardiogram signals from radar signals more accurately and robustly under various noise conditions and has broad application prospects.

radarODE-MTL: A Multi-Task Learning Framework with Eccentric Gradient Alignment for Robust Radar-Based ECG Reconstruction

radarODE: An ODE-Embedded Deep Learning Model for Contactless ECG Reconstruction from Millimeter-Wave Radar

A Novel Aggregated Multipath Extreme Gradient Boosting Approach for Radar Emitter Classification

Human Recognition with the Optoelectronic Reservoir-Computing-based Micro-Doppler Radar Signal Processing.

ECG waveform generation from radar signals: A deep learning perspective

Physiology and Geometry-Informed Deep Learning Framework for 12-Lead ECG Cardiac Transmembrane Potential Recovery

Contactless Electrocardiogram Monitoring with Millimeter Wave Radar

RadarNet: Noncontact ECG Signal Measurement Based on FMCW Radar

A Robust and Accurate FMCW MIMO Radar Vital Sign Monitoring Framework With 4-D Cardiac Beamformer and Heart-Rate Trace Carving Technique

Deep Learning Radar for High Fidelity Heart Sound Recovery in Real-World Scenarios

Multi-task Learning for Radar Signal Characterisation

Multi-task cascaded assessment of signal quality for long-term single-lead ECG monitoring

Noncontact Multi-Point Vital Sign Monitoring with mmWave MIMO Radar

MEDUSA: Scalable Biometric Sensing in the Wild through Distributed MIMO Radars

Radar-Beat: Contactless beat-by-beat heart rate monitoring for life scenes

E-BDL: Enhanced Band-Dependent Learning Framework for Augmented Radar Sensing

MAE-Based Self-Supervised Pretraining Algorithm for Heart Rate Estimation of Radar Signals

3M: Measuring Vital Signs With Markov-Gauss Model

Mmhrr: Monitoring Heart Rate Recovery with Millimeter Wave Radar