Waveform Morphology Recovery for Non-Contact Physiological Monitoring

27 concepts19 papers1 proposal

Traditional non-contact physiological monitoring systems—such as remote photoplethysmography (rPPG)—frequently operate like simple pitch detectors. When analyzing a video of a face or a wireless signal, they focus primarily on identifying the dominant frequency to compute an average heart rate. This paradigm is evident in foundational problem framings discussed by Fulan Li et al. in A Temporal Encoder-Decoder Approach to Extracting Blood Volume Pulse Signal Morphology from Face Videos and Yogesh Deshpande et al. in Camera-based Recovery of Cardiovascular Signals from Unconstrained Face Videos using an Attention Network.

While frequency tracking works well for basic fitness metrics, it discards the rich structural information embedded within the pulse wave itself. The full shape of the blood volume pulse (BVP) contains critical markers like the dicrotic notch, systolic slopes, and diastolic decay, as highlighted by Brendan Chwyl et al. in A Bayesian Multi-Scale Framework for Photoplethysmogram Imaging Waveform Processing and by Kavish Doshi in A novel system to collect dual pulse oximetry data for critical congenital heart disease screening research. Recovering this waveform morphology is essential for unlocking advanced clinical diagnostics—such as tracking arterial stiffness, blood pressure variations, and vasomotor tone—completely contactlessly, a direction championed by Xin Liu et al. in Learning Higher-Order Dynamics in Video-Based Cardiac Measurement.

Why the Pulse Wave is Complex

Why isn't the pulse wave just a clean sine wave? The cardiovascular system is an intricate network of elastic vessels. When the left ventricle contracts, it drives a forward pressure wave through the arteries. As this wave encounters branching points and changes in vessel diameter, a portion of the energy reflects backward toward the heart.

The observed peripheral pulse waveform is the mathematical superposition of these forward and reflected waves. As Hong Tang notes in Changes of Femoral Photolethysmographic Waveform Characteristics in Anesthetized Dogs with Increased Blood Pressure Induced by Epinephrine, standard mathematical models often use too few components to capture this complexity, obscuring subtle physiological markers. This superposition creates distinct morphological features: the systolic phase (the initial blood surge), the dicrotic notch (the closure of the aortic valve), and the diastolic phase (systemic blood runoff), which are detailed by Banafsheh Adami et al. in rPPG-SysDiaGAN: Systolic-Diastolic Feature Localization in rPPG Using Generative Adversarial Network with Multi-domain Discriminator.

Simple metrics like pulse wave velocity compress this high-dimensional physical interaction into a single number, failing to fully capture the underlying cardiovascular state, as argued by Jian Li et al. in Thin, soft, wearable system for continuous wireless monitoring of artery blood pressure. Furthermore, capturing advanced metrics like heart rate variability (HRV) or stress levels requires an exceptionally high signal-to-noise ratio (SNR) to resolve precise temporal fluctuations between individual beats, a challenge documented by Zhiwei Wang in Accurate face alignment and adaptive patch selection for heart rate estimation from videos under realistic scenarios and Wei Jing et al. in Transdermal Optical Imaging Reveal Basal Stress via Heart Rate Variability Analysis: A Novel Methodology Comparable to Electrocardiography.

The Destructive Nature of Traditional Filtering

When a raw non-contact signal contains noise from motion, lighting changes, or poor sensor contact, the standard engineering response is to apply a bandpass filter around the expected heart rate frequencies (e.g., 0.5 to 3.0 Hz).

However, this approach is fundamentally destructive to waveform morphology. Bandpass filtering acts as a mathematical smoother; it rounds off sharp peaks, fills in dicrotic notches, and obliterates higher-order dynamics—such as acceleration and second derivatives—that define cardiac valve timings. As Xiang Zhang et al. explain in Wital: A COTS WiFi Devices Based Vital Signs Monitoring System Using NLOS Sensing Model, noise-reduction techniques that favor frequency estimation smooth out the fine-grained temporal details required for clinical diagnostics. This loss of high-fidelity mechanical signatures limits the diagnostic utility of non-contact sensors, as echoed by Luzhou Xu et al. in Soli-enabled noncontact heart rate detection for sleep and meditation tracking.

Mechanisms for High-Fidelity Morphology Recovery

To overcome the limitations of traditional filtering, modern frameworks focus on reconstructing the entire structural shape of the pulse wave using learning-based and statistical frameworks.

Generative Adversarial Networks and Structural Losses

Rather than minimizing mean squared error—which often results in blurry, oversmoothed waveforms—researchers employ adversarial training to force models to generate physiologically realistic pulse shapes. Rencheng Song et al. developed PulseGAN: Learning to Generate Realistic Pulse Waveforms in Remote Photoplethysmography to capture subtle temporal dynamics and spectral characteristics for accurate HRV estimation.

Similarly, the multi-domain discriminator architecture proposed by Banafsheh Adami et al. explicitly localizes and preserves the distinct boundaries of the systolic and diastolic phases. To handle real-world distortions like poor skin-to-sensor contact on wearables, Huy Pham et al. utilized a "PPG-aware" loss function within an adversarial framework in Reliable wrist PPG monitoring by mitigating poor skin sensor contact, proving that restoring the structural integrity of pulse waveforms significantly enhances secondary health metrics.

Multi-Scale and Multi-Wavelength Fusion

Information about pulse morphology is often distributed across different spatial scales and optical wavelengths. The multi-scale framework introduced by Brendan Chwyl et al. uses a Bayesian Residual Transform to suppress illumination artifacts while explicitly preserving fine-grained cardiovascular structures.

In the domain of multi-channel sensing, Manuel Meier et al. demonstrated in Tri-Spectral PPG: Robust Reflective Photoplethysmography by Fusing Multiple Wavelengths for Cardiac Monitoring that fusing green, red, and infrared light channels allows a deep neural network to reconstruct a single high-fidelity waveform, preserving underlying physiological shapes better than single-channel frequency estimation.

Key Foundational Work

A Bayesian Multi-Scale Framework for Photoplethysmogram Imaging Waveform Processing (2016) — contributed 1 concept elements to this direction.
Accurate face alignment and adaptive patch selection for heart rate estimation from videos under realistic scenarios (2018) — contributed 1 concept elements to this direction.
Validation of Instantaneous Respiratory Rate Using Reflectance PPG from Different Body Positions (2018) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_waveform_morphology_recovery_afc7acba,
  title = {waveform morphology recovery},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/waveform-morphology-recovery/rdr_afc7acbac30147539ef76ecd44ff1b20},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_waveform_morphology_recovery_afc7acba_v1,
  title = {waveform morphology recovery},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/waveform-morphology-recovery/rdr_afc7acbac30147539ef76ecd44ff1b20/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake