Radar Signal Processing · Multimodal Cardiac Research

Find cardiac timing candidates
in contactless chest motion.

A research prototype that aligns ECG electrical anchors, SCG mechanical references, and non-contact FMCW radar morphology on a beat-relative axis for AO/AC candidate and PEP/LVET/QS2 analysis.

3 signalsECG · SCG · FMCW Radar
Beat-wiseR-peak anchored alignment
AO / ACMorphology-based candidates
PEP · LVET · QS2CTI analysis outputs

Actual Experiment Evidence

Hardware built and used in the study

These user-captured photographs show the actual ECG, SCG, and FMCW radar hardware and measurement setup. They are not AI-generated or composited equipment; public derivatives use only orientation correction, deterministic cropping, resizing, and privacy blur.

Collage of six actual ECG, SCG, and FMCW radar experiment photographs
Six actual experiment views: ECG acquisition, SCG module, STM32 acquisition, wearable session, Lead II electrodes, and BGT60TR13C radar.

01 · Radar Path

From range bin to cardiac component

The pipeline selects a region of interest after range FFT, extracts phase displacement, and suppresses respiration/motion components before beat analysis.

FMCW radar range FFT and region of interest example
1. Range FFT and ROI
Phase displacement extraction from the selected range bin
2. Phase displacement
Respiration component cancellation example
3. Respiration suppression

02 · Alignment

One beat axis for three signals

ECG R-peaks provide stable beat anchors; SCG references and radar morphology are compared on the same relative time axis.

ECG R peak detection example
Electrical beat anchor
Beat alignment before and after example
Beat-relative alignment

03 · Detection

Candidate fusion, not one rule

Slope, morphology, template, wavelet, and notch detectors contribute timing candidates; confidence and SQI reject unstable beats.

Fusion of AO and AC timing candidates
Detector ensemble and fusion
SQI beat acceptance and rejection
Signal-quality gate
AO candidate

Early-systolic shape

Upstroke, slope, and template similarity propose an opening-related landmark.

AC candidate

Late-systolic shape

Notch, inflection, and temporal continuity track a closure-related candidate.

Quality

Record acceptance

Beat quality and detector agreement make both outputs and exclusion reasons inspectable.

04 · Paper-source evidence

Reviewed figures from the supplied paper

The paper PDF was audited at the embedded-image level. Four substantive figures are public; three companion mask layers were not duplicated. No participant-level raw data or subject CSV is included.

Paper workflow from ECG SCG and radar acquisition to AO AC comparison
Paper workflow · Source-Derived
Normalized ECG SCG and radar waveforms on a common time interval
Multimodal waveform excerpt · Source-Derived
Beat-relative ECG SCG landmarks and radar AO AC candidates
Beat-relative landmarks · Source-Derived
Radar minus SCG relative AO and AC timing difference distributions
Relative timing distribution · Source-Derived

05 · Evidence Boundary

Separate implementation from validation

The repository includes analysis code, firmware, configurations, synthetic algorithm diagrams, and four paper-source figures. It does not publish participant raw data or subject-level result CSV, and it makes no clinical performance claim.

Conceptual PEP LVET and QS2 cardiac time intervals
CTI definitions after beat-wise landmark selection
ItemRepository stateMeaning
Analysis pipelineImplementedPython processing, detectors, SQI, and export documented
Embedded acquisitionDocumentedSTM32 ECG and ESP32 MPU6050 SCG firmware included
Public raw dataNot includedOnly synthetic explanatory figures are public
Absolute valve-event validationFuture validationRequires independent echo, ICG, or PCG reference