Interpretation boundary for radar AO/AC candidates after ECG-anchored beat alignment

Concurrent acquisition system

ECG, SCG, and FMCW radar are acquired over the same interval, with ECG R-peaks serving as common anchors. The radar is a BGT60TR13C, with STM32F411, ESP-32S, and MPU6050 devices in the acquisition and reference paths.

Physical setup containing ECG electrodes, an SCG sensor, BGT60TR13C radar, and acquisition boards — The concurrent ECG, SCG, and radar hardware used in the experiment.

The streams cannot be compared beat by beat until timestamps, sample rates, start offsets, and missing-sample handling are understood.

From radar phase to beats

The processing path calculates a range FFT from FMCW chirps and selects a range bin or ROI containing chest reflections. The phase of the selected complex sample is unwrapped and converted using wavelength-related scaling to obtain a displacement-related waveform.

ROI movement and respiration can alter cardiac morphology. ROI selection, unwrap discontinuities, filtering, and motion rejection are therefore part of the candidate-timing input conditions. One visual peak is not enough to declare a valve event.

ECG R-peak alignment

ECG, SCG, and FMCW radar cardiac waveforms aligned on a common time axis — SCG and radar waveforms placed on a beat-relative axis anchored to ECG R-peaks.

Each R-peak defines a window around one beat. SCG reference timing and radar morphology are compared within that beat. Resampling or length normalization can change timing, so original sample indices and transformed beat coordinates should both be retained.

The paper-reported scope is 600 seconds of concurrent acquisition and 848 corresponding beats analyzed. These remain paper-reported values rather than a new rerun.

AO/AC candidate timing

AO/AC-related SCG reference points are constructed, then temporally corresponding radar peaks, slope changes, or detector-fusion outputs are selected. Low-SQI or detector-disagreement beats need separate handling. Candidate Timing Interval summaries describe the selected-point distribution.

The AO and AC labels mean morphology-based candidate timing. Radar did not image the aortic valve, and valve opening or closure was not independently observed by Doppler echo. The claim is limited to temporal candidates relative to ECG and SCG references.

Supported results and limits

Concurrent acquisition setup with an STM32 board, ECG electrodes, and FMCW radar — Hardware evidence for the experiment and acquisition path.

I checked the experiment path against the saved analysis source, firmware, configuration files, and hardware photographs. The paper reports 600 seconds of acquisition and 848 corresponding beats. Clinical sensitivity, specificity, and direct valve detection were not measured in this experiment.

Next validation

Validation closer to an anatomical event requires a synchronized independent reference such as echocardiography, ICG, or PCG. The protocol should quantify clock error, annotation rules, beat exclusion, inter-rater agreement, and timing error or Bland–Altman behavior. Until then, candidate terminology remains necessary.