Robust synchronization of cardiac-linked biosensors under peak detection failures: a modality-agnostic approach via structural repair and Huber-IRLS regression

Problem: timing alignment between independently clocked sensors (e.g. ECG and PPG) using inter-beat intervals only, without waveform alignment. Classic estimators assume one-to-one beat correspondence, which missed beats, motion-induced false peaks and gross interval errors break. Pipeline: (i) Dynamic Repair of abnormally short split intervals caused by false peaks; (ii) greedy peak-failure-aware mapping with q ∈ {1, 2} that models isolated missed beats as merged intervals; (iii) Huber-IRLS regression with MAD-based residual gating.

• Form factor: other

Cooperative sensing in clinical instruments, distributed medical devices, and body-area networks requires accurate relative timing between independently clocked sensors, such as electrocardiography (ECG) and photoplethysmography (PPG). Interval-only synchronization operates directly on inter-beat intervals and avoids waveform alignment, but conventional estimators generally assume one-to-one reference–device beat correspondence. Missed beats, motion-induced false peaks, and gross interval errors violate this assumption by creating merged or split intervals and index shifts. We propose a robust interval-only synchronization pipeline comprising: (i) Dynamic Repair of abnormally short split intervals caused by extra (false) peaks; (ii) greedy peak-failure-aware mapping with

q ∈

1 , 2

to represent isolated missed beats as merged intervals; and (iii) Huber-iteratively reweighted least squares regression with median absolute deviation-based residual gating. Approximate uncertainty is quantified using an OLS-style Newey–West HAC covariance estimate (approximate and conditional when gating) and Fieller confidence sets for the ratio-defined delay-dynamics parameter. Monte Carlo ablation on synthetic ECG/PPG data (ORACLE and SIGNAL pipelines) shows that structural peak-detection failures can be catastrophic for OLS (e.g.

RMS

E

after , all

≈

227

ms

under Extra 10% and

≈ 232

ms

under a mixed extreme regime). In contrast, the proposed pipeline maintains robustness; in the mixed extreme regime it achieves

RMS

E

after , in

≈ 37.4

ms

(SIGNAL) and ≈35.9 ms (ORACLE). The real-data experiment was performed as a semi-synthetic feasibility validation using matched ECG/PPG beats from a real recording, with known clock skew imposed on the PPG timestamps and controlled peak failures injected into the resulting device interval sequence.

Verbatim from the publisher's DOI record (opens in a new tab) · Publisher licence: https://creativecommons.org/licenses/by/4.0/

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.