Noise-Robust Ensemble Empirical Mode Decomposition with SNR-Adaptive Complementary Perturbations

Noise-assisted ensemble empirical mode decomposition (EEMD) alleviates mode mixing by averaging decompositions of noise-perturbed replicas. In experimental records, however, a fixed injection amplitude may either over-perturb high-SNR data or fail to stabilize envelope estimation when ambient fluctuations are strong, and the ensemble size is limited. We introduce an SNR-adaptive complementary robust ensemble EMD (SCR-EMD). SCR-EMD estimates the ambient noise level from the observation, injects only the incremental perturbation needed to reach a target assistance scale while enforcing a nonzero floor, pairs ± perturbations to suppress injected-noise bias, and aggregates trials using correlation-based weights to down-weight outlier decompositions. Across synthetic multicomponent benchmarks, structured shared-core comparisons with CEEMDAN/ICEEMDAN, added amplitude-modulated and colored-noise stress tests, reconstruction-policy ablations, and ECG-related evaluations, SCR-EMD is most advantageous in high- SNR and low-to-moderate noise conditions, where reducing unnecessary firststage assistance helps avoid over-perturbation, while complete variants remain attractive in heavier-noise regimes. The ECG-related evidence is broadened through a single-record real ECG proof-of-concept and a morphology/artifactdiversity proxy study, whereas the external motorcycle impact-acceleration case is interpreted separately as cautious cross-domain generalizability beyond ECG. SCR-EMD should therefore be viewed as a practical ambient-noise-adaptive complement to complete-ensemble methods rather than a universal replacement, and the preferred IMF cutoff remains signal- and noise-dependent.