S4-Based Self-Supervised Framework Targets Long-Range ECG Representation

SL-S4Wave combines contrastive self-supervised learning with a structured state space model encoder using multiscale subkernels for multichannel physiological waveform modeling, targeting ECG tasks with limited labeled data. The approach addresses known weaknesses of CNN-based SSL methods in capturing long-range dependencies across high-sample-rate signals.