Rethinking PPG-based Sleep Staging: Datasets, Metrics, and Benchmarks

Summary

Automated sleep staging conventionally assigns one discrete label per 30-second epoch, inheriting the minimum temporal resolution of the clinical AASM scoring standard. Wearable PPG has long been pursued as an ambulatory alternative to polysomnography, but PPG-based staging still trails EEG-based methods by a wide margin. This paper argues that the gap is substantially a mismatch between signal and task rather than an intrinsic limitation of optical sensing.