Decomposition of surface EMG under arbitrary non-stationarities

Characterize the feasibility and limits of decomposing surface electromyography signals into motor neuron spike trains using linear convolutive blind source separation when motor unit action potentials exhibit arbitrary non-stationarities, and determine conditions under which identification remains robust.

Background

Convolutive BSS methods often assume stationarity of MUAPs, which is approximately satisfied during isometric, non-fatiguing contractions but can be violated due to physiological and experimental factors. Recent ultrasound-based results suggest some tolerance to non-stationarity, yet general decomposition under arbitrary non-stationary conditions lacks established theory and methods.

Resolving this question is key for applications involving dynamic contractions, fatigue, and clinical scenarios where stationarity cannot be guaranteed, and for tracking motor units across sessions.

References

Although MU identification during slow dynamic contractions, i.e., where the stationarity violation is modest (at least within limited temporal windows), is possible, the problem of decomposing surface EMG signals under arbitrary non-stationarities is still unresolved and the subject of ongoing research.

— Revisiting convolutive blind source separation for identifying spiking motor neuron activity: From theory to practice (2502.04065 - Klotz et al., 6 Feb 2025) in Discussion, Role of motor unit responses