A Scale Mixture-Based Stochastic Model of Surface EMG Signals With Variable Variances

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Surface electromyogram (EMG) signals are typically modeled as Gaussian.
• Recent studies indicate non-Gaussian EMG signals during muscle activity.
• A unified model for both Gaussian and non-Gaussian EMG distributions is lacking.

Purpose of the Study:

• To develop a novel stochastic model for surface EMG signals.
• To account for both Gaussian and non-Gaussian EMG distributions within a single framework.
• To investigate the relationship between EMG signal distribution and muscle activity.

Main Methods:

• Formulated a non-Gaussian EMG model using a scale mixture distribution.
• Modeled EMG signal variance as a random variable following an inverse gamma distribution.
• Estimated variance distribution parameters via marginal likelihood maximization.

Main Results:

• The proposed scale mixture model demonstrated a superior fit to experimental EMG data compared to conventional models.
• Experiments with nine participants validated the model's performance.
• Variance distribution parameters were found to correlate with underlying motor unit activity and muscle force.

Conclusions:

• Introduced a scale mixture distribution-based stochastic EMG model capable of capturing non-Gaussianity in muscle activity.
• Experimental validation confirmed the model's effectiveness.
• The model provides a unified scheme for understanding surface EMG signal distributions and their relation to muscle force.