On the selection of the cost function for gradient-based decomposition of surface electromyograms

Area of Science:

• Signal processing
• Biomedical engineering
• Machine learning

Background:

• Blind source separation is crucial for analyzing complex signals.
• Existing methods for sparse pulse sequence (PS) decomposition face limitations.
• Gradient-based optimization offers a promising avenue for improving PS assessment.

Purpose of the Study:

• To discuss and compare cost functions for gradient-based optimization in PS decomposition.
• To provide an analytical framework for evaluating different cost functions.
• To validate theoretical findings with simulations and experimental data.

Main Methods:

• Utilizing a gradient Convolution Kernel Compensation method for blind assessment of sparse pulse sequences (PS).
• Employing multichannel recordings for automatic signal decomposition.
• Applying a gradient algorithm to optimize estimated PSs after compensating for mixing channels.
• Developing an analytical framework to compare different cost functions for optimization.

Main Results:

• The study provides a theoretical framework for comparing cost functions in gradient-based PS optimization.
• Analytical derivations showed strong agreement with numerical simulations using synthetic and real electromyogram data.
• The proposed method demonstrates effectiveness in compensating for unknown mixing channels.

Conclusions:

• The analytical framework offers guidelines for selecting optimal cost functions in gradient-based decomposition.
• The validated methods contribute to the development of more robust blind assessment techniques for sparse pulse sequences.
• This research advances the field of signal processing for biomedical applications.