Electromyography Signals in Embedded Systems: A Review of Processing and Classification Techniques

Area of Science:

• Biomedical Engineering
• Computer Science
• Signal Processing

Background:

• Electromyography (EMG) signal classification is crucial for real-time wearable applications.
• Implementing EMG algorithms on embedded systems presents challenges in performance and resource constraints.

Purpose of the Study:

• To provide an overview of implementing EMG signal classification algorithms across diverse embedded system architectures.
• To analyze various architectures (microcontrollers, DSP, FPGA, SoC, neuromorphic) for their suitability in wearable EMG applications.

Main Methods:

• Analysis of embedded system architectures based on specifications like movement count and classification type.
• Evaluation of architectures considering precision, processing time, energy consumption, and cost.
• Focus on local inference for artificial intelligence models to optimize execution and resource usage.

Main Results:

• Different embedded architectures offer varying trade-offs in performance and cost for EMG classification.
• Microcontrollers, DSP, FPGAs, SoCs, and neuromorphic chips present unique advantages for specific real-time wearable needs.
• Device selection is contingent upon system specifications, model robustness, classification complexity, and budget.

Conclusions:

• The optimal embedded system for EMG signal classification depends on specific application requirements and constraints.
• This work serves as a reference for selecting appropriate technologies for developing embedded biomedical solutions utilizing EMG.
• Understanding the capabilities of each architecture is key for efficient and effective wearable device development.