Real-time edge computing design for physiological signal analysis and classification.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Embedded Systems

Background:

• Physiological signals (Electromyography - EMG, Electroencephalography - EEG) offer vital data for applications like rehabilitative robotics and device control.
• Current PC-based signal analysis is compute-intensive, hindering deployment on portable, resource-constrained embedded systems for real-world use, such as improving Activities of Daily Living (ADL).

Purpose of the Study:

• To develop and validate an embedded solution for processing physiological signals (EMG and EEG) on a resource-constrained microcontroller.
• To investigate the use of Cepstrum features for high classification accuracy with minimal input features on an ARM Cortex-M4 processor.

Main Methods:

• Algorithm design, testing, and validation were performed on a PC using Python and Matlab.
• The developed algorithms were deployed onto an ARM-based Cortex-M4 embedded processor.
• Cepstrum features were extracted from EMG and EEG signals for classification.

Main Results:

• The embedded solution achieved an average classification accuracy of 95.34% for five EMG classes.
• The system demonstrated an average classification accuracy of 96.16% for EEG signals on the embedded board.
• Cepstrum features provided high classification accuracy with minimal input, suitable for embedded deployment.

Conclusions:

• The proposed ARM Cortex-M4 based embedded system effectively processes physiological signals (EMG, EEG) with high accuracy.
• This research facilitates the development of portable and user-friendly rehabilitative robotic and assistive device solutions.
• The use of Cepstrum features is validated as an efficient method for accurate signal classification in embedded environments.