A Wearable Bio-signal Processing System with Ultra-low-power SoC and Collaborative Neural Network Classifier for Low...

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Artificial Intelligence

Background:

• Real-time physiological signal classification is crucial for remote health monitoring and human-computer interaction.
• Existing systems often face challenges with high power consumption and data transmission bandwidth.
• Need for efficient, low-power solutions for continuous physiological data analysis.

Purpose of the Study:

• To develop an ultra-low-power, real-time physiological signal classification system.
• To implement an integrated collaborative neural network classifier on a custom System-on-Chip (SoC).
• To demonstrate reduced power consumption and data communication for wearable sensing applications.

Main Methods:

• Designed a custom System-on-Chip (SoC) for ultra-low-power sensing and classification of physiological signals (e.g., EMG, ECG).
• Implemented a collaborative neural network classifier enabling multiple chips to work together.
• Developed a wireless communication module transmitting classification results to a smartphone app for real-time training.

Main Results:

• Achieved 1100X less power consumption in EMG-based gesture classification compared to conventional embedded solutions.
• Reduced data communication by 50X due to low-dimensional data transmission from the collaborative classifier.
• Demonstrated real-time classification and training capabilities via a smartphone interface.

Conclusions:

• The developed SoC and collaborative neural network classifier offer a significant advancement in low-power, real-time physiological signal processing.
• This technology enables more efficient and accessible wearable health monitoring and human-computer interfaces.
• The system's reduced power and data requirements pave the way for widespread adoption in various sensing applications.