Promise of embedded system with GPU in artificial leg control: enabling time-frequency feature extraction from...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computational Neuroscience

Background:

• Electromyographic (EMG) signal pattern recognition for artificial limb control faces challenges due to the non-stationary nature of leg EMGs.
• Extracting time-frequency features is crucial for non-stationary signals but computationally intensive for current embedded systems.
• Existing embedded systems lack the real-time computing power required for advanced EMG analysis in prosthetics.

Purpose of the Study:

• To quantify the computational speed of a Graphic Processor Unit (GPU) for EMG time-frequency feature extraction.
• To compare the performance of a GPU-based system against a traditional Central Processing Unit (CPU).
• To assess the feasibility of GPUs for real-time EMG processing in medical applications.

Main Methods:

• Implemented EMG time-frequency feature extraction on both GPU and CPU platforms.
• Measured and compared the computational time required by each system.
• Evaluated performance across different EMG analysis window sizes, specifically noting a 100 ms window.

Main Results:

• The GPU demonstrated a significant increase in computational speed compared to the CPU.
• With a 100 ms EMG analysis window, the GPU was over 50 times faster than the CPU.
• The study confirmed the GPU's capability for high-speed, real-time computation.

Conclusions:

• High-performance GPUs offer a promising solution for accelerating EMG time-frequency feature extraction.
• GPU acceleration is crucial for enabling sophisticated real-time control of EMG-based artificial legs.
• This technology has broad potential for other medical applications demanding rapid computational analysis.