PaWFE

PaWFE extracts signal features from surface electromyography (sEMG) recordings using parallel windowed processing to accelerate feature computation for machine learning applications in prosthetic hand control.

Key Features:

• Parallel windowed processing: Extracts signal features from multiple time windows simultaneously using parallel processing.
• Time-domain features: Computes widely used time-domain features from sEMG signals.
• Frequency-domain features: Computes frequency-domain features from sEMG signals.
• MATLAB implementation: Implemented in MATLAB.
• Extensibility: Allows addition of new signal feature extraction scripts into the framework.
• Parallel scalability and performance: Scales to multi-core systems with reported speedups up to 20-fold using 32 cores and over 15× improvement versus traditional methods, enabling extraction in seconds for entire acquisitions and under 100 ms per single window.
• Benchmark hardware: Performance benchmarked on a server with four Xeon E7-4820 processors and 128 GB of RAM.
• Benchmark datasets: Evaluated using the first five datasets from the Ninapro database recorded under different acquisition setups.

Scientific Applications:

• sEMG-based prosthetic hand control: Provides features for machine learning pipelines aimed at controlling dexterous and robotic prosthetic hands using sEMG.
• Real-time feature extraction: Enables real-time sEMG feature extraction due to sub-100 ms per-window computation.
• Large-scale sEMG data analysis: Facilitates processing and rapid analysis of large sEMG datasets for algorithm development and evaluation.
• Benchmarking and comparative studies: Supports benchmarking feature extraction performance across acquisition setups and datasets such as Ninapro.

Methodology:

Parallel extraction of signal features from multiple time windows implemented in MATLAB, computing time-domain and frequency-domain features; benchmarking performed on a server with four Xeon E7-4820 processors (128 GB RAM) using the first five Ninapro datasets, reporting up to 20-fold speedups with 32 cores.