Training-free compressed sensing for wireless neural recording using analysis model and group weighted [Formula: see...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Data compression is vital for wireless neural recording due to bandwidth limitations.
• Compressed sensing (CS) offers a promising approach for neural data compression.
• Existing CS methods often require training or dictionary learning, increasing complexity.

Purpose of the Study:

• To propose an analytical, training-free CS recovery method for wireless neural recording.
• To enhance the performance of CS in neural signal processing.
• To address the challenges of resource-constrained applications.

Main Methods:

• Developed group weighted analysis [Formula: see text]-minimization (GWALM).
• Utilized an analysis model to enforce signal sparsity, unlike conventional synthesis models.
• Employed a multi-fractional-order difference matrix as the analysis operator, avoiding dictionary learning.
• Incorporated a group weighting strategy based on statistical properties of analysis coefficients.

Main Results:

• GWALM demonstrated superior performance compared to state-of-the-art CS methods.
• The method achieved higher spike recovery quality.
• Improved classification accuracy was observed in experiments on synthetic and real datasets.

Conclusions:

• GWALM is energy and area efficient, suitable for large-scale, resource-constrained wireless neural recording.
• The training-free nature enhances robustness to spike shape variations, making it practical for long-term use.