Compressed sensing of large-scale local field potentials using adaptive sparsity analysis and non-convex optimization

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Energy consumption and limited bandwidth are critical challenges in wireless neural recording.
• Compressed Sensing (CS) offers efficient data compression for such applications.

Purpose of the Study:

• To propose a novel CS-based approach, Simultaneous Analysis Non-Convex Optimization (SANCO), for large-scale, multi-channel local field potentials (LFPs) recording.
• To enhance energy efficiency and data compression in neural recording systems.

Main Methods:

• Utilized an analysis model to enhance LFP sparsity, addressing limitations of conventional synthesis models.
• Developed an optimal continuous order difference matrix as the analysis operator for improved recovery and resource savings.
• Implemented a non-convex optimizer solvable via the alternating direction method of multipliers for multi-channel LFP reconstruction.

Main Results:

• The SANCO approach demonstrated superior recovery quality and computational efficiency compared to existing CS methods on real datasets.
• Key LFP features were preserved with minimal degradation even at 16x data compression.

Conclusions:

• SANCO is an energy-efficient solution ideal for resource-constrained, large-scale wireless neural recording.
• The method's effectiveness in preserving signal integrity makes it highly suitable for long-term neural monitoring.