A Vector Quantization-Based Spike Compression Approach Dedicated to Multichannel Neural Recording Microsystems.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Electrical Engineering

Background:

• Implantable neural recording systems generate large data volumes, straining wireless transmission bandwidth.
• Efficient data compression is crucial for practical, long-term neural monitoring applications.

Purpose of the Study:

• To develop a hardware-friendly compression module for implantable neural recording microsystems.
• To significantly reduce data bandwidth requirements while preserving neural spike information.

Main Methods:

• A novel approach combining a spike extractor and a vector quantization (VQ)-based spike compressor.
• Utilizing an unsupervised learning process for vector quantization of extracted neural spikes.
• Developing efficient hardware architectures for the compression module.

Main Results:

• Achieved a high spike compression ratio (CR) ranging from 10 to 80.
• Demonstrated significant data reduction while preserving spike waveshapes.
• Implemented on a 180-nm CMOS process, achieving 99.62% classification accuracy at CR 20, with low power (4 µW) and area (0.16 mm²).

Conclusions:

• The proposed compression approach effectively reduces neural data for wireless transmission.
• The developed module is suitable for efficient hardware implementation in implantable microsystems.
• This method enables high-density neural recordings with reduced bandwidth demands.