A Framework for Compressive On-Chip Action Potential Recording

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Scaling neural recording systems to thousands of channels presents extreme bandwidth challenges for implantable devices.
• Resource constraints in implantable devices necessitate efficient data compression for high-bandwidth neural interfaces.

Purpose of the Study:

• To introduce an adaptive, multi-stage compression framework for high-bandwidth neural interfaces.
• To reduce bandwidth demands for neural recording systems while preserving signal fidelity.

Main Methods:

• Implemented a Wired-OR analog-to-digital compressive readout.
• Developed a digital core for adaptive requantization, selective sampling, and encoding.
• Utilized a mutual information-based criterion for spike sample selection.
• Employed a static entropy coder optimized for neural signal statistics.

Main Results:

• Achieved a 1098× total compression ratio on 512-channel macaque retina data.
• Preserved 90% of recorded spikes.
• Demonstrated that quantization levels can be matched to electrode SNR ($\bm {\lceil \log _{2} \rm{SNR} \rceil }$ bits) to reduce precision without degrading waveform fidelity.

Conclusions:

• The adaptive compression framework effectively addresses bandwidth demands in neural recording systems.
• The system successfully preserves critical neural waveform features and spike data.
• This approach offers a viable solution for resource-constrained, high-channel-count implantable neural interfaces.