Comparison of spike-sorting algorithms for future hardware implementation.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-machine interfaces (BMIs) require efficient hardware spike sorting for single-unit activity isolation and data compression.
• Real-time, low-power, low-area, and high-accuracy systems are crucial for wireless BMI data transmission.

Purpose of the Study:

• To evaluate spike detection and feature extraction algorithms for hardware implementation in BMIs.
• To identify optimal algorithms balancing accuracy and computational complexity for real-time processing.

Main Methods:

• Evaluation of various spike detection algorithms, focusing on robustness to noise and simplicity.
• Assessment of feature extraction methods, comparing accuracy and computational cost against traditional techniques like Principal Component Analysis (PCA).

Main Results:

• The nonlinear energy operator (NEO) was selected as the optimal spike detection algorithm due to its noise robustness and simplicity.
• The discrete derivatives method was chosen as the optimal feature extraction method, providing high accuracy across signal-to-noise ratios (SNRs) with significantly lower complexity than PCA.

Conclusions:

• The NEO and discrete derivatives methods provide an efficient solution for hardware spike sorting in BMIs.
• These algorithms enable high-accuracy, real-time processing with reduced computational load, facilitating wireless data transmission.