Random projection for spike sorting: Decoding neural signals the neural network way
Aakash Patil, Shanlan Shen, Enyi Yao, Arindam Basu · 2015
This paper showcases the hardware implementation of Spike Sorting using Extreme Learning Machine (ELM) or its variants based on a first stage of random projections. The Random Projection (RP) operation in the proposed system is implemented on an analog co-processor for ultra low energy operation while the second stage of the classifier consisting of multiply and accumulate is to be performed on a DSP to provide accuracy. Compared to log-normal weight distribution in earlier implementations, we show simple digital post processing to get difference of log normal distribution performs much better. The system accuracy is verified for synthetic dataset with real spike waveforms and achieves misclassification rate of 5%. Fabricated in 0.35μm CMOS and consuming only 14.8 nJ/classify, it has the potential of being integrated into implantable ICs for brain machine interfaces.