Study On Design Of Application Specific Integrated Circuits For Implantable Neural Signal Processing

This paper is for practical application of biomedical engineering and neuralelectrophysiology, based on the intersection of bioscience and information science, researchimplantable neural signal processing algorithm and its application specific integrated circuitdesign with the application specific integrated circuit as the basic means of realization, withimplantable neural recording and neural signal processing as the breakthrough point.As an accurate record of nerve activity means, implantable neural record which is thebasis of nervous system study, diagnosis and treatment of neurological diseases such asepilepsy and Parkinson’s disease, is paid more and more attention by clinic specialist andother experts who engaged in neurophysiology, cognitive science, behavior science andartificial neural network science. With the development of biomedical and microelectronictechnology, the development trend of neural recording system is full implanted, wirelesstransmission and multichannel neural recording, facing the challenge of limited power andbandwith.In order to solve these problems, this paper studies and designs a neural signalprocessing application-specific integrated circuit which is suitable for implantable neuralrecording system. The ASIC could provide online and original neural signal data of a lot ofneurons for neuroscientists and other professional researchers to realize neuroscience researchsuch as Spike-Sorting on the large scale equipments.This paper studies and analyzes the neural signal processing algorithm which is used byexisting implantable neural recording systems, compares advantages and disadvantages ofSpike-Sorting and neural signal compression, choose the vector quantization algorithmwhich is one of the neural signal compression as basic method of neural signal processing forthis paper. On the basis of this, we study and design a improved training sequence algorithm(Linde-Buzo-Gray, LBG) which is based on vector quantization, completed algorithmmodeling on Matlab, finished model simulation with artificial neural signal data and theactual rat cortical neural signal data.On this basis, this paper designs a neural signal processing application-specificintegrated circuit which is suitable for implantable neural recording system based on UMC0.18um1p6m Mixed-Mode1.8V Twin-Well technology. The core area is3.5mm*5mm. With the actual rat cortical neural signal data, we got the postsimulation results:compression ratio124.57, peak signal-to-noise ratio (PSNR)30.16dB, normalized meansquare error of3.82e-04, Spike-Sorting error less than5%. The results show that the neural signal processing application-specific integrated circuitcompress neural signal datas efficiently,keeping the useful information of the original signal.Combined with the existing implantable neural recording system, this application-specificintegrated circuit can provide neural datas of many neuros for neuroscience research, whichsolves the problem of multi-channel neural recording with the limited wirelesscommunication bandwidth for the traditional implantable neural recording system.