Research On The Key Techniques Of Neural Recording And Spike Detection In Motor Brain-machine Interface

There are a large number of patients with physical disabilities in the world.The techniques aiming to the recovery of their motor function have been receiving worldwide attention.The motor brain-machine interface is a recently developed technique,which decodes the motion information from the neural spiking activities recorded in the motor cortex and turns them into electric signals that directly control the motor-assist devices like artificial limbs and wheel chairs.Thus helps the disabled recover part of their basic motor function required in their daily lives.So far,the motor brain-machine interface technique has been successfully developed to the stage that uses primates and human as the subjects of the experiments with quite convincing results.It is a promising technique for the rehabilitation of the physical disabilities.However there are still qiute a number of difficulties and challenges need to be solved.For instance,long-term recording requires robust spike detection algorithm;the increasing channel count of implant electrodes challenges neural frontend circuit design on the input impedance,power consumption and chip area;the wireless transfer of neural signal relies on the system integration and miniaturization.Aimint to solve these tough issues,this work studied the key techniques of neural recording and spike detection in the motor brain-machin interface.The research content includes:1.The research on low noise,low power,and small area instrumentation amplifier(IA)design techniques.IA is used to amplify the weak signals recorded by the implanted electrodes.This work analysed the design requirements in the application of neural recordings,including signal amplitude and bandwidth,electrode DC offset voltage cancellation,high input impedance,etc.The EOV cancellation techniuque and input impedance boosting technique have been studied.A noval technique is proposed in this work which introduces both chopping and auto-zero techniques into the capacitively-coupled topology.It achived both low noise and low power and did not require extra area for low pass filtering after chopper modulation.The studied IA design techniques is designed and fabricated in 0.18?m technology.The test showed good results.2.The research on low power,small area analog to digital converter(ADC)design techniques.ADC is used to digitize the neural signal followed by amplifier.Considering the power and area restriction,this work studied successive pproximation ADC(SAR ADC).SAR ADC has relatively fixed structure,including a charge redistribution DAC,a time-domian comparator and a contrl logic.This work studied the design techniques of these building blocks aiming to achieve low power and small area.The capacitve-DAC array was studied and a mixed switching strategy was used to save 75%area and power consumption compared to common voltage based DAC.This work proposed a ladder-based time-domain comparator with a noval voltage controlled delay lines.It achieved larger gain and better noise performance compared to traditional ones.In the control logic,this work used a noval clock distribution circuitry to further decrease power consumption.The studied SAR ADC is designed and fabricated in 0.18?m technology.The test showed good results.3.The research of a probability based spike detection algorithm.Spike detection is to extract spiking activities from neural signals.There have been many long-term recording requirements in the brain-machine interface applications during which the neural environment will inevitably change thus the spike detection algorithm should be unsupervised,robust and adaptive to the changes.This work proposed a noval probability based algorithm called EC-PC spike detection algorithm.Unlike the exsiting arts,this algorithm deals with signal under a probability density function framework.It is able to track the noise and spike power of a neural signal,adaptively refresh its parameters and revise its threshold thus is highly robust to the different neural recroding experiments.At the same time,by chosing the probability threshold,the detection precision can be approximately predicted.So for the first time,the algorithm threshold is closly connected to the detection results and is an usful instruction for the experimentalists.This paper introduced the theory and verified the algorithm through experiments using both simulated data and real data.The algorithm supports online neural spike detection and has been implemented using ASIC.At last,a noval spike detection algorithm based on noise estimation is proposed to reduce the algorithm complexity of EC-PC method from O(NLogN)to O(N).It works under the same probability framework as EC-PC thus has the simillar robustness.It is very suitable for system miniaturization and online spike detection.