Research On The Extraction Method Of Few-Trial Evoked Potentials

Evoked potential(EP)is one of the important signals to judge the integrity of the nervous system pathway or whether it has damage or lesions,its real-time monitoring and analysis has great significance in clinical medical research.However,evoked potentials are often accompanied by strong background noise and low signal-to-noise ratio.Therefore,how to accurately extract weak evoked potentials from strong background noise has been an important issue in the field of neural signal processing.In this dissertation,we have conducted an in-depth study on the methods for the extraction of single-channel evoked potential.The prominent research contents are as follows:1 Single-trial evoked potential extraction method based on sparse representation(SP)and autoregressive model with exogenous input(ARX)is deeply studied.Sparse representation is a powerful tool for signal denoising and separation,and has a significant effect on non-stationary signal processing.In ARX,using sparse representation instead of moving average(MA)model to model EP,in addition,the usage of the over-complete dictionary and the corresponding coefficients to represent signal is more flexible when compared with the application of the autoregressive moving-average(ARMA)model.2 Single-trial evoked potential extraction method based on multi-input single-output autoregressive modeling with exogenous input(MISO-ARX)is deeply studied.In this method,the single-trial EP is considered as a complex containing many components,which may originate from different functional brain sites;these components are extracted simultaneously by ARX.Avoided the mutual interference among various components in the ARX method.In addition,all the parameters are calculated at the same time to ensure that the estimated EP signal reaches an overall optimum.3 Single-trial evoked potential extraction method which based on singular spectrum analysis and sparse representation under ? stable distribution noise is deeply studied.The background signal of EP presents non-Gaussian pulse characteristic according to the clinic.We use ? stable distribution noise instead of traditional Gaussian white noise to simulate spontaneous EEG signals.Firstly,we use SSA to preprocess the non-stationary observation signal to obtain a relatively stable signal and improve the signal-to-noise ratio of the EP signal.Then reconstruct EP signal from the preprocessed signal with SP.Compared with separately using SSA and SP,it has better results.