The Fast Extraction Methods For EP Signals With Low SNR

Evoked Potential (EP) plays a key role in many fields, the exact and fast extraction of which is the precondition of application. However, EP signals obtained from clinical application directly are often contaminated by the noise of Electroencephalogram (EEG), and the SNR is lower than 0dB, even to-20dB. The traditional superposed average method will make a big error as which needs to stimulate patients for many times. For a fast extraction of EP purpose, it needs to stimulate as fewer times as possible, which makes it possible tracing the variety of EP dynamically. This task is named "dynamic extraction" or "fast extraction", which has been the hot field of EP extraction.This thesis proposes a new method for the fast extraction of EP based on the knowledge of array signal processing after digging into Blind Source Separation (BSS), which can be concluded into two aspects as follows:(1) According to the consistency between array signal processing model and BSS model, a novel algorithm of BSS is proposed. Based on Minimum Output Energy (MOE) principle, the condition for BSS model is presented. What's more, the result under the minimum variance constrained linearly is reached and the independent sources are separated thereby. This method utilizes the knowledge of array signal processing to resolve BSS problems rather than using BSS to resolve array signal processing problems. Compared with typical BSS approaches, this new algorithm does not need to solve the unmixing matrix, so it runs fast, computes at a low complexity and figures out sources precisely at a lower SNR.(2) After setting up the mathematical model according to the physical model of the visual evoked potential (VEP), this thesis puts forward a single-trail VEP extraction method based on the MOE criterion. The new method does not need a prior knowledge on the signal and noise. It can also extract the VEP component fast and high-efficiently at a low SNR from single stimulation compared with BSS methods or adaptive signal processing methods. After being verified with simulations, the proposed method obtains satisfied results with the real VEP data.In addition, it's noted that a new method combining Non-negative Matrix Factorization (NMF) and signal's sparseness is used to extract EP fast in the fifth chapter of thesis, which is a novel try from another aspect of fast extraction, namely the extraction in single-channel and few trials and some valid results are reached by experiments.