| Epilepsy is a chronic brain disorder caused by abnormal firing of neurons in the brain.Clinical diagnosis of epilepsy is usually achieved by performing epileptiform waveform(EW)detection on the recorded electroencephalogram(EEG)using electronic instruments.The wearable ambulatory EEG(WAEEG)can transmit EEG data to the base station wirelessly,which has become the frontier research of current epilepsy diagnosis technology.However,long-term recording can generate huge amounts of EEG data,and make wireless transmission very power-hungry,that is unsuitable for the battery powered WAEEG which has stringent power budget.To break through above bottleneck,on-line data reduction for WAEEG has been proposed,whose core task is the low-power hardware design of epileptic event detection algorithm(EEDA).Owing to the good balance between detection accuracy and computational complexity,EEDA based on continuous wavelet transform(CWT)has become the research focus in realizing on-line data reduction,in which the design of low power analog wavelet filter(AWF)for realizing CWT plays the important role.Under this background,this thesis has conducted the research on the design of EEDA for on-line data reduction,and also the implementation theory and method of Gm-C wavelet filter,which mainly involving:1.The epileptic event detection algorithm is investigated,and a novel EW detection method based on the combination of multiple wavelet scales is proposed.A combination model of two wavelet bases for representing epileptic feature is constructed,based on which the high performance EEDA is designed.In order to select the optimal scales of two combined wavelet bases,the scale selection is converted into the parameter optimization problem with upper and lower limits,in which genetic algorithm is used to solve the problem so as to obtain the optimal combination model for EEDA.Experimental results show that the EW detection sensitivity can be improved to 92% by proposed EEDA with low computational complexity and 50% data reduction.2.The approximation method of analog wavelet base is investigated,and an approximation method based on frequency-domain sampling and hybrid genetic algorithm is proposed.According to the magnitude-frequency characteristics of wavelet base,the mathematical model of wavelet approximation is constructed,and the hybrid genetic algorithm combining the quasi-newton method and genetic algorithm is used to find optimum solution.Simulation results show that,compared with the existing methods,the proposed method can improve the performance in the aspects of detection delay,system stability and approximation accuracy.Also,to verify the excellent performance in EW detection,the approximated wavelet base is applied in the designed EEDA with EW detection sensitivity being 92%.3.The design of ultra-low power Gm-C wavelet filter is investigated.The performance of AWF depends on not only the wavelet filter structure,but also Gm cells.Therefore,this thesis uses LC ladder filter structure to synthesize the obtained wavelet approximation function,and then employs quasi-newton method to determine the capacitance values.In addition,the simple differential pair is employed to construct the transconductor with pS-magnitude transconductance,which makes ultra-low frequency EEG signal processing possible.Finally,AWF is designed based on SMIC 0.18 ?m CMOS process.Simulation results show that the proposed Gm-C wavelet filter based on LC ladder structure can achieve high EW detection accuracy,and also low power consumption and small chip size. |
PDF Full Text Request