Properties Of Deconvolution Sequence And Their Effects On Auditory Evoked Potentials

A transient auditory evoked potential(AEP)can be recorded from the scalp electrode when a short sound stimulation delivered to the auditory system.The amplitude of AEP is so weak that requires reparative stimuli to enhance the signal-to-noise ratio.However,fast stimulus repetition rate may is cause the overlap of AEP components yielding a continuous response.This process can be described as a linear convolution model between the transient response and the stimulus sequence.If taking into account of the temporal interaction between adjacent stimuli,the nonlinear convolution model with nonlinear AEP components are also needed.AEP can be classified into auditory brainstem response(ABR)and middle-latency response(MLR)based on latency.Several deconvolution algorithms with various pros and cons have been proposed to restore transient AEP components.In this thesis,we systematically introduced the deconvolution techniques and experimental paradigms,and put forward solutions to some key problems.1.Evaluation and optimization of CLAD stimulation sequencesCLAD(Continuous loop averaging deconvolution)is a recent developed linear deconvolution method,the existence of the solution is guaranteed by the random jittering of the stimulus sequences.However,there are two problems in the application which are the noise attenuation performance for a sequence,and the optimization and generation of a suited sequence.This study indicates that the noise attenuation capability of a sequence needs to be estimated in combination with the statistical properties of the underlying noise.We thus specifically confirmed that under the ABR and MLR recording conditions,EEG noise is in line with 1/f model.We therefore presented a new noise gain factor to measure the sequence performance.We validate this metric using sufficient measured EEG data to demonstrate that new metric is highly correlated with the true noise attenuation ability.As an objective function,we used classical genetic algorithm as an example to optimize the sequence under given constraint conditions.2.Comparison of MLSs in 40Hz stimulation rateIt is well-known that the MLS is sensitive to the stimulation rate.The inphase enhancement of MLR components are used to account for the enhanced 40 Hz steady-state response.However,there are also evident experiments explained differently the generation of 40Hz steady state response.Since existing linear deconvolution techniques can restore 40Hz transient AEP components,we attempts to address two issues:the basic characteristics of the 40Hz transient response compared to conventional MLR;and whether 40Hz AEPs can identify the sequencing effect.We chose CLAD and MSAD without real time jittering sequence to restive 40Hz transient AEPs and compared them with conventional MLR.The results showed that 40Hz MLR was more stable and evident,while the rapid stimulation of different jittering had a significant effect on some components of MLR,indicating that MLR was significantly affected by temporal property of sequence.3.Analysis of the noise property of MLS for nonlinear deconvolutionThe maximum length sequence(MLS)is a classical deconvolution sequence,which can be used to recover linear and nonlinear AEP components by cross-correlation techniques.MLS is a pseudo-random sequence based on primitive polynomials of different orders,which may affect the additive noise during deconvolution.The purpose of this study is to evaluate the relationship between the noise attenuation of the MLS and the sequence(or polynomial)order.We derived theoretical formulas to measure the noise attenuation for the average process and cross correlation process,and verified with the simulation and the real ABR nonlinear component extraction experiment.Theoretical and experimental results show that noise attenuation is independent of the order,but is determined by the total length of valid EEG data and the stimulus rate of MLS.This study provides the basis for the selection of MLS and can be used to estimate the time and signal-to-noise ratio required for MLS experiments.