Method For Evoked Potential Latency Change Estimation Under Fractional Lower Order Statistics

With the development of science and technology, the application of non-Gaussian stochastic signal processing becomes more and more popular in recent years. In the past, we often analysis the stochastic signal with traditional Gaussian processing because it is easy and suitable in practice application, also it is very helpful to analysis the stochastic process in theory.Although Gaussian model can describe many common signals and noises well, there are so many non-Gaussion signals and noises in the actual applications, for example, there exsit Radar echo, low-frequency atmospheric noise, sound signal from water, artificial signals and noises and some biomedical signals which all satisfy the non-Gaussion processing. Various non-Gassian signals and noises have distinct spiky and impulsive characteristics comparing the graph of the probability densition function for both Gaussion processing and non-Gaussion processing, if such processes are still modeled with Gaussian distribution, the designed signal processor will degenerate for the miss-match between the models and signals, while a-stable distribution is the useful for these processes. As the only type of distribution which satisfy the broad sense center extreme limit axioms, a-stable distribution was first used in signal processing in the year1993, and then large numbers of reaserchers domestic and overseas pay attention to it.This article mainly introduces the parameter estimation of a-stable distribution and the application in Evoked Potential (EP) latency change detection. This article first recalls the histrtorical background of α-stable distribution, and describes the foundmental conceptions, features, principle, the application prospects and the definition of EP signal.Secondly it introduces the parameter estimation method of SaS stable distribution. The characteristic function of SaS distribution is determined by the parameters such as α、β,γand a, so it is important to estimated these parameters. This article uses the log|SaS|method and the samples method for simulation experiment. Experimental results show that both the two effects are good, but the log|SaS|method has a smaller calculation, and has a more superior performance than the quantiles of samples method.According to the non-Gaussian property of the noises in evoked potential latency change detection, we use SaS distribution noise model to describe the property of evoked potential signals and noises. In DLMP (Direct Least Mean p-normal) algorithm, the selection of parameters p relies on the correct estimation of α in noise signals. This new algorithm avoids the shortcoming of estimated the characteristic exponent a. compared with the DLMS (Direct Least Square) and DLMP, Computer simulation shows that this monitoring method has higher estimation accuracy and faster convergence rate.According to the non-stationary property of the noise in EP lantency change decetion, this paper proposes a new parameter estimation method for a-stable noise in evoked potentials. Computer simulation shows that this method can track the change of signal feature effectively and it improves good performance with the dynamic EP latency change detection methods.Finally, because of the non-stationary features of the α-stable distribution. The parameter of noise changes over the time, so this paper research the dynamic analysis on the parameters, we improved DLMP algorithm. The reasults show that the dynamic analysis response to changes in signal and noise characteristics effectively.