The Study On Adaptive Active Control Method Of Non-Gaussian Noise

With the continuous progress of industrial technology,sound noise that damages or hinders hearing and affects attention has become the third major pollution after water pollution and air pollution.Adaptive Active Noise Control(ANC)had the advantages of good noise reduction and easy implementation,and has become the first choice to solve the problem of noise control.However,most existing ANC studies are focused on steady noise with Gaussian distribution.In fact,non-Gaussian noise with high energy,strong randomness and short duration widely exists in many application scenarios such as remote conference room telephone makes it difficult to obtain effective control of common ANC systems.In this thesis,the α steady-state distribution was used to model the non-Gaussian noise.Based on entropy theory,the following research was carried out on the single-channel nonGaussian noise active control algorithm:Firstly,the basic theory of α steady-state distribution was studied in depth,and the principle and properties of entropy theory can suppress non-Gaussian noise are analyzed.It included the FxLMS(Filtered-x Least Mean Square)algorithm based on maximum correlation entropy,the FxLMS algorithm based on maximum mixed correlation entropy and the convex combination FxLMS algorithm based on maximum correlation entropy.Secondly,on the basis of a deeper understanding of entropy theory,considering that the correlation entropy was a measure of the similarity of two random variables,although it was not a correlation operation in the original measurement space,but in the kernel spaces its still the concept of second-order statistics,and non-Gaussian noise generated in line with the α-stable distribution only exists as a low-order moment statistic,and there was no concept of secondorder statistics,which destroyed the performance of the algorithm based on second-order statistics in entropy theory and made it appear divergent.Therefore,this thesis uses the p-norm to modify the kernel function of the entropy theory and proposes an improved maximum correlation entropy algorithm(Improved Maximum Correntropy Criterion Filtered-x Least Mean Square,IMCCFxLMS)and an improved maximum mixture correlation entropy algorithm(Improved Filtered-x Maximum Mixture Correntropy Criterion,IFxMMCC)to avoid this phenomenon and derive the update formula for their weight coefficients.Finally,the convergence and complexity of the proposed IMCCFxLMS algorithm and IFxMMCC algorithm were analyzed respectively,simulation experiments on noise control were also carried out for two noise sources: α steady-state distributed noise and mixed noise.The extensive simulation results showed that although the computational effort of this thesis was slightly larger than that of some existing algorithms,it achieved better results in terms of convergence speed and steady-state error.