The Research Of Adaptive Filtering Algorithm Based On Distributed Network In Non-Gaussian Noise Environment

With the continuous development of signal processing technology,distributed estimation based on wireless sensor networks is widely used in practical applications such as agriculture,military and industrial production.In distributed estimation,the data on the network is self-processed by each node and then interacts with its neighboring nodes locally to estimate the target parameters.The essence of how to improve the speed and accuracy of estimation is an improvement of the distributed adaptive filtering algorithm.Since distributed estimation is mostly applied to real environments,which are mostly non-Gaussian noise.Therefore,the research of distributed adaptive filtering algorithms in non-Gaussian noise environments has practical significance.In this thesis,the corresponding algorithms based on distributed network are investigated in non-Gaussian noise environments to improve the shortcomings of existing algorithms,and the main research work includes the following:(1)The step size is one of the most important parameters affecting the performance of adaptive filtering algorithms.The traditional Least Mean Square algorithm is improved by the variable step size strategy to address the problem that the performance of the algorithm is affected by the fixed step size.Based on the inverse hyperbolic tangent function,the functional equation is transformed and adjusted by introducing parameters to obtain the new step size expression.The performance of the proposed algorithm is compared with several newer algorithms in system identification application,and its applicability is verified in sinusoidal signal denoising and adaptive linear prediction.Finally,the proposed variable step function is introduced into the Diffusion Least Mean Square algorithm with several comparative variable step functions,and then the performance is compared.The simulation results show that the improved Diffusion Least Mean Square algorithm using the proposed variable step function has better convergence performance and stronger robustness.(2)To improve the performance of the Diffusion Maximum Versoria Criterion algorithm in non-Gaussian noise environments,two Diffusion Maximum Versoria Criterion algorithms are proposed by combining the continuous mixed p-norm and recursion ideas.The two algorithms have fast convergence and low steady-state error,respectively,and are based on the same criterion,so they can be further combined convexly and then optimised using the instantaneous transfer structure.Experimental simulations in distributed network with non-Gaussian noise environments show that the resulting algorithm performs well and effectively overcomes the quadratic convergence problem due to convex combination,while providing good tracking performance.(3)In order to solve the problem that the performance of the traditional Diffusion Least Mean Square type algorithm degrades or even diverges in non-Gaussian noise environments,an Improved Diffusion Normalized Least Mean Square algorithm based on the inverse hyperbolic sine function is proposed.Comparative experiments are conducted in non-Gaussian noise environments,and the performance of the algorithm is proved to be improved.Subsequent experimental results in Gaussian noise and non-Gaussian noise environments constructed by the Alpha-stable distribution show that the performance of the algorithm maintains better in both noise environments,and there is no performance degradation in non-Gaussian noise environments.