Research On Adaptive Filtering Algorithm Based On RBFNN

The nonlinear adaptive filter based on radial basis function neural network has the advantages of simple topology,strong nonlinear mapping ability and easy implementation.It is widely used in nonlinear signal processing problems such as system identification,time series prediction and channel equalization.Due to the problems of low filtering accuracy,imbalance between convergence speed and steady-state error in its application,this thesis mainly improves the adaptive filtering algorithm of radial basis function neural network under the traditional network structure and adaptive core network structure in the simulation experimental environment such as system identification and time series prediction.The main research contents are as follows:(1)In the face of the impact of fixed step size on filtering performance,a radial basis function neural network adaptive filtering algorithm based on the improved variable step size minimum mean square algorithm is proposed by improving the variable step size algorithm based on the inverse hyperbolic tangent function through a variable scale function.Apply the improved variable step size algorithm to update three parameters in the network.It is applied to system identification and chaotic time series prediction for simulation experiments.The experimental results show that the algorithm solves the problem of step size parameter selection and the contradiction between convergence speed and steady-state error.(2)A combined radial basis function neural network adaptive filtering algorithm is proposed to address the problem of poor network filtering performance caused by poor robustness of filtering algorithms under noise interference.This algorithm first improves the minimum mean square algorithm based on the sigmoid function,and then combines it with the minimum mean square algorithm using a variable scaling function to update the weight parameters in the network.Finally,a new step size function is constructed for updating and training the network center and width parameters.The algorithm is applied to nonlinear system identification under different noises to carry out simulation experiments.The simulation results show that the algorithm improves the filtering performance of radial basis function neural network under different noises.(3)In order to improve the filtering performance of the radial basis function neural network with adaptive kernel structure,supervised learning and unsupervised learning are mixed,and an adaptive kernel radial basis function neural network filter under the normalized least mean square algorithm is proposed.The supervised learning algorithm is used to update the weights and bias parameters in the network,and the unsupervised learning algorithm is used to select and update the network center and width.The algorithm is applied to different nonlinear system identification and pattern classification for simulation experiments.The simulation results show that the algorithm effectively improves the filtering performance of the adaptive radial basis function neural network.