Abstract: In order to further improve the convergence and stability of Radial basis function neural network nonlinear adaptive filtering, an improved variable step size least mean square adaptive filtering algorithm based on Radial basis function neural network (RBFNN) was proposed. On the basis of the variable step size least mean square algorithm and the inverse hyperbolic tangent function, a variable scale function is used to replace the fixed parameters in the variable step size model in order to solve the drawbacks of fixed step size in the algorithm and the problem of selecting fixed parameters in the variable step size model. Then, the improved algorithm is applied to RBFNN to update and train the center, width, and output weight parameters of the network, in order to improve the performance of RBFNN filtering under gradient algorithm. Finally, the simulation comparison experiments are carried out in nonlinear system identification and chaotic time series prediction. The results show that the proposed algorithm has obvious advantages in terms of convergence rate and steady-state error performance.

Key words: adaptive filtering, radial basis function neural network, variable step size least mean square algorithm