Commonly used in the array to test the algorithm,signal source is generally considered as the number of prior parameters,when the signal source estimation is error,the performance of position estimation is serious decline,or even completely ineffective.Radial basis is introduced,signal source estimation is changed into a model classification to solve problems,using RBF neural network structure is simple,concise training and learning fast convergence can approximate any of the characteristics of nonlinear function.Finally,implementation of the algorithm is given,and the simulation results with the existing algorithm performance is compared.