Abstract: A novel algorithm for optimizing the structure and parameters of Direction of Arrival(DOA) estimation model based on radial basis function neural network is presented. By using the astringency of error criteria function, the number of hidden neurons can be decided reasonably, according to the distribution of signal phase difference between the antenna array, the representative hidden neuron centers can be selected. By this way, the constructed RBF model can be represented the direction finding capacity of the antenna array. Compared with the other Radial Basis Function(RBF) methods, the proposed model has the features of more generalization and accuracy DOA estimation. Experimental results show the effectiveness of the proposed approach.

Key words: direction of arrival wave, Radial Basis Function(RBF) neural network, error criteria function, hidden neuron, initial center

摘要： 提出一种优化径向基函数神经网络来波方位(DOA)估计模型结构和参数的方法。利用误差准则函数的收敛性，合理确定模型的隐层神经元数目，根据阵列信号相位差特征的空间分布特点，选择具有代表性的隐层神经元的中心，构建的RBF神经网络更能反映阵列的测向能力。相比于目前的径向基函数神经网络测向模型的构建方法，改进的DOA估计模型具有更好的泛化性能，能够提高测向精度。实验结果验证了该方法的有效性。

关键词: 来波方位, 径向基函数神经网络, 误差准则函数, 隐层神经元, 初始中心

CLC Number: 

• TP391