互耦补偿的神经网络算法用于均匀圆阵波达方向估计

Neural Network Algorithms for DOA Estimation with Compensation of Effects on Mutual Coupling

径向基函数神经网络方法在阵列信号处理中得到广泛应用。但是 ,阵列天线单元间耦合对基函数中心的影响会降低波达方向估计的精度 ,因此有必要对神经网络的输入数据进行互耦补偿 ,以生成正确的基函数中心。本文首先利用矩量法计算天线阵的广义阻抗矩阵 ,再使用直接数据域算法对神经网络的训练数据进行互耦补偿 ,神经网络训练完成后 ,神经网络用于实现波达方向估计。仿真结果表明 ,神经网络算法和直接数据域算法结合具有补偿效果好 。

Radial-basis function neural network (RBFNN) method is widely used in array-signal-processing techniques. However, mutual coupling between the elements of the array affects the computation of basis functions centers and the accuracy of direction of arrival (DOA) estimation is degraded. So compensation of training data for RBFNN is necessary for calculating correct centers. Firstly generalized impedance matrix of array is calculated by the moment method; then the training data for RBFNN are compensated by direct data domain (DDD) algorithm using the generalized impedance matrix; after RBFNN's training, DOA can be accurately estimated. Simulational results show that the algorithm combining RBFNN with DDD has the characteristics of excellent compensation and reducing computation.