降维训练神经网络的圆阵列二维角估计

在信号二维达波方向估计中,神经网络训练集往往较大,一定程度上影响了神经网络在二维角估计中的应用,为此,提出一种降维训练方法,构造两个输入量,并各自形成独立训练集,在两个网络中分别训练俯仰角和方位角,极大地缩简了训练集。论文对降维训练方法进行了仿真,结果表明,所提方法十分有效。

Dimension-Degraded Training Method for 2D Direction Estimation Based on Neural Network

Size of the training set is very large in the two-dimensional (2D) direction estimation based on neural network, as a result it prevents neural network from being widely used in estimation for 2D DOA. In order to reduce the training set, a so-called dimension-degraded training (DDT) method is proposed in this paper. In the DDT method, two input variants are constructed to generate two corresponding sets which are used to train two different networks for elevation and azimuth respectively. The DDT method dramatically reduces the training set while keeping the high resolution of angle. Simulations show the validity of the proposed method.