实时自适应波束形成器的神经网络方法

A Neural Network Based Approach to Realtime Adaptive Beamforming

作者在Ｈｏｐｆｉｅｌｄ神经网的基础上构造了一种实时处理的自适应波束形成网络。由于实际应用中神经元的联结矩阵有时是奇异的，文章引入正则化方法，有效地提高了网络解的稳定性和处理速度。为了降低编程复杂性，避免计算求协方差矩阵，作者还改进了一种网络，该网络用模拟电路实现不需要任何计算，并且便于用常规的模拟电路实时处理。仿真实验结果表明，这种网比传统的自适应方法有较好的收敛和跟踪性能。

new Hopfield-type neural network approach to adaptive beamforming is pres-ented.Because the link matrix of the elementry neurons is sometimes singular in practical application, a regularization method is introduced in this paper to efficiently increase the stability of solution and the speed of network.In addition,in order to simplify programming and avoid compu-tation of covariance,an improved network is introduced ,which need no computation and easy for realtime processing by the conventional analog cireuit. Simulation results show that it takes advan-tages in both convergence and tracking performances over the conventional techniques.