谱峰搜索的神经网络实现

本文提出了谱峰搜索的神经网络方法,作为一个线性规划问题,通过构造适当的代价函数把谱峰搜索与Hopfield线性规划网络联系起来。网络中神经元的个数与参数空间搜索的分辨率无关,而与传感器阵列中传感器的个数成线性关系,因此神经元的个数并不要求很多,理论分析和仿真实验证明,该方法实际上与MUSIC方法或RD(实数域波达方向估计)法是等价的。

APPROACH FOR APPROXIMATION OF SIGNAL SUBSPACE VIA NEURAL NETWORKS

A neural method to solve the orthogonality search problem arising in direction-of-arrival (DOA) estimation is proposed. The problem is mapped onto the snergy function of the neural network as a linear programming problem. The most important feature of this method hinges upon the fact that the number of neurons in the network is linearly dependent on the number of sensors in the sensor array, rather than the discretization levels of parameter space. Theoretical analysis and simulation results show that the performance of the neural method is exactly equivalent to that of the standard MUSIC method or the real domain DOA estimation method.