高阶带阻滤波器优化设计研究

众所周知，传统BP神经网络收敛速度慢、学习效率低。之所以如此，主要原因在于人工神经元输出函数的同一化。本文提出的神经网络模型的主要特点是：用正交基函数作人工神经元的输出函数，而且每个神经元的输出函数各不相同。该神经网络模型有效克服了传统BP神经网络收敛速度慢、学习效率低的致命缺陷。本文还详细研究了FIR线性相位滤波器的幅频特性与余弦基函数神经网络算法的关系，给出了高阶带阻滤波器优化设计实例。计算机仿真结果表明了该算法在高阶带阻滤波器设计中的有效性和优异性能。

Study on Optimization Design of the High-degree Band-block Filters

It is very known that the convergence speed of the conventional BP neural networks is slow and its learning efficiency low. So is that, it attributes to the identity of the output functions of the artificial neural networks. The main characteristic of the neural networks presented in this paper is that its output functions are perpendicular and different each other. It conquers effectively the disadvantage about the conventional BP neural networks in the convergence speed and learning efficiency. Furthermore, this paper discusses in detail the relations between the amplitude- frequency characteristic of the FIR high-degree band-block filters with linear phase and the algorithm of the neural networks based on the cosine basis functions. The simulation results show that the algorithm is efficient and excellent in the designs of the high-degree band-block filters.