模糊神经网络控制的混合小波神经网络盲均衡算法

摘要

Aiming at the contradiction between the convergence rate and mean square error of traditional Constant Modulus Algorithm (CMA),a hybrid wavelet neural network blind equalization algorithm based on fuzzy neural network controlling (FHWNN) is proposed. In this proposed algofithm,a transversal filter is cascaded to the front end of the wavelet network input layer, outputs of the transversal filter nodes are divided into real and imaginary pans, these two parts signals are merged into one complex signal after passing wavelet network. The proposed algorithm can improve the control precision of step-size via using fuzzy rules of Fuzzy Neural Network (FNN) to control the step-size of scale factor and displacement factor of the WNN. The weight coefficients iterative formulas of the transversal filter and the wavelet neural network are obtained via constant modulus cost function. The theory analysis and simulation result demonstrate that the proposed algorithm has faster convergence rate and smaller steady-state error. Accordingly, it can overcome the contradiction between the convergence rate and mean square error effectively.%针对传统恒模算法(CMA)收敛速度与均方误差之间的矛盾,提出了模糊神经网络控制的混合小波神经网络(FHWNN)盲均衡算法.该算法在小波神经网络输入层之前级联一个横向滤波器,将横向滤波器的节点输出分为实部和虚部两路经过小波神经网络后再合成为一路复数信号;利用模糊神经网络(FNN)设计的模糊规则控制小波函数的尺度因子和平移因子的迭代步长,以提高步长控制的精度;通过常数模代价函数分别获得横向滤波器和小波神经网络的权系数迭代公式.理论分析与仿真结果表明,该算法具有较快的收敛速度和较小的稳态误差,较好地克服了收敛速度与均方误差之间的矛盾.