摘要
基于RLS-BP(Recursion Least Square-Back Propagation,简称RLS-BP)算法提出了一种新的应用于复信道的神经网络盲均衡算法。算法实现了对一个输入、输出和权值都为复数的网络的训练。网络的误差传递采用后向传播(Back Propagation,简称BP)结构,用RLS算法实现网络的训练,这样不仅加快了网络的收敛速度,而且使得均方误差也进一步减小。为了适应复信道,新算法采用常数模(Constant Modulus algorithm,称CMA)算法的代价函数实现算法对复信道的盲均衡。最后对线性复信道和非线性复信道的均衡进行了仿真,结果表明新算法有较快的收敛速度,且稳态均方误差较CMA算法和传统的神经网络盲均衡算法有大幅度的降低。
A novel blind equalization algorithm based on RLS-BP was proposed for equalizing complex communication channel The algorithm could complete to train the network whose input, output and weights were complex. The neural network used error back propagation （BP） structure. And for increasing the converge speed and decreasing the mean square error （MSE）, RLS was applied to train the BP neural network. For equalizing complex channel, the cost function of the network is the same as the CMA algorithm. At last, the numerical simulations for the linear complex channel and the non-linear complex channel were completed. The results show that the proposed algorithm has faster convergence, and much lower MSE than those of BP and CMA algorithm.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2009年第17期5553-5555,5561,共4页
Journal of System Simulation
关键词
神经网络
RLS—BP
复信道
盲均衡
算法
neural network
RLS—BP
complex communication channel
blind equalization
algorithm
