基于BP神经网络的盲均衡器设计

为构建用于某语音信号传输系统盲均衡器的BP神经网络模型,编写了基于BP神经网络的盲均衡算法伪代码,计算了算法的时间复杂度,分析了BP神经网络输入层神经元个数、隐含层神经元个数和隐含层层数对盲均衡算法性能的影响,评估了基于Sigmoid的变步长算法、基于误差补偿的变步长算法和基于误差的变步长算法对基于BP神经网络的盲均衡器性能的改进效率,据此设计了一种含双隐层结构的BP神经网络盲均衡器,并对其性能进行了数值仿真分析,明确了其适用范围,为该语音信号传输系统设计提供了技术支撑。

Design of a blind equalizer based on BP neural networks

In order to construct an efficient BP neural network for the blind equalizer in some voice signal communication systems, the pseudo-codes of the blind equalization method based on BP neural networks are programmed, and the time complexity is calculated. Efficiencies of the BP neural networks with variable numbers of the input-layer neurons, of the mid-layer neurons and of the mid layer are evaluated. The efficiency improving the capability of the BP neural networks based on three variable step-size methods, which are developed from the sigmoid function, the error compensatory method and the residual errors, are also evaluated. A blind equalizer based on a two-layer BP neural network structure is obtained. Simulations show its capability and verify its scope of application.