非线性信道判决反馈快速收敛神经网络均衡器

针对严重符号间串扰及轻度非线性畸变的数学信道，提出了一种神经网络判决反馈均衡器。它将传统的线性信道判决反馈结构巧妙地随入二层感知器中，学习时充分利用训练信号的信息，在可调权数较少的情况下，取得了较好的均衡效果；人为选择初始权值大大提高了收敛速度。模拟结果表明：该均衡器简单实用、便于实时处理，且均衡性能有较大的改善。

A DECISION FEEDBACK NEURAL NETWORK EQUALIZER WITH RAPID CONVERGENCE FOR NON-LINEAR CHANNELS

A decision feedback equalizer eniploying two-layer perceptron structure used for non-linear channels with severe linear d istortion and mild non-linear distortion is proposed in this paper. We put the two-layer neural network skilfully into the conventional structure of the decision feedbackequalizer which is mainly effective on linear channels and substitute trainning signal instead of thedecision feedback signal in the learning process, thus better performance is gained with a small amount of weights; In adition ,rapid convergence is also obtained by choosing initial values of the adaptiveweights,It is shown by simulations that the equalization performance of the proposed equalizer isimproved significantly