神经网络理论在高速通信中应用-软件工程专业论文.docxVIP

摘要摘要均衡技术一直是双绞线接入研究的重点之一。HDSL 中普遍使用的均衡技术是 传统的基于 LMS 算法的自适应判决反馈均衡，它在线路较长、信道非线性影响显 著或者遇到连续突发的冲激噪声时性能迅速下降。而神经网络均衡擅长解决非线 性问题，在无线、移动或者卫星通信中都表现了卓越的性能。因此，本文尝试将 神经网络均衡方法使用到 HDSL 信道中，并比较它与传统 HDSL 均衡方法的性能优 劣。论文首先分析并总结了目前 HDSL 中均衡器的发展现状。根据资料，建立了 HDSL 传输模型。并通过阅读大量文献，学习了神经网络理论及其在信道均衡的应 用方法。选择了两种前向型神经网络结构——多层感知机 MLP 和径向基神经网络 RBF、选择了两种递归神经网络结构——内部递归神经网络 IRN 和 Elman 神经网络 构建了 HDSL 均衡器。最后用 MATLAB 对双绞线下的 HDSL 神经网络均衡器性能进行 了仿真，比较了 DFE 和 MLP、RBF、Elman、IRN 在线性环境和非线性环境下的收敛 特性、正常工作后的均方误差性能，分析了当存在信道漂移和冲激噪声时 DFE 和 MLP、RBF、Elman、IRN 的误码性能，得出了一些有益的结论。关键词：均衡，多层感知机，径向基神经网络，内部递归神经网络AbstractAbstractEqualization techniques are always one of the fatal techniques in twisted-pair access network. Traditional equalization techniques in HDSL are based on LMS algorithm, which is adaptive decision feedback algorithm, though it is widely used, it is not fit in nonlinear conditions. While ANN (Artificial Neural Network) is good at resolving nonlinear problems and it was successfully applied in radio communications and satellite communications. So the paper tries to apply the ANN equalization methods in HDSL and compares its performances with the traditional DFE.First, the paper analyses and summarizes the developments and actuality of equalization in HDSL. Then, according to many papers, a HDSL transfer model is established. And after large study and comparison, the paper chooses Multilayer Perceptron (MLP) , Radial-Basis Function (RBF),Internally Recurrent Net(IRN) and Elman Neural Network(ENN) as the Equlization’s structure. Thirdly, The Neural Network Equlization’s various performances in HDSL are simulated and compared with DFE by MATLAB. Finally, this paper gives some original and useful conclusions.Key Word: equalization, MLP, RBF, IRN目录目录第一章 引言 ....................................................... 11.1 研究背景 ..................................................... 11.2 HDSL 中的均衡技术的现状 21.2.1 自适应判决反馈均衡技术 .................................. 21.2.2 神经网络均衡技术 ........................................ 21.3 主要工作 ..