通信工程文献翻译——摘要：我们看到数字调制信号自动识别的需求一直在增加，自从20世纪90年代以来，用人工神经网络来实现该功能十分普遍。在此，我们包含了数种为识别的调制类型，例如QAM16，V29，V32，QAM64，是通过新提议的根据统计学特点相加的。通常有两种算法，一种是回归的的BP（RPROP）算法，它带有动量和自适应学习方法，可以自己调整学习率，它有广泛的应用，而RPROP是这个领域中的首选算法，我们可以用独立的样本不断的训练它，让它达到更高的识别率。基因算法（GA）基于习惯性的自组统计值的谱向量，最优特征子集，采用遗传基因法RPROP网络可以在大多数的信噪比下达到99％的识别率。

关键字：调制识别，人工神经网络，基因遗传，特征提取

1.简介

自动调制识别源于军事上通讯智能应用程序例如计算机频道干扰，威胁估计，波长检测，干扰识别等等。所踢出的诸多方法立即为模拟调制设计，最近该科目焦点贡献，更多在数据通讯。根本上，这是由于在新的应用程序中，不断增长数字调制功能的增加。新的应用程序，例如移动电话，私人拨号上网，室内无线网络等等，随着软件无线电系统的上升发展，自动数字调制识别逼从前引起更大的注意。这是单元能作为SDR前端，在解调产生以前，因而一个SDR系统可稳定地处理多个调制。

在早写年，调制识别着中以来于人类操作者对测得数据假设地分类信号。信号特性，如IF波形，信号波长，振幅和相位，常用于传统的方法，较近识别器形式包括解调。没个都设计为一台特别调制类型，既然人类操作员仍然需要”听”输出，我们考虑这作为半自动，但这对数字通讯不实际。

自从20实际80年代以来。人们提议调制石壁诶新的分类。该分类能自动决定引入的调制类型。总的来讲，这些方法分成梁祝要种类。决定理论以及统计模式识别。决定理论方法是利用豁然论说及假设测试论点阐明识别问题。这种方法最大的缺点是形成正确的假设，尽可能详尽的分析很困难。分析用来设置正确的极限值。决定理论方法的例子包括Azzouz及Nandi提出为模拟和数字通讯喜好全球化软件。

然而，模式识别方法不需要如此细致的对待，尽管选择正确特性设置仍然是个问题。这种分类方法可进一步分解为两种：特性释放子系统及识别子系统。特性释放子系统主要负责从原始数据中得出显著特点。已用特性地例子：较高秩序光谱地积累量，星座修整，动力瞬间等等。

第二个字系统，模式识别器，负责分类引入基于解析特性的数据。这方法应用于许多方面，例如：KNN、PNN、SVM等等。选择多层感知作为他们调制识别器系统。

Louis and Sehier 提出使用算法训练的分等级神经网络。他们也给出与其他算法如串级相结合的分析、二进制决定数及KNN。Lu et al提出新颖的基于多层感知的使用几十频率带宽调......

Abstract

Automatic recognition of digital modulation signals has seen increasing demand nowadays. The use of arti"cial neural networks for this purpose has been popular since the late 1990s. Here, we include a variety of modulation types for recognition,e.g. QAM16, V29, V32, QAM64 through the addition of a newly proposed statistical feature set. Two training algorithms for multi-layer perceptron (MLP) recogniser, namely Backpropagation with Momentum and Adaptive Learning Rate is investigated, while resilient backpropagation (RPROP) is proposed for this problem, are employed in this work. In particular,the RPROP algorithm is applied for the "rst time in this area. In conjunction with these algorithms, we use a separate data set as validation set during training cycle to improve generalisation. Genetic algorithm (GA) based feature selection is used to select the best feature subset from the combined statistical and spectral feature set. RPROP MLP recogniser achieves about 99% recognition performance on most SNR values with only six features selected using GA.

Keywords: Modulation recognition; Arti"cial neural networks; Genetic algorithm; Feature subset selection

1. Introduction

Automatic modulation recognition (AMR) has its roots in military communication intelligence applications such as counter channel jamming, spectrum surveillance, threat evaluation, interference identi"-cation, etc. Whilst most methods proposed initially were designed for analogue modulations, the recent contributions in the subject focus more on digital communication [6,17,23,24]. Primarily, this is due to the increasing usage of digital modulations in many novel applications, such as mobile telephony, personal dial-up network, indoor wireless network, etc.

With the rising developments in software-de"ned radio (SDR) systems, automatic digital modulation recognition (ADMR) has gained more attention than ever. Such units can act as front-end to SDR systems before demodulation takes place, thus a single SDR system can robustly handle multiple modulations.

In the early days, modulation recognition relied heavily on human operator’s interpretation of measured parameters to classify signals. Signal properties such as IF waveform,signal spectrum,......