【作者】 吴丹；

【导师】 顾学迈；

【作者基本信息】 哈尔滨工业大学 ， 信息与通信工程， 2007， 博士

【摘要】 通信技术正以日新月异的速度发展,其中以通信信号各种调制方式的变化和进步尤为突出,通信信号调制方式分类识别研究也相继发展起来。由于信号环境的日趋密集,使得常规的识别方法和理论很难适应实际需要,无法有效地对通信信号进行识别,对数字通信信号的识别研究提出了更高的要求。调制方式分类识别在合作领域和非合作领域都具有重要的意义。在合作领域中,调制方式分类识别是软件无线电通用接收机和认知无线电智能接收机的关键技术基础,同时也是瓶颈和核心。调制方式也广泛应用于政府实施有效无线频谱管理、监视民用信号等方面。在非合作领域中,调制方式分类识别更起到了关键的作用。未来战争是以信息优势为基础,这使得通信在战争中的地位变得尤为突出,通信对抗已经成为现代战中电磁领域斗争的焦点。通信对抗需要截获敌方的信号,掌握信号中承载的信息内容,而调制方式分类识别是这一过程的基础。在各种无线通信信号类型中,卫星通信信号占有很大的比例,因此能对卫星通信信号进行调制方式的分类识别具有很高的实用价值。由于卫星无线通信环境的复杂性,特别是战场环境下的不可预知性,使得传统的调制分类识别方法性能下降,难以满足现代战争的需要。因此非理想环境下的调制方式分类识别逐渐引起了人们的重视。在此背景下,本文重点对非理想情况中的大范围信噪比变化下和衰落信道下卫星通信信号的调制方式分类识别算法进行深入和系统的研究。首先,对盲信噪比估计进行研究。信噪比是信号质量的重要衡量标准,对后续的盲均衡和调制方式分类识别算法起到铺垫作用。提出基于子空间和联合信息标准的盲信噪比估计算法,不仅更适用于小样本估计,还可以扩大估计范围。这种算法既适用于加性高斯白噪声信道,也适用于衰落信道。其次,为减弱非理想信道对信号的影响,对信道盲均衡进行深入研究。分析在卫星信号截获中存在的星间链路信道和星地信道,并给出对应的数学模型。在信道盲均衡算法基础上,对超指数算法进行改进,提出一种新颖稳健的超指数算法。这种算法先对信号的噪声功率进行估计,并通过这个噪声功率估计值来减小噪声影响,然后采用二阶累积量和四阶累积量相结合的方法对信道进行均衡。该算法减弱了噪声,克服了原超指数算法在低信噪比下不能维持收敛的缺点,同时保持了收敛速度。再次,将支持向量机和神经网络进行比较,支持向量机不仅结构简单,而且有较强的泛化能力,它可以避免神经网络中的过学习、欠学习和局部最小点等缺陷。针对目前调制识别算法存在的问题,提出了基于支持向量机的三种调制方式分类识别的算法。第一种算法为支持向量机模糊网络,它将多个分类器的结果通过一种新的模糊积分融合在一起,使得算法可以在大的信噪比范围内对信号进行识别,尤其在低信噪比下的识别率比较高。第二种算法为支持向量机自适应调制分类识别算法,它需要借助第二章中的信噪比估计值来选择合适的分类器对信号进行分类,仅使用单个分类器就能够实现大范围信噪比识别能力。第三种算法是基于小波和小波支持向量机调制分类识别算法,将改进的重点放在训练过程上,提出一种改进的训练分类器的方法,来扩大单个分类器信噪比的识别范围。本文也分析了载波误差对识别正确率的影响,仿真结果表明调制方式分类识别算法允许在一定范围内的载波偏差,而且组合分类器比单个分类器具有更好的抑制载波偏差的能力。最后,利用调制方式分类识别试验系统,将部分调制分类识别算法移植到硬件平台上。由信号源发生器E4438产生信号,对算法在硬件平台上的性能进行深入的研究。本文也给出了调制方式分类识别试验系统与卫星测控通信地面站相结合,对某一在轨卫星侦收识别的结果,验证算法的工程可行性和有效性。更多还原

【Abstract】 Communication technologies are developed at a very fast speed, and it is especially represented by the changes and improvements of various communication modulation types. Thus, the algorithms to classify and recognize communication signal modulations are developed. Because the signal environment is becoming more crowded, conventional methods can not meet the needs of real situations, which puts forward high demands for modulation classification and recognition of digital communication signals.Modulation classification and recognition is very significant for both cooperative and non-cooperative field. In cooperative field, modulation classification and recognition is the technical base for universal receiver based on software radio and intelligent receiver based on cognitive radio, and they are also the bottleneck and kernel. Besides, governments apply modulation classification and recognition to implement effective wireless frequency management, surveil civil signals and so on. In non-cooperative field, modulation classification and recognition plays a key function. In the future, wars mainly depend on the superiority of information, which makes communication more outstanding. Anti-communication is the focus in electromagnetic battle field.Anti-communication requires intercepting and capturing enemy’s signals and then getting the information embedded in signals. Modulation classification and recognition is the base of this process. In different kinds of wireless communication signal types, satellite communication signal occupies a big proportion. Thus, modulation classification and recognition for satellite communication signal has very high practical values. Because of the complexity of satellite wireless communication environment, especially the unpredictability of the battlefield, performances of traditional modulation classification and recognition algorithms become worse and can not meet the demands of wars. Thus, modulation classification and recognition algorithms for non-ideal situations have been paid much attention.Based on this background, the problem about modulation classification and recognition algorithms for wide range SNR and fading channels are deeply and systematically developed in this dissertation.Firstly, blind SNR estimation is investigated. SNR is an important criterion of signal quality and it is the groundwork for the following bind equalization and modulation classification and recognition. A blind SNR estimation algorithm is proposed based on signal subspace and combined information criterion. It not only fits for small sample estimation but also enlarges the range of estimation. What’s more, it can be applied in both AWGN channels and fading channels.Secondly, blind equalization is studied in details to reduce the affection of non-ideal channels. Inter-satellite link channel and satellite-land channel are analyzed and their mathematical models are presented. For blind equalization, super-exponential method (SEM) is improved and a novel robust super-exponential method (NRSEM) is proposed. In the method, noise power of intercepted signals is estimated which is used to decrease the influence of noise. Then second order and fourth order comulants are combined to equalize channels. Since noise is decreased effectively, the drawback of bad convergence property in low SNRs of the original SEM is overcame,and the convergence speed is promised.Thirdly, comparisons are made between SVM and neural networks and the results show that SVM has not only simple structure but also strong generalization ability. It avoids over-fitting, under-fitting and local minimum in neural networks. Three modulation classification and recognition algorithms by virtue of support vector machine (SVM) are presented aiming at the problems existing in present modulation recognition algorithms. The first algorithm is SVM fuzzy network. It uses several classifiers and fuses recognition results by a new fuzzy integral, which can widen the range of SNR for modulation recognition. Especially, it has good performance in low SNRs. The second algorithm is adaptive modulation classification and recognition based on SVM. It employs the SNR estimation in the second chapter to select suitable classifier to classify signals. This algorithm only uses single classifier to realize the modulation recognition in wide SNR range. The third algorithm is modulation recognition based on wavelet and wavelet SVM (WSVM). It puts forward an improved classifier training method, and also widen the SNR range of modulation classification with a single classifier. The effects of carrier frequency errors on the classification success rate are analyzed and computer simulations show that carrier frequency errors in a certain range can be accepted in modulation classification and recognition algorithms. What’s more, combined classifiers have better capability to overcome the affections of carrier frequency errors compared with single classifiers.Lastly, a modulation classification and recognition test system is emplyed and a simplified modulation classification and recognition algorithm is transferred to the hardware board. The performance on the hardware board is deeply investigated using the signal source generator E4438. Furthermore, the modulation recognition results of an in-orbit model satellite are given using modulation classification and recognition test system and remote sensing and controlling earth station in Harbin Institute of Technology, which proves the engineering validity of the algorithm.更多还原