【作者】 于志明；

【导师】 郭黎利；

【作者基本信息】 哈尔滨工程大学 ， 通信与信息系统， 2010， 博士

【摘要】 随着无线通信技术以及互联网技术的迅猛发展,无线频谱资源日趋饱和,为了提高频谱的利用率和保证不同体制无线网络的协同工作,满足多种通信业务需求,认知无线电技术孕育而生,其中的频谱感知技术就是解决这一问题的关键技术之一,其目的就是要对各频段所存在的授权及非授权信号类型进行检测和识别,仍然属于通信信号调制识别的范畴。本文针对适用于非协作频谱感知的信号特征提取及识别算法进行了深入的研究,主要内容包括基于统计模式识别的单载波调制信号特征提取技术,单载波与多载波调制信号的类间识别技术,以及多载波调制(MCM)信号的参数估计和类内盲识别技术。首先,在分析单载波调制信号传统统计模式识别算法的基础上,提出了一种基于方向数据统计理论的信号特征提取新算法,利用信号相位服从圆周分布这一特点,在载波频率、带宽和调制指数均未知的情况下,通过一定的角度变换将单载波信号瞬时频率值处理后,作为数据样本,提取分类特征。该算法提取的特征稳定性明显优于传统算法,不因时间的推移或环境的变化而发生显著改变,对样本长度依赖程度较小,类间可分离度好。当信噪比大于01Bd的情况下,特征趋于平稳,具有较高的置信度。为非协作频谱感知中单载波通信信号调制类型的盲识别提供了新的途径。同时指出了算法需要进一步完善和改进的地方。其次,针对瑞利衰落信道条件下的单载波信号和多载波信号的类间盲识别问题,提出了一种改进的高阶累积量组合特征参数提取算法。理论分析证明算法能有效抑制瑞利多径衰落及高斯噪声对接收端识别性能的影响。算法不需任何先验知识,避免了载波同步处理的繁琐过程,可直接对中频采样信号进行处理。仿真验证了改进的识别特征与传统的特征相比具有更好的稳健性,并且对子载波数目不敏感。解决了特征参数动态变化所导致的识别门限确定的困难,降低了误判的概率,提高了识别精度,通过门限判别法或者结合简单的分类器即可取得良好的识别效果。为下一步实现多载波调制信号的类内识别奠定了基础。再次,对基于高阶循环累积量估计多载波CDMA信号子载波频率以及利用多尺度Haar小波变换估计多载波CDMA信号码速率的可行性进行了初步的研究,证明了通过检测特定循环频率处高阶循环累积量较大峰值的位置,来对多载波CDMA信号的子载波频率进行盲估计的切实可行的。随后在信号子载波估计结果的基础上,对基于多尺度Haar小波变换的码片速率估计算法进行了仿真,分析了载波估计频偏对不同多载波信号码速率估计性能的影响。为四类多载波调制信号的类内盲识别,提供了必要的理论基础和参数支持。最后,在多载波调制信号参数盲估计的基础上,提出了基于对构造数据矩阵进行奇异值分解的多载波调制信号盲识别新算法,给出了算法模型及实现框图。以典型的基于IFFT实现的OFDM, MC-CDMA, MC-DS-CDMA和MT-CDMA四种常见但难以区分的多载波调制信号为例,分别在理想高斯白噪声信道以及瑞利多径信道下做了详细的算法理论分析和仿真验证,并针对存在多址干扰时,MC-CDMA信号构造矩阵的较大非零奇异值个数n随用户数量线性增长而造成的单次判定结果失效问题,给出了修正的判定准则,更加适用于实际信道情况。算法无需知道任何多载波调制信号数据信息以及扩频码类型和长度,仅通过构造数据矩阵奇异值梯度序列中较大非零奇异值的个数,即可准确判断多载波调制信号的类型,避免了传统识别算法中特征提取之后的分类器设计的繁琐过程,大大简化了识别流程,而且算法中构造数据矩阵的阶数N值选取不必严格遵守与子载波个数的整数倍关系,可以选取相对较小的N值,以减少算法的运算量,仿真分析证明算法在较低信噪比条件下取得了良好的效果。为多载波调制信号的类内识别,提供了一条新思路,具有较高的实际参考价值。更多还原

【Abstract】 Along with the rapid development of wireless communications technologies and Internet technologies, wireless spectrum resources become increasingly saturated, in order to improve the utility ratio of spectrum, ensure the different heterogeneous wireless networks to work together and meet the needs of a variety of communication services, in this case, cognitive radio technology is born pregnant, of which the spectrum sensing is one of the key technologies to solve this problem, its objective is to detect and recognize the type of authorized or unauthorized signal exist in various frequency bands, it still belongs to communication signal modulation recognition category. In this dissertation, the signal feature extraction and recognition algorithm suitable for non-cooperative spectrum sensing have been deeply researched, mainly include single-carrier modulation signal feature extraction techniques based on statistical pattern recognition, inter-class recognition technology of single-carrier and multi-carrier modulation signals, as well as the multi-carrier modulation (MCM) signal parameter estimation and intra-class blind recognition technologies.First of all, after the analysis of the traditional statistical pattern recognition algorithm of single-carrier modulation signals, a new signal feature extraction algorithm based on the direction data statistical theory is proposed. This algorithm utilizes the characteristic that signal phase obeys the distribution of circular, take the instantaneous frequency values of single-carrier signal processed by a certain angle transform as data samples to extract the classification features, without any prior knowledge of carrier frequency, bandwidth and modulation index. The features extracted by this algorithm is more stable than conventional algorithms, has little changes with the time or the environment, smaller dependence of the sample length and better degree of inter-class separability. When the signal to noise ratio is greater than lOdB, the feature tends to stabilize and has a high degree of confidence. All of this provides a new way for blind recognition of single-carrier communication signal modulation type in non-cooperative spectrum sensing. Then, the further refinement and improvement are also indicated.Secondly, in order to solve the blind recognition problem of single-carrier signals and multi-carrier signals in Rayleigh fading channel, put forward an improved higher-order cumulants combination feature extraction algorithm and prove that algorithm can effectively suppress the effect on recognition performance caused by multi-path Rayleigh fading and Gaussian noise on the receiving end. Without any a priori knowledge, algorithm avoids tedious process of carrier synchronization and processes the sampling IF signals directly. Simulation shows that the improved features is not sensitive to the number of sub-carrier and have better robustness, compared with traditional algorithm. It also solves the difficulties of determining the recognition threshold caused by the dynamic changes of characteristic parameters. In addition, algorithm not only reduces the probability of misjudgment but also improves the identification accuracy, through the threshold discriminance or combining with simple classifiers it can achieve good recognition effect and lay a good foundation for the recognition of multi-carrier modulation signal next.Thirdly, the estimation of sub-carrier frequencies based high-order cyclic cumulants and bit rate based on multi-scale Haar wavelet transform are discussed about multi-carrier CDMA signal. The research shows that it is feasible to estimate sub-carrier frequencies by detecting the peak position of specific cycle-frequency. After this, bite rate estimation algorithm based on multi-scale Haar wavelet transform and the analysis of its performance under the conditions of existence of carrier frequency offset are simulated. All the experimental results provide the necessary theoretical basis and parameters support for the blind recognition of four kinds multi-carrier modulation signals.Finally, on the basis of the blind parameters estimation of the multi-carrier modulation signal, a new blind identification of. multi-carrier modulation signal algorithm based on the structural data matrix singular value decomposition is proposed, including algorithm model and implementation diagram. Take OFDM, MC-CDMA, MC-DS-CDMA and MT-CDMA four typical kinds of multi-carrier modulation signals based on IFFT implementation for example, which are common but difficult to distinguish. Detailed theoretical analysis and algorithm simulation are made respectively in the ideal Gaussian white noise channel and Rayleigh multi-path channel. Since the number of larger non-zero singular value of the MC-CDMA signal’s structure matrix increases linearly with the number of users which leads to the misjudgment caused by the existence of multiple access interference, in one time, so a modified criteria more applicable to the actual channel conditions is presented. Algorithm doesn’t need to know any multi-carrier modulation signal data information as well as the type and length of the spreading code. Only by counting the number of large non-zero singular value in gradient sequence of structural matrix singular values, it can accurately determine the type of multi-carrier modulation signal, not only avoids the tedious process of classifier design in traditional recognition algorithms after feature extraction, but also greatly simplifies the identification process. Furthermore, the order of structural data matrix do not have to strictly abide by the relationships that its value is an integer multiple of the number of sub-carriers, so a relative smaller value can be selected to reduce the computational complexity of algorithm. The simulation and analysis verifies that this algorithm can achieve good results in low SNR conditions. All of above providing a new idea for the intra-class recognition of multi-carrier modulation signals, with a high practical value.更多还原