【作者】 梅桂圆；

【导师】 郜丽鹏；

【作者基本信息】 哈尔滨工程大学 ， 通信与信息系统， 2011， 硕士

【摘要】 在现代战争中,电子对抗所起的作用越来越大,对侦查信号的处理其好坏与否会影响电子战的胜负,其作用也显得越来越重要。由于现代战争中雷达体制多样化,战场的信号环境越来越复杂,所以对密集交叠的信号流分选其实时性和可靠性要求也越来越高。雷达信号分选是侦查引导系统的一重要组成部分。目前对雷达信号分选算法的研究比较多,传统方法主要包括试探法、统计评估技术和最近邻分类器以及参量范围匹配法、相关函数重频鉴别法等。这些算法理论成熟,应用较多,但是在密集信号环境条件下进行分选时,算法的缺点也比较突出,实时性差,可靠性低,有时甚至会出现无法分选的情况。针对目前的密集的电磁环境,在传统算法上出现了许多改进方法,这其中应用较多的就是直方图法(包括累计差直方图法CDIF和序列差直方图法SDIF)、基于平面变换的分选、聚类分析法、PRI变换法和小波变换法等。利用聚类方法来实现密集脉冲流的初分选,对脉冲流进行稀释,目前的方法比较多,本文主要研究了K-means聚类算法在雷达信号分选中的应用。利用K-means聚类算法对脉冲流进行预分选,其中除了利用常规的载频(CF)、脉宽(PW)、到达角(DOA)这三个参数外,还加入了雷达极化特征这个参数,利用极化可能的四种状态,即水平极化、垂直极化、左旋圆极化和右旋圆极化对信号进行分选。由于极化特征只有这四种状态,所以可先对其进行分选,将脉冲流稀释,这样极化参数可以看成除到达角之外又一个可靠地分选参数。利用K-means算法进行分选时,其中一个较突出的缺点就是初始聚类中心确定后,通过反复迭代,可能会出现局部最优,由于对雷达信号的分选采用了先分选极化参数,然后进行脉宽和到达角的分选,最后进行载频的分选这样一个顺序,所以不论初始聚类中心如何,在经过极化参数及到达角的分选后,通常分选结果不会陷入局部最优,即不会受噪音等的影响,通过仿真实验,也进一步得到了验证。在对密集的信号流分选统计分析后,可以发现该算法可靠性强,分选效果良好,K-means聚类算法另一个缺点就是反复迭代时间较长,这个目前还没有很好的解决,是下一步研究的重点。对于预分选后的信号流,要进行主分选。本文主要研究了相关函数重频鉴别法、直方图法及PRI变换法。相关函数重频鉴别法是PRI变换法的基础,是一种典型的主分选算法,由于利用传统的相关函数重频鉴别法对谐波的抑制较弱,所以常采取谐波压缩的方法,使谐波得到抑制。PRI变换法的基础是自相关函数法,在经过变换加入相位因子后,使谐波分量由向量表示,对其求和后向量和为零,这表示谐波得到抑制,在此基础上,针对该算法的缺点提出了两点改进,改进后基本上可以消除噪声对分选结果的影响。直方图法的改进方法即累计差直方图法(CDIF)和序列差直方图法(SDIF)本文也做了一些研究,这两方法各有优缺点。CDIF算法准确性高,可靠性好,但是运算量大,门限也不是最佳门限,若有大量脉冲丢失,会检测出谐波；SDIF算法是在CDIF算法上的改进,比CDIF算法更快,具有最佳检测门限,但是不适合PRI随机变化的信号分选。论文的最后,给出了整个的一个分选过程,预分选后对其序列检索,并进行参差识别,及对虚假信号进行处理等,仿真实验表明,算法可靠性较强,分选效果较为理想。更多还原

【Abstract】 The electronic warfare plays a more and more important role in modern warfare because the performance of detection signal processing can affect win or lose of warfare, and its role has become increasingly important. Due to diversity of the Radar system in modern warfare, battlefield signals environment has become increasingly complex, so requirements for real-time sorting and reliability to the overlapped dense signal flow are also increasing. Radar signal sorting is a important component of Investigation Guidance System. Currently several algorithms for radar signal sorting have been well studied, more about traditional methods including heuristics, statistical assessment technique, nearest neighbor classifier, parametric range matching method, the correlation function repetition frequency identification method and so on. These algorithms have mature theory basis and widely applied. However, when used to dense signal environment for separation, the algorithm’s shortcomings are prominent, that is, poor real-time capability, low reliability, and sometimes even can not complete separation. Aiming to the current dense electromagnetic environment, there have been many improved method based on traditional algorithms, among of which some method are used more often, for instance, the histogram (including the cumulative difference histogram CDIF and sequence difference histogram SDIF), sorting method based on the sub-plane transform, cluster analysis, PRI transform, wavelet transform method and so on.The initial pulse flow separation for dilution of the pulse stream by use of clustering methods has been widely used. This paper mainly addresses the K-means clustering algorithm application in the radar signal sorting. Using K-means clustering algorithm to pre-pulse flow separation, which in addition to using the conventional carrier frequency (CF), pulse width (PW), angle of arrival (DOA) of these three parameters, also join the polarization characteristics of radar parameter to make use of the four possible polarization states, namely, horizontal polarization, vertical polarization, left-hand circularly polarized and right-hand circular polarization of the signal separation. Because the only four states exist in the polarization characteristics, so we can sort them first. After diluting the pulse flow, the polarization can be seen as another reliable parameter than sorting parameters. Using K-means algorithm for sorting, one of the disadvantage is that, the initial cluster centers determined through repeated iterations may lead to local optimum because of the radar signal sorting separation using the way in such a sequence, first pole parameter, then the pulse width and arrival angle separation, the final sorting of the carrier frequency. Thus, no matter how poor the initial cluster centers is, after sorting DOA and polarization parameters, the results will not fall into local optimum, that is not affected by the impact of noise. The simulation results have further verified the effectiveness of the method. The analysis to Signal flow in the intensive statistical of sorting demonstrates that the algorithm possesses high reliability and good sorting effect. K-means clustering algorithm has a drawback that needs a longer iteration computing time, which will be the next focus of the study.For the pre-sorted signal flow, the primary sorting procedure will be performed later. This paper mainly studies the correlation function method of repetition frequency identification, the histogram method and the PRI transform, among of which Correlation function method of repetition frequency identification based on PRI transform is a typical primary sorting algorithm. The harmonic compression method is often taken due to the poor suppression effect of traditional correlation function method to harmonic components. PRI transform algorithm based on the autocorrelation function method after transformation after joining phase factor expressing by the vector, we find its summation becomes zero, which means that harmonics are suppressed, on the basis of this, to overcome the disadvantage of the algorithm we made two improvements. The improved algorithm can eliminate the noise effect on separation results. Additionally, the paper also addresses the improved Method for the cumulative difference histogram (CDIF) and the sequence difference histogram (SDIF). The two methods have both advantages and disadvantages. CDIF algorithm possesses high accuracy and reliability as well as the weakness of the large computation amount and not optimal threshold. If a large number of pulses are lost CDIF algorithm can detect the harmonic. SDIF, improved on the basis of the CDIF, which is faster than the CDIF algorithm, has the best detection threshold. But it does not apply for sorting the random PRI signal.Lastly, the paper gives a whole sorting process. In the process, First step is to search the Pre-sorted sequence, and then to recognize the residual, finally to process the false signal. Simulation experiments show that the algorithm has good robustness and sorting performance.更多还原