基于二次星座聚类提取信号特征参数的方法

摘要

针对传统单一的聚类算法在低信噪比时对调制信号识别率低,以及进行特征提取的算法复杂并且难以实现的问题,为了提高星座调制信号在低信噪比的识别率,在对比了不同的MPSK和MQAM的星座图的差异性的基础上,提出了一种二次星座聚类提取信号特征值的方法.其利用改进核函数的DENCLUE(基于密度的聚类)提取信号星座图中密度最大点作为k均值聚类的初始聚类中心,并通过k均值聚类实现了一种新型的提取特征值方法,提取出一组特征值,采用支持向量机构造信号识别模型,并进行分类识别.仿真结果表明,所提出的方法相比传统单一的聚类算法,尤其是在低信噪比下对星座图调制信号的识别率和健壮性较好,并且基于该算法的实际系统简单可靠,具有广阔的发展空间.%Aimed at the problems that traditional single clustering algorithm has the low recognition rate of the modulation signal in low SNR (Signal-Noise Ratio) and feature extraction algorithm is complex and difficult to achieve,in order to improve the recognition rate of the constellation modulation signal in low SNR,an innovative method for twice constellations clustering has been proposed based on the comparison of the differences of MPSK and MQAM constellation diagram,where improved kernel function DENCLUE (density-based clustering) is used to extract the maximum density of the signal constellation as the initial clustering center of K-means clustering.The new feature value extraction method has been implemented for acquisition of characteristic values.The support vector machine has been used to construct the signal recognition model and its classification and identification has been conducted.Simulation results show that the method is better than traditional single clustering algorithm,which is good at the recognition rate and robustness of constellation modulation signal in low SNR especially and that the practical system with this proposed algorithm is simple and reliable with broad development space in future.