中文摘要：

传统的聚类算法用在MQAM（multilevel quadrature amplitude modulation,多进制正交幅度调制）信号的调制识别中,算法的迭代次数多,特别对高阶调制信号运算时间长.针对此问题,提出了一种半监督聚类重构星座图的方法,由自适应减法聚类确定初始聚类中心,在其周围标记部分样本点并赋予初始隶属度值fik,根据标记的样本点数目确定可信度参数α的值.用fik和α来监督隶属度和聚类中心的更新,误差平方和函数迭代次数减少1/2.接收端识别时,提出基于星座图圆半径的调制识别方式,该方法能很好应对初始聚类中心数目不准确的情况,不需要进行聚类中心的合并与分裂.通过提取接收端星座图的特征参数R并与标准参数Rs进行比较,实现对MQAM信号调制方式的识别.仿真结果表明运算时间是传统聚类算法的1/3,对4～256QAM信号的调制方式识别率在93％以上。

英文摘要：

Traditional clustering algorithms were used in MQAM （multilevel quadrature amplitude modulation） modulation recognition of signals; the number of iterations is more,particularly to the high-order modulation signals.To solve this problem,a method of semi-supervised clustering was presented to reconstruct constellation diagram.The initial clustering centers were determined by the subtractive clustering adaptive.Around them marking part of the sample point and given membership value fik,the value of parameter α of credibility labeled was determined according to the number of sample points.With fik and α to monitor and update the membership of the cluster center,iterations of squared error function reduced by 1/2.When identifying at the receiving end,the modulation recognition method was put foryard based on constellation radius.The method can cope with the number of initial cluster centers inaccurate well,rather than merge with or split cluster centers.At the receiving end,extracting characteristic parameters R and compared with the standard parameters Rs,the recognition of MQAM signal modulation was achieved.Simulation results show that the computation time is 1/3 of the traditional clustering algorithms,the modulation recognition rate of 4 ～256QAM signals over 93％.