Abstract: Via defining comparability of two FH sequences，mapping FH sequences into high dimension space，calculating inner product of the FH sequences，constructing appropriate kernel matrix，using cluster analysis on FH sequences set，finding cluster structure of characteristic space into which the FH sequences are projected，and then finding out the centroid of every cluster，so the FH sequence corresponding to the nearest point to centroid can be used in FH communication.The cluster analysis algorithm，such as k-means algorithm clustering and spectral clustering，minimize the distance inner cluster and maximize the distance between clusters，so the collision probability is reduced，and the efficiency of FH radio network and utilize of frequency are enhanced.

Key words: Frequency-Hopping（FH） sequences, Frequency-Hopping（FH） collision, K-means clustering algorithm, kernel matrix, cluster analysis

摘要： 通过对任意序列之间相似性的定义，将序列影射到合适的高维空间当中，给出在高维特征空间当中序列之间的内积，构造出可选序列集的核矩阵，进而对跳频序列集进行聚类分析，发现序列集投影在特征空间中的聚类结构，进而分析得出各个聚类的质心，特征空间中离质心最近的点所对应的序列即为跳频电台异步组网的可用跳频序列。k-均值法与谱方法等聚类分析算法保证了聚类内部距离最小化和聚类之间距离最大化，从而减少了跳频序列的碰撞几率，提高了组网的效率和频率利用率。

关键词: 跳频序列, 跳频碰撞, 核矩阵, k-均值法, 聚类分析