论文部分内容阅读
为了提高通信信号的识别精度,提出了一种基于核Fisher判别分析(KFDA)的数字调制信号分类器设计方法.将接收信号的高阶累积量作为分类特征向量,利用核函数的思想把非线性向量映射到一个高维空间,并在高维空间中利用线性Fisher判别分析实现数字信号的分类.将多类分类器分解成一系列二类问题,并给出了KFDA用于信号分类的详细流程.仿真实验结果表明,当选择合适的核参数时,基于KFDA的分类精度与支持向量机相当,但由于避免了求解非线性优化问题,故计算复杂度低,训练时间短.
In order to improve the recognition accuracy of communication signals, a design method of digital modulation signal classifier based on Kernel Fisher Discriminant Analysis (KFDA) is proposed. Using the higher order cumulants of received signals as classification eigenvectors, The vector is mapped to a high-dimensional space, and linear Fisher discriminant analysis is used to classify digital signals in high-dimensional space.Multi-class classifiers are decomposed into a series of second-class problems and a detailed flow of KFDA for signal classification is given. Simulation results show that the classification accuracy based on KFDA is equivalent to that of SVM when choosing the appropriate kernel parameters. However, the computational complexity is low and the training time is short because it avoids solving nonlinear optimization problems.