基于Fisher判别字典学习的辐射源调制特征识别

摘要

The limited forms of atoms in analytical dictionary lead to sub-optimal matching of atoms and complex emitter signal,resulting in low recognition rate of signal modulation.A dictionary learning method based on Fisher discrimination criterion is proposed to improve the recognition efficiency.The timefrequency transformation of emitter signal is made.The feature vectors are extracted from time-frequency graph using image processing method,which are added class labels.In the dictionary training,the Fisher criterion with small within-class scatter and big between-class scatter is introduced,by which the dictionary not only represents signal more suitably,but also owns better classification performance.The simulated result proves that,compared to analytical dictionary and non-supervision dictionary,the proposed method can obtain a better recognition rate,especially under low SNR.For the atom number Ns =20,Fisher discrimination dictionary can achieve a pretty good balance in recognition rate and calculation amount.%针对基于字典的信号调制类型识别方法中解析字典原子形态单一、无法与复杂辐射源信号最优匹配的问题,提出一种基于Fisher判别准则的字典学习方法.对辐射源信号进行时频分析,借鉴图像处理的方法提取信号时频特征列向量,在字典训练过程中加入信号调制类型信息,根据Fisher准则训练字典,使字典原子类间距离最大同时类内距离最小,以增强字典的识别性能;通过仿真分析Fisher判别字典的识别性能以及原子个数对字典性能的影响.研究结果表明:该方法相比于解析字典法和无监督字典法,具有更好的识别性能,在低信噪比时识别性能突出、抗噪声干扰性能好;综合考虑识别性能和计算量,当字典原子数取20时该方法性能最优.