基于堆栈式LSTM网络的通信辐射源个体识别  

Communication emitter individual identification based on stacked LSTM network

作　　者：吴子龙 陈红[1] 雷迎科 李昕 熊颢 WU Zilong;CHEN Hong;LEI Yingke;LI Xin;XIONG Hao(College of Electronic Countermeasures,National University of Defense Technology,Hefei 230037,China)

机构地区：[1]国防科技大学电子对抗学院,安徽合肥230037

出　　处：《系统工程与电子技术》2020年第12期2915-2923,共9页Systems Engineering and Electronics

基　　金：国防科技重点实验室基金(9140C130502140C13068);国家自然科学基金(61272333,61473237)资助课题。

摘　　要：针对现有通信辐射源个体识别方法预处理过程复杂及特征提取较难的问题,提出了一种基于堆栈式长短期记忆(long short-term memory,LSTM)网络的辐射源个体识别算法。该算法直接使用IQ时间序列信号训练LSTM网络,即可实现对通信辐射源个体的高效识别,避免了复杂的信号预处理过程。为使LSTM网络能更好地适用于通信辐射源个体识别,利用3层LSTM网络提取辐射源深层特征,并通过实验优化了网络参数。然后对该算法的实际应用泛化性进行了实验探究,结果表明该算法在其他辐射源数据集上也取得了较好的效果。最后,通过实验对算法进行了验证,结果表明相比于传统算法,在样本数较多时,该算法的识别准确率可以达到98%,而且简单快速智能,便于工程化与实用化。For solving the problems of complicated preprocessing and difficult feature extraction for the existing communication emitter individual identification,an algorithm of emitter individual identification based on stacked long short-term memory(LSTM)network is proposed.The algorithm directly uses IQ time series signal to train the LSTM network for realizing efficient individual identification of communication emitter and avoiding complex signal preprocessing.In order to make the LSTM network more suitable for communication emitter individual identification,the three-layer LSTM network is used to extract the deep features of the communication emitter and the network parameters are optimized by experiments.Then the generalization of the algorithm for practical application is investigated experimentally,and results show that the algorithm obtains good effects on other emitter data sets.Finally,the algorithm is verified by experiments,and the results show that compared with the traditional algorithm,the identification accuracy of the algorithm can reach 98% when the number of samples is larger.And the algorithm is more simple,faster and more intelligent,which is suitable for engineering and practical application.

关 键 词：通信辐射源个体识别 长短期记忆 参数优化 泛化性 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]