基于卷积神经网络联合多域特征提取的干扰识别算法

干扰识别技术是智能抗干扰通信系统中的关键技术，通过对接收信号中干扰类型的准确判别，可为无线通信系统生成最佳的抗干扰方式提供决策依据。针对无线通信系统中典型压制式干扰的识别问题，本文提出了一种基于卷积神经网络联合多域特征提取（Convolutional Neural Network-based Joint Multi-Domain Feature Extraction，CNN-JMDFE）的干扰识别算法，通过CNN同时对两种预处理增强的数据对象:时频图像与频域序列提取干扰特征，有效利用了两种数据对象的优势，提升了干扰识别性能。仿真结果表明，在对于包含动态和参数随机的干扰识别场景下，CNN-JMDFE算法在干噪比（Jamming-to-Noise Ratio，JNR）≥-2 dB时可准确识别14种类型的干扰，识别性能明显优于基于时频图像或频域序列单一数据对象的基于卷积神经网络自动特征提取（Automatic Feature Extraction-based Convolutional Neural Network，AFE-CNN）算法；与传统的人工特征提取的深度神经网络（Manual Feature Extraction-based Deep Neural Network，MFE-DNN）相比，本文算法显著提升了在低JNR下分类准确率，增强了干扰特征的抗噪性能；对于复合干扰，本文算法同样可取得良好的分类效果，当JNR≥0 dB时可准确分类10种复合干扰。 

Jamming Classification Using Convolutional Neural Network-Based Joint Multi-domain Feature Extraction

? ?Interference recognition technology is a key technology in the intelligent anti-interference communication system. Interference identification can provide a decision-making basis for the system to generate the best anti-interference method by accurately distinguishing the type of interference in the received signal. Aiming at the identification of typical suppression interference in wireless communication systems， this paper proposes an interference identification algorithm based on Convolutional Neural Network-based Joint Multi-Domain Feature Extraction （CNN-JMDFE）. Two types of pre-processing enhanced data， time-frequency images and frequency-domain sequences， are extracted through CNN at the same time， which effectively utilizes the advantages of the two data objects and improves the performance of interference recognition. The simulation results show that， in the case of dynamics and random parameters of interference， the CNN-JMDFE algorithm can accurately identify 14 types of single interference when the Jamming-to-Noise Ratio （JNR） ≥ -2 dB， and the recognition performance is significantly better than that of a single data format based on time-frequency images or frequency-domain sequences by Automatic Feature Extraction based on Convolutional Neural Network （AFE-CNN） algorithm. Compared with traditional manual feature extraction-based Deep Neural Network （MFE-DNN）， the proposed method significantly improves the classification accuracy under low JNR and enhances the anti-noise performance of interference features. For composite interference， the proposed algorithm can accurately classify 10 types of composite interference when JNR ≥ 0 dB.