基于深度残差网络的(n,1,m)卷积码盲识别  

Blind Recognition of ( n,1,m ) Convolutional Codes Based onDeep Residual Network

作　　者：刘杰 朱宇轩 马钰 LIU Jie;ZHU Yuxuan;MA Yu(Academy of Systems Engineering,Academy of Military Sciences,Beijing 100191,China;Laboratory of Electromagnetic Space Cognition and Intelligent Control,Beijing 100191,China)

机构地区：[1]军事科学院系统工程研究院,北京100191 [2]电磁空间认知与智能控制技术重点实验室,北京100191

出　　处：《无线电通信技术》2023年第6期1052-1058,共7页Radio Communications Technology

基　　金：信息系统安全技术重点实验室基金~~。

摘　　要：针对传统(n,1,m)卷积码识别方法容错性能较差或所需数据量较大的问题,提出了一种基于深度残差网络(Residual Network, ResNet)的方法。对图像识别领域常用的二维ResNet模型进行结构调整,使其适用于一维卷积编码序列的处理;仿真生成大量卷积码比特序列,以不同的误比特率在序列中随机加入误比特,并按固定长度从序列截取片段作为ResNet的训练样本,分别完成编码类型和起点识别模型的训练;将待识别卷积码序列输入网络,即可输出识别结果。仿真结果表明,相比传统方法,该方法以略高的计算复杂度为代价,获得了更好的容错性和较低的识别数据量需求。Since the traditional recognition methods for(n,1,m)convolutional codes have the problem of poor fault tolerance or demand of large data amount,a method based on deep Residual Network(ResNet)is proposed.Firstly,the structure of two-dimensional ResNet model commonly used in image recognition field is adjusted to adapt to the processing of one-dimensional convolutional coding sequence.Then large amounts of convolutional code bit sequences are generated by simulation,with bit errors randomly adding to sequences under different bit error rates.After which fragments are intercepted from the sequence at a fixed length as training samples for ResNet to complete the training of coding type and starting point recognition model;Finally,network output recognition results as the convolutional code sequence to be identified is input.Simulation results show that the proposed method,compared with traditional methods,achieves better fault tolerance and lower recognition data requirements at the cost of slightly higher computational complexity.

关 键 词：信道编码 盲识别 (n 1 m)卷积码 残差网络 

分 类 号：TN911.22[电子电信—通信与信息系统]