摘要
为了解决当前识别算法只能识别一种或者两种码字类型以及人工提取特征复杂的问题,提出了两种基于深层神经网络模型的信道编码类型识别器,即卷积神经网络(convolutional neural network,CNN)识别器和递归CNN(recursive CNN,RCNN)识别器,用于识别接收数据中不同类型的信道码字。将待识别的软解调序列作为自然语言处理中文本分类问题的句子向量进行处理,输入到预先训练好的深层神经网络识别器中进行识别,并分析了字长度对识别准确率的影响,得出了最合适的字长度。实验结果表明,两种识别器都能够有效识别接收数据中多种类型的信道编码,且在信噪比为3 dB时CNN识别器的识别准确率能够达到99%以上,而RCNN识别器在1 dB时就能够达到99%以上的识别准确率。
In order to solve the problem that the current recognition algorithm can only recognize one or two code types and the complexity of manually extracting features,two channel coding type recognizers based on the deep neural network model are proposed,namely,convolutional neural network(CNN)recognizer and recursive CNN(RCNN)recognizer,used to identify different types of channel codewords in received data.The soft demodulation sequence to be recognized is treated as the sentence vector of text classification in natural language processing,input into the pre trained deep neural network recognizer for recognition,and analyze the influence of word length on recognition accuracy,and obtain the most appropriate word length.The experimental results show that both types of recognizers can effectively recognize various types of channel codes in the received data,and the recognition accuracy of the CNN recognizer can reach over 99%when the signal-to-noise ratio is 3 dB,while the RCNN recognizer can achieve over 99%recognition accuracy at 1 dB.
作者
杨宗方
张天骐
马焜然
邹涵
YANG Zongfang;ZHANG Tianqi;MA Kunran;ZOU Han(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Chongqing Key Laboratory of Signal and Information Processing,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2024年第5期1820-1829,共10页
Systems Engineering and Electronics
基金
国家自然科学基金(61671095,61702065,61701067,61771085)
信号与信息处理重庆市市级重点实验室建设项目(CSTC2009CA2003)
重庆市自然基金(cstc2021jcyj-msxmX0836)
重庆市教育委员会科研项目(KJ1600427,KJ1600429)资助课题。
关键词
深层神经网络
信道编码识别器
盲识别
字长度
deep neural networks
channel code recognizer
blind identification
word length