基于二维异步同相正交直方图的调制方式识别  

Modulation recognition based on the two-dimensional asynchronous in-phase quadrature histogram

作　　者：万鹏武 惠茜 陈东瑞 吴波 WAN Pengwu;HUI Xi;CHEN Dongrui;WU Bo(School of Communications and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;SUNNY SCIENCE&TECHNOLOGY CO.,LTD,Xi’an 710075,China)

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121 [2]陕西山利科技发展有限责任公司,陕西西安710075

出　　处：《西安电子科技大学学报》2024年第4期78-90,共13页Journal of Xidian University

基　　金：国家自然科学基金(62101441);重点实验室基金资助项目2022-JCJQ-LB-006(6142411222203)。

摘　　要：自动调制识别技术能够准确识别信号的调制类型,是信号处理领域中的一项关键技术。传统的识别方法在低信噪比下存在着识别准确率低的问题,并且当信号频率不稳定,或存在异步采样时,常规的识别方法性能将会出现恶化甚至失效。基于此,在信噪比和延迟变化的信道条件下,针对低速异步采样信号,研究基于深度学习的调制方式识别技术。首先对低速异步采样信号进行建模,利用其同相正交分量生成二维异步同相正交直方图,然后通过径向基函数神经网络提取该二维图像的特征参数,完成对输入信号的调制方式识别,最后经过大量的计算机仿真验证了该方法对7种调制方式的识别准确率。实验结果表明,在受到加性高斯白噪声的信道模型下,在低速异步采样的输入信号信噪比为6 dB时,可达到约95%以上的平均识别准确率,并通过对比实验验证了所提方案的有效性和鲁棒性。Automatic modulation recognition technology accurately identifies the modulation type of signals,making it a key technology in the field of signal processing.Traditional recognition methods suffer from low accuracy at low signal-to-noise ratios,and performance degradation or failure when dealing with signal frequency instability or asynchronous sampling.In this paper,we investigate modulation recognition technology based on deep learning for low-speed asynchronous sampled signals under channel conditions with varying signal-to-noise ratios and delays.We start by modeling low-speed asynchronous sampled signals and generating a two-dimensional asynchronous in-phase quadrature histogram using their in-phase and quadrature components.Subsequently,we employ a Radial Basis Function Neural Network to extract feature parameters from this two-dimensional image,thus achieving modulation type recognition for the input signal.Extensive computer simulations validate the proposed method’s accuracy in recognizing seven modulation types under the influence of additive white Gaussian noise.Experimental results demonstrate that,in the presence of additive white Gaussian noise in the channel model and with an input signal-to-noise ratio of 6 dB,the average recognition accuracy can exceed 95%.Comparative experiments further verify the effectiveness and robustness of the proposed approach.

关 键 词：调制方式识别 二维异步同相正交直方图 深度学习 径向基神经网络 

分 类 号：TN911.6[电子电信—通信与信息系统]