基于多任务学习的MIMO-OFDM信噪比估计与调制识别  

MIMO-OFDM Signal to Noise Ratio Estimation and Modulation Recognition Based on Multi Task Learning

作　　者：张天骐[1] 汪锐 安泽亮 王雪怡 方竹 ZHANG Tianqi;WANG Rui;AN Zeliang;WANG Xueyi;FANG Zhu(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065

出　　处：《北京邮电大学学报》2022年第6期95-100,121,共7页Journal of Beijing University of Posts and Telecommunications

基　　金：国家自然科学基金项目(61771085);重庆市自然科学基金项目(cstc2021jcyj-msxmX0836)。

摘　　要：非协作通信中多输入多输出正交频分多路复用(MIMO-OFDM)信号信噪比盲估计和子载波的调制识别研究仅针对单一任务。对此,提出了一种将深度神经网络与多任务学习模型相结合,同时完成信噪比盲估计与子载波调制识别的算法。首先利用特征值矩阵联合近似对角化算法恢复发送信号,并提取恢复信号的同向正交分量作为浅层特征;然后搭建基于一维卷积神经网络的多任务学习模型,通过联合训练信噪比估计和子载波调制识别2个任务,实现优势互补。仿真结果表明,所提算法可获得比单任务学习模型更优的性能;当信噪比为-10 dB时,信噪比估计的均方误差降低了66.21%,子载波调制识别的精度提高了4.75%。The research on blind estimation of signal-to-noise ratio and modulation recognition of subcarriers in multi input multi output orthogonal frequency division multiplexing(MIMO-OFDM) signals in non cooperative communication is only aimed at a single task. For this reason, an algorithm combining deep neural network with multi task learning model is proposed to complete blind estimation of signal-to-noise ratio and modulation recognition of subcarriers at the same time. Firstly, the transmitted signal is recovered using the joint approximate diagonalization(JADE) algorithm of the eigenvalue matrix, and the co orthogonal components of the recovered signal are extracted as shallow features;Then, a multi-task learning model based on one-dimensional convolutional neural network is built. Through joint training of signal-to-noise ratio estimation and subcarrier modulation recognition, the advantages are complementary. Simulation results show that the proposed algorithm can achieve better performance than the single task learning model;When the signal-to-noise ratio is-10 dB, the mean square error of signal-to-noise ratio estimation is reduced by 66.21%, and the accuracy of subcarrier modulation recognition is improved by 4.75%.

关 键 词：多输入多输出信号 信噪比估计 调制识别 神经网络 多任务学习 

分 类 号：TN911.7[电子电信—通信与信息系统]