基于互信息量估计的几何与概率联合整形技术  

Joint Shaping of Geometry and Probability based on Mutual Information Neural Estimation

作　　者：梁家熙 牛泽坤 胡卫生[1] 义理林[1] LIANG Jia-xi;NIU Ze-kun;HU Wei-sheng;YI Li-lin(State Key Laboratory of Advanced Optical Communication System and Networks,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学区域光纤通信网与新型光通信系统国家重点实验室,上海200240

出　　处：《光通信研究》2022年第3期1-6,共6页Study on Optical Communications

基　　金：科技部重点研发计划资助项目(2018YFB1800904)。

摘　　要：针对正交振幅调制在高信噪比的加性高斯白噪声(AWGN)信道与香农极限有1.53 dB容量间隙的问题,文章提出了一种基于互信息量估计的几何与概率联合整形的方法,将几何整形与概率整形相结合,提升通信系统的互信息量。文章将互信息量估计作为计算系统互信息量的方式,以最大化互信息量为目的训练发端的编码器,实现几何整形与概率整形。通过在不同信噪比下AWGN信道中的仿真,验证了基于互信息量估计的几何与概率联合整形系统的性能要优于单独进行几何整形或概率整形的性能。在信噪比为10 dB的AWGN信道中,系统的互信息量与几何整形相比有0.0417 bit/symbol的增益,与概率整形相比有0.0279 bit/symbol的增益。In view of the problem that quadrature amplitude modulation has 1.53 dB capacity gap at the Shannon limit in high signal noise ratio,the paper proposes a joint shaping method of geometry and probability based on mutual information neural estimation.Geometric shaping and probability shaping are combined to improve the mutual information of the communication system.In this paper,the mutual information neural estimation is used to calculate the mutual information of the system,and the encoder of the transmitter is trained for the purpose of maximizing the mutual information,so as to realize geometric shaping and probability shaping.Through the simulation of Additive White Gaussian Noise(AWGN)channels with different signal noise ratios,it is proved that the performance of joint shaping of geometry and probability based on mutual information neural estimation is better than that of geometric shaping or probabilistic shaping separately.In AWGN channel with signal-to-noise ratio of 10 dB,the mutual information of the system has a gain of 0.0417 bit/symbol over geometric shaping,and a gain of 0.0279 bit/symbol compared over probability shaping.

关 键 词：深度学习 互信息量估计 物理层通信 

分 类 号：TN919.3[电子电信—通信与信息系统]