基于Bi-LSTM的同时同频全双工数字域自干扰抑制方法  

A CCFD digital domain self-interference suppression approach based on Bi-LSTM

作　　者：鲁帆[1,2] 范占春 马超[1,2] 陈远祥 汪予晗 程竟爽 杜海龙 胡聪 Fan LU;Zhanchun FAN;Chao MA;Yuanxiang CHEN;Yuhan WANG;Jingshuang CHENG;Hailong DU;Cong HU(Beijing Institute of Spacecraft System Engineering,Beijing 100094,China;Beijing Engineering Research Center of EMC and Antenna Measurement,Beijing 100094,China;School of Electronic Engineering,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京空间飞行器总体设计部,北京100094 [2]北京市电磁兼容与天线测试工程技术研究中心,北京100094 [3]北京邮电大学电子工程学院,北京100876

出　　处：《中国科学：信息科学》2023年第10期1982-1993,共12页Scientia Sinica(Informationis)

基　　金：北京市科技新星计划(批准号:Z211100002121138);国家自然科学基金(批准号:62271079,61875239);国家重点研发计划(批准号:2021YFB2900405)资助项目。

摘　　要：同时同频全双工(co-frequency co-time full duplex, CCFD)系统在相同的频率上同时进行信号的收发,理论上可使通信频谱利用率提高一倍.但是由于收发天线等前端模块距离较近,系统中会存在很强的自干扰信号.当前常用的自适应滤波、最小二乘法估计等自适应干扰抑制方法存在着不能有效抑制多径信道和功放非线性产生自干扰信号的不足.针对此问题,本文提出一种基于双向长短时记忆神经网络(bi-directional long short-term memory, Bi-LSTM)的CCFD数字域自干扰抑制方法.首先根据多径信道的特征,采用记忆多项式对自干扰信道进行建模;然后采用Wild Horse优化算法(Wild Horse optimizer, WHO),通过迭代寻找到最优时延单位以确定训练数据的特征数;最后搭建Bi-LSTM网络进行训练,重构出自干扰信号,并在接收端减去,以达到自干扰抑制的目的.在仿真实验中采用OFDM (orthogonal frequency division multiplexing)信号作为参考信号,实现了47.17 dB自干扰信号抑制比,较传统最小二乘(least square, LS)算法有31.58 dB的提升.结果表明,本文所提出的方法可高效准确地提高CCFD系统的自干扰信号抑制能力.Co-frequency co-time full duplex(CCFD)systems can transmit and receive signals simultaneously at the same frequencies and bring on double spectrum utilization,but CCFD systems suffer from self-interfering signals caused by the short distance between the Tx and Rx antennas.The commonly applied self-interference suppression methods,such as adaptive filtering and least-squares estimation,suffer from not suppressing multipath channels and self-interfering signals caused by non-linearity in the power amplifier at the same time.Based on this,we propose a CCFD digital domain self-interference suppression approach based on bi-directional long shortterm memory(Bi-LSTM).First,according to the multipath channel characters,we build a self-interfering channel model using memory polynomials.Then,we search for the optimal unit of delay through iteration to confirm the number of features,by using Wild Horse optimizer.Finally,we build the Bi-LSTM to train and reconfigure self-interfering signals and suppress self-interference by canceling these signals at the Rx ports.We use orthogonal frequency division multiplexing signals at simulations and achieve a 47.17 dB self-interference suppression ratio result,with a 31.58 dB improvement compared with a traditional least-squares algorithm.Simulation results indicate the effectiveness of this approach.

关 键 词：同时同频全双工(CCFD) 双向长短时记忆神经网络(Bi-LSTM) Wild Horse优化算法(WHO) OFDM 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TN975[自动化与计算机技术—控制科学与工程]