门控递归单元神经网络坐标变换盲均衡算法  

Coordinate transformation blind equalization algorithm based on gated recurrent unit neural network

作　　者：魏海文 郭业才[1,2] WEI Hai-wen;GUO Ye-cai(Jiangsu Key Laboratory of Meteorological Observation and Information Processing,Nanjing University of Information Science & Technology,Nanjing 211800, China;Jiangsu Collaborative Innovation Center on Atmospheric Environment and Equipment Technolog,Nanjing 211800,China)

机构地区：[1]南京信息工程大学江苏省气象探测与信息处理重点实验室,江苏南京211800 [2]江苏省大气环境与装备技术协同创新中心,江苏南京211800

出　　处：《微电子学与计算机》2019年第9期89-93,98,共6页Microelectronics & Computer

基　　金：国家自然科学基金(61673222);江苏省高校自然科学研究重大项目(13KJA510001);江苏省六大人才高峰项目(2008026)

摘　　要：针对数字信号传输过程中的码间干扰问题,提出了门控递归单元神经网络坐标变换盲均衡算法(GRUNN-CT-CMA).首先,在递归神经网络基础上加入门控结构,使门控递归单元神经网络(GRUNN)对长时间跨度信息的感知能力更强、记忆力更持久.其次,在GRUNN中引入坐标变换得到的盲均衡算法,进一步降低了稳态误差、加快了代价函数收敛速度.结果表明,与常模盲均衡算法(CMA)和延迟单元递归神经网络盲均衡算法(BRNN-CMA)相比,GRUNN-CT-CMA在均衡高阶多模信号时,稳态误差最小、收敛速度最快、输出信号星座图最清晰.In order to solve the problem of inter symbol interference in the process of digital signal transmission, a coordinate transformation constant modulus blind equalization algorithm based on gated recurrent unit neural network(GRUNN-CT-CMA) is proposed. Firstly, based on the recurrent neural network, the gated recurrent unit neural network(GRUNN) with a gate structure was added, which has stronger perception and longer-lasting memory of long-span information. Secondly, the coordinate transformation blind equalization algorithm was introduced in GRUNN, which further reduced the residual error and corrected the phase offset. The simulation results show that comparing with the constant modulus blind equalization algorithm(CMA) and the bias-unit recurrent neural networkconstant modulus blind equalization algorithm(BRNN-CMA), when GRUNN-CT-CMA equalizing high order multi-mode signals, the steady-state error is minimal, the speed of convergence is the fastest and the constellation of the output signal is the clearest.

关 键 词：盲均衡 门控递归单元 神经网络 代价函数 坐标变换 码间干扰 

分 类 号：TN911.7[电子电信—通信与信息系统]