GFDM中基于高阶长短时记忆神经网络的自适应均衡器  

An adaptive equalizer based on high order LSTM in GFDM

作　　者：牛安东 苗硕 刘佳宁 李英善[1,2] Niu Andong;Miao Shuo;Liu Jianing;Li Yingshan(College of Electronic Information and Optical Engineering,Nankai University,Tianjin 300350,China;Tianjin Key Laboratory of Optoelectronic Sensor and Sensing Network Technology,Nankai University,Tianjin 300350,China)

机构地区：[1]南开大学电子信息与光学工程学院,天津300350 [2]南开大学光电传感器与传感网络技术重点实验室,天津300350

出　　处：《电子技术应用》2022年第8期95-100,共6页Application of Electronic Technique

基　　金：国家自然科学基金项目(62171239)。

摘　　要：在广义频分复用系统(GFDM)中,为解决5G网络下车载移动通信在Sub-6 GHz频段信道中信号严重失真的问题,提出一种基于高阶长短时记忆神经网络(HO-LSTM)结构的自适应均衡器。HO-LSTM自适应均衡器在传统高阶前馈神经网络(HO-FNN)的基础上,采用复杂度更低的广义记忆多项式模型(GMP)代替Volterra模型,并引入LSTM神经网络使其更适用于复杂非线性模型的预测。结果表明,相比于传统HO-FNN均衡器和LSTM均衡器,所提出的HO-LSTM均衡器的均衡效果显著提升,系统性能也得到进一步改善。In the generalized frequency division multiplexing system(GFDM),in order to solve the problem of severe signal distortion in the sub-6 GHz frequency band channel of the vehicle-mounted mobile communication under the 5G network,an adaptive equalizer based on high order long short-term memory(HO-LSTM)neural network structure is proposed.Based on the traditional high-order feedforward neural network(HO-FNN),HO-LSTM adaptive equalizer uses the generalized memory polynomial model(GMP)with lower complexity instead of Volterra model,and introduces LSTM neural network to make it more suitable for the prediction of complex nonlinear models.The results show that,compared with the traditional HO-FNN equalizer and LSTM equalizer,the equalization effect of the proposed HO-LSTM equalizer is significantly improved,and the system performance is further improved.

关 键 词：广义频分复用技术 长短时记忆神经网络 高阶神经网络 广义记忆多项式 

分 类 号：TN911.7[电子电信—通信与信息系统]