摘要
针对毫米波Massive MIMO系统中基于导频的信道估计精度较差和导频开销过大问题,提出基于DNN的两阶段信道估计方案。第一阶段通过训练数据对网络模型进行训练,在误差阈值范围内得到网络最优效果;第二阶段通过测试和验证数据进行信道参数估计,在此过程中采用Dropout策略防止过拟合。仿真结果表明,与LS、MMSE、OMP算法和LSTM相比,相同导频数量下,所提出的DNN方法信道估计精度平均提升约1.58 dB;不同导频数量下,使用较少导频数量的DNN方法估计精度仍优于使用较多导频数量的LS、MMSE、OMP和LSTM算法,同一估计精度下,DNN方法平均降低导频开销约32.4%.
In order to solve the problems of poor pilot-based channel estimation accuracy and excessive pilot overhead in a millimeter-wave massive MIMO system,a two-stage channel estimation scheme based on Deep Neural Network(DNN)is proposed.In the first stage,the network model is trained through data training to obtain the optimal one within the error threshold.In the second stage,the channel parameters are estimated through test and verification data,where the dropout strategy is adopted to prevent overfitting.Simulation results show that compared with the Least Squares(LS),Minimum Mean Square Error(MMSE),Orthogonal Matching Pursuit(OMP)and Long Short-Term Memory(LSTM)algorithms,the proposed DNN method averagely improves the channel estimation accuracy by about 1.58dB with the same number of pilots.Under different numbers of pilots,the accuracy of the DNN method with fewer pilots is still better than the LS,MMSE,OMP and LSTM with more pilots.Under the same estimation accuracy,the DNN method reduces pilot overhead by about 32.4%on average.
作者
马珊珊
刘紫燕
MA Shanshan;LIU Ziyan(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)
出处
《移动通信》
2020年第10期43-48,共6页
Mobile Communications
基金
国家自然科学基金资助项目(61863006)
贵州省科学技术基金资助项目(黔科合基础[2016]1054)
贵州省联合资金资助项目(黔科合LH字[2017]7226)
贵州省科技计划重点项目([2019]1416)
贵州大学2017年度学术新苗培养及创新探索专项(黔科合平台人才[2017]5788)。
