针对在采用massive MIMO（multiple-input multiple-output）系统的5G网络规划中，使用传统的系统级仿真方法获得信道幅值的计算量和时间开销非常大的问题，提出一种基于BP（back propagation）神经网络的自适应神经网络来预测massive MIMO系统的信道幅值。自适应神经网络由基本BP子神经网络和特征降维BP子神经网络组成，可实现对给定训练集和预测集的自适应，基于用户射线追踪数据快速准确地预测用户的信道幅值。仿真结果表明，所提出的自适应神经网络在得到与系统级仿真方法精度接近的信道幅值的同时，可大幅降低获得信道幅值的时间开销；并且与采用传统BP神经网络相比，可以明显降低训练时间、预测误差大的用户数和平均预测误差。
Aiming at the problem that the computation and time overhead of obtaining channel amplitude is very large when implementing network planning of 5G massive MIMO (multiple-input multiple-output) networks with the traditional system-level simulation methods, this paper proposes a BP (back propagation) network based adaptive neural network to predict the channel amplitude of massive MIMO systems. The adaptive neural network consists of a basic BP sub neural network and a feature-reduced BP sub neural network, which can realize self-adaption to the given training and test set and can quickly and accurately predict the user channel amplitude based on the ray-tracing data. Simulation results show that the proposed adaptive neural network can achieve close accuracy to the traditional system-level simulation methods when obtaining channel amplitude but with significantly reduced time overhead and can obviously reduce the training time, the amount of users with large prediction error and the average prediction error compared with the traditional BP neural network.
[1] 尤肖虎, 潘志文, 高西奇, 等. 5G移动通信发展趋势与若干关键技术[J].中国科学:信息科学, 2014, 44(5):551-563.
[2] Mehmood Y, Afzal W, Ahmad F, et al. Large scaled multi-user MIMO system so called massive MIMO systems for future wireless communication networks[C]//2013 19th International Conference on Automation and Computing. September 13-14,2013, London, UK, IEEE, 2013:1-4.
[3] 邓瑞琛, 姜之源, 刘景初, 等.利用信道学习获取超蜂窝网络休眠基站的信道信息[J]. 中国科学:信息科学, 2017, 47(11):1583-1591.
[4] 叶新荣, 张爱清, 麻金继. 一种仿真瑞利信道的有效方法[J]. 中国科学院研究生院学报, 2010, 27(2):228-233.
[5] 郑建涛. 基于MATLAB的LTE系统级仿真研究[D]. 北京:北京邮电大学, 2013.
[6] Wang Y, Xu J, Jiang L S. Challenges of system-level simulations and performance evaluation for 5G wireless networks[J]. IEEE Access, 2014, 2:1553-1561.
[7] Kitao K, Benjebbour A, Imai T, et al. Development of 5G system evaluation tool[C]//2018 Asia-Pacific Microwave Conference(APMC). November 6-9, 2018, Kyoto, Japan. IEEE, 2018:675-677.
[8] 李凯, 徐景, 杨旸. 5G环境下系统级仿真建模与关键技术评估[J]. 中兴通讯技术, 2016, 22(3):41-46.
[9] Liao Y, Yao H M, Hua Y X, et al. CSI feedback based on deep learning for massive MIMO systems[J]. IEEE Access, 2019, 7:86810-86820.
[10] Wen C K, Shih W T, Jin S. Deep learning for massive MIMO CSI feedback[J]. IEEE Wireless Communications Letters, 2018, 7(5):748-751.
[11] Chun C, Kang J, Kim I. Deep learning based channel estimation for massive MIMO systems[J]. IEEE Wireless Communications Letters, 2019, 8(4):1228-1231.
[12] Huang H J, Yang J, Huang H, et al. Deep learning for super-resolution channel estimation and DOA estimation based massive MIMO system[J]. IEEE Transactions on Vehicular Technology, 2018, 67(9):8549-8560.
[13] Dong P H, Zhang H, Li G Y, et al. Deep CNN-based channel estimation for mmWave massive MIMO systems[J]. IEEE Journal of Selected Topics in Signal Processing, 2019, 13(5):989-1000.
[14] Strouse D, Schwab D J. The deterministic information bottleneck[J]. Neural Computation, 2017, 29(6):1611-1630.
[15] 3GPP TR 38.901. Study on channel model for frequencies from 0.5 to 100 GHz (Release 15)[R/OL]. (2018-06-29)[2019-12-01].https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=3173.
[16] Tishby N, Pereira F C, Bialek W. The information bottleneck method[J]. University of Illinois, 2000, 411(29-30):368-377.
[17] Tishby N, Zaslavsky N. Deep learning and the information bottleneck principle[C]//2015 IEEE Information Theory Workshop (ITW). April 26-May 1, 2015, Jerusalem, Israel, IEEE, 2015:1-5.
