Beam Configuration for Millimeter Wave Communication Systems Based on Distributed Federated Learning
-
摘要: 针对超密集组网中毫米波通信系统复杂的波束配置问题,该文提出一种基于分布式联邦学习(DFL)的波束配置算法(BMDFL),旨在利用有限的波束资源实现用户覆盖率最大化。考虑到传统集中式学习存在用户数据安全问题,基于分布式联邦学习框架构建系统模型,从而减少用户隐私信息的泄露。为了实现波束的智能化配置,引入双深度Q学习算法(DDQN)训练系统模型,并通过马尔可夫决策过程将长期的动态优化问题转化为相应的数学模型进行求解。仿真结果从系统的网络吞吐量和用户覆盖率方面验证了该方法的有效性和鲁棒性。Abstract: Considering the complex beam configuration problem of ultra-dense millimeter wave communication systems, a Beam management Method based on Distributed Federation Learning (BMDFL) is proposed to maximize the beam coverage by using the limited beam resources. To solve the problem of user data security in traditional centralized learning, the system model is constructed based on DFL, which can reduce the leakage of user privacy information. In order to realize intelligent configuration of beams, Double Deep Q-Network (DDQN) is introduced to train the system model, and the long-term dynamic optimization problem is transformed into the corresponding mathematical model through the Markov decision process. Simulation results demonstrate the effectiveness and robustness of the proposed method in terms of network throughput and user coverage.
-
算法1 基于分布式联邦的毫米波通信系统波束配置算法 输入:$ \mathcal{U} $,$ \mathcal{M} $, ${\varPhi _m}$,$ \kappa $, $ {S_t} $, $ {A_t} $,$ \gamma $, $ \rho $,$ \lambda $, $ \tau $ 输出:波束配置策略$ \mathcal{C}{\text{(}}t{\text{)}} $ (1) 初始化经验回访池,初始化全局模型 (2) for j = 1 : 1 : J do (3) $\triangleright $模型初始化 (4) 根据式(4)初始化局部模型 (5) $\triangleright $数据筛选 (6) 通过${d_{u,m} } \le {\varPhi _m}$和 $ {{{k_{u,m}}} \mathord{\left/ {\vphantom {{{k_{u,m}}} {{k_{\mathcal{U},m}}}}} \right. } {{k_{\mathcal{U},m}}}} $筛选有效的用户样本参与
模型训练(7) $\triangleright $局部模型训练 (8) for t = 1 : 1 : $ \tau $ do (9) 通过贪心算法求解$ {A_t} = \arg {\max _{{A_t}}}Q({S_t},{A_t}) $ (10) 执行$ {A_t} $得到$ {\mathcal{R}_t} $,$ {S_{t + 1}} $ (11) 将$ ({S_t},{A_t},{\mathcal{R}_t},{S_{t + 1}}) $存储到经验回访池 (12) 从经验池中随机抽取样本进行训练 (13) 根据式(5)计算$Y_t^{{\rm{DDQN}}}$ (14) 根据式(7)更新$ {\vartheta _{t,m}} $ (15) end for (16) $\triangleright $临时中心控制节点选择 (17) 通过${T_m}{\text{ = } }\min\left( { {T_1},{T_2}, \cdots ,{T_M} } \right)$选择执行时间最短的
mBS m(18) $\triangleright $全局模型更新 (19) 根据式(9)更新全局模型 (20) end for (21)得到全局最优的波束配置策略$ \mathcal{C}(t) $ 表 1 神经网络结构设置
层数 结构 1 nn.Sequential(nn.Linear($ \left| \mathcal{U} \right| $+2*$ {b_m} $, 40), nn.ReLU()) 2 nn.Sequential(nn.Linear(40,60), nn.ReLU()) 3 nn.Sequential(nn.Linear(60,40), nn.ReLU()) 4 nn.Sequential(nn.Linear(40,1)) -
[1] XU Yongjun, XIE Hao, WU Qingqing, et al. Robust max-min energy efficiency for RIS-aided HetNets with distortion noises[J]. IEEE Transactions on Communications, 2022, 70(2): 1457–1471. doi: 10.1109/TCOMM.2022.3141798 [2] XUE Qing, FANG Xuming, XIAO Ming, et al. Beam management for millimeter-wave beamspace MU-MIMO systems[J]. IEEE Transactions on Communications, 2019, 67(1): 205–217. doi: 10.1109/TCOMM.2018.2867487 [3] 闫莉, 方旭明, 李毅, 等. 面向高铁毫米波通信智能资源管理研究综述[J]. 电子与信息学报, 2023, 45(8): 2806–2817.YAN Li, FANG Xuming, LI Yi, et al. Overview on intelligent wireless resource management of millimeter wave communications under high-speed railway[J]. Journal of Electronics & Information Technology, 2023, 45(8): 2806–2817. [4] GIORDANI M, POLESE M, ROY A, et al. A tutorial on beam management for 3GPP NR at mmWave frequencies[J]. IEEE Communications Surveys & Tutorials, 2019, 21(1): 173–196. doi: 10.1109/COMST.2018.2869411 [5] 梁应敞, 谭俊杰, NIYATO D. 智能无线通信技术研究概况[J]. 通信学报, 2020, 41(7): 1–17. doi: 10.11959/j.issn.1000-436x.2020145LIANG Yingchang, TAN Junjie, and NIYATO D. Overview on intelligent wireless communication technology[J]. Journal on Communications, 2020, 41(7): 1–17. doi: 10.11959/j.issn.1000-436x.2020145 [6] XUE Qing, SUN Yao, WANG Jian, et al. User-centric association in ultra-dense mmWave networks via deep reinforcement learning[J]. IEEE Communications Letters, 2021, 25(11): 3594–3598. doi: 10.1109/LCOMM.2021.3108013 [7] LI Lixin, REN Huan, CHENG Qianqian, et al. Millimeter-wave networking in the sky: A machine learning and mean field game approach for joint beamforming and beam-steering[J]. IEEE Transactions on Wireless Communications, 2020, 19(10): 6393–6408. doi: 10.1109/TWC.2020.3003284 [8] ZHOU Yuhao, YE Qing, and LV Jiancheng. Communication-efficient federated learning with compensated overlap-FedAvg[J]. IEEE Transactions on Parallel and Distributed Systems, 2022, 33(1): 192–205. doi: 10.1109/TPDS.2021.3090331 [9] XUE Qing, LIU Yijing, SUN Yao, et al. Beam management in ultra-dense mmWave network via federated reinforcement learning: An intelligent and secure approach[J]. IEEE Transactions on Cognitive Communications and Networking, 2023, 9(1): 185–197. doi: 10.1109/TCCN.2022.3215527 [10] CHEN Mingzhe, POOR H V, SAAD W, et al. Wireless communications for collaborative federated learning[J]. IEEE Communications Magazine, 2020, 58(12): 48–54. doi: 10.1109/MCOM.001.2000397 [11] KHAN L U, SAAD W, ZHU Han, et al. Dispersed federated learning: Vision, taxonomy, and future directions[J]. IEEE Wireless Communications, 2021, 28(5): 192–198. doi: 10.1109/MWC.011.2100003 [12] LIAO Xiaomin, SHI Jia, LI Zan, et al. A model-driven deep reinforcement learning heuristic algorithm for resource allocation in ultra-dense cellular networks[J]. IEEE Transactions on Vehicular Technology, 2020, 69(1): 983–997. doi: 10.1109/TVT.2019.2954538 [13] BUSARI S A, MUMTAZ S, HUQ K M S, et al. System-level performance evaluation for 5G mmWave cellular network[C]. GLOBECOM 2017 - 2017 IEEE Global Communications Conference, Singapore, 2017: 1–7.