Online reliability optimization for URLLC in HetNets: a DQN approach
Online reliability optimization for URLLC in HetNets: a DQN approach
复制标题
HetNets 中 URLLC 的在线可靠性优化：一种 DQN 方法
DOI：
10.1007/s00521-020-05492-4
复制
发表时间：
2020-11
影响因子：
6
通讯作者：
Wang Xingwei
中科院分区：
文献类型：
--
作者：
Yang Leyou;Jia Jie;Chen Jian;Wang Xingwei
异质蜂窝网络(HetNets)已被证明是处理不断增长的数据流量的一种很有前途的方法。支持超可靠和低延迟通信(URLLC)也被认为是即将到来的无线网络的一个新功能。由于HetNet的重叠结构和小区间的相互干扰，现有的资源分配方法不能直接应用于实时应用，特别是对于URLLC业务。作为一种新的无监督算法，深度Q网络(DQN)已经成功地应用于许多在线复杂优化模型。然而，由于HetNet中状态变化微小、动作空间规模大的特点，在HetNet中进行资源分配优化时可能会表现不佳。为了应对这些问题，我们首先提出了一种自动编码器来干扰相邻状态的相似性来增强特征，然后将整个决策过程分成两个阶段。每个阶段分别用DQN进行求解，迭代整个过程以求出联合最优解。我们在具有不同数量的用户设备(UE)、冗余链路和子载波的6个场景中实现了我们的算法。仿真结果表明，该算法对优化目标具有良好的收敛性能。此外，通过进一步优化功率分配，在恶劣条件下，可靠性提高了1-2个9。最后，实验结果表明，该算法在常见场景下达到了8-9的可靠性。作为一种在线方法，本文提出的算法平均只需要S的0.32%。
Heterogeneous cellular networks (HetNets) have been proven as a promising approach to deal with ever-growing data traffic. Supporting ultra-reliable and low-latency communication (URLLC) is also considered as a new feature of the upcoming wireless networks. Due to the overlapping structure and the mutual interference between cells in HetNets, existing resource allocation approaches cannot be directly applied for real-time applications, especially for URLLC services. As a novel unsupervised algorithm, Deep Q Network (DQN) has already been applied to many online complex optimization models successfully. However, it may perform badly for resource allocation optimization in HetNets, due to the tiny state change and the large-scale action space characteristics. In order to cope with them, we first propose an auto-encoder to disturb the similarity of adjacent states to enhance the features and then divide the whole decision process into two phases. DQN is applied to solve each phase, respectively, and we iterate the whole process to find the joint optimized solution. We implement our algorithm in 6 scenarios with different numbers of user equipment (UE), redundant links, and sub-carriers. Simulations results demonstrate that our algorithm has good convergence for the optimization objective. Moreover, by further optimizing the power allocation, a 1–2 nines of reliability improvement is obtained for bad conditions. Finally, the experiment result shows that our algorithm reaches the reliability of 8-nines in common scenarios. As an online method, the algorithm proposed in this paper takes only 0.32 s on average.
登录
查看更多内容
影响因子：
6.8
作者：
Chongtao Guo;Le Liang;Geoffrey Y. Li
通讯作者：
Chongtao Guo;Le Liang;Geoffrey Y. Li
DOI：
--
发表时间：
1991-07
期刊：
arXiv: Optics
影响因子：
--
作者：
S. Whitehead
通讯作者：
S. Whitehead
DOI：
10.1109/tnn.1998.712192
发表时间：
1998
期刊：
IEEE Trans. Neural Networks
影响因子：
--
作者：
R. S. Sutton;A. Barto
通讯作者：
R. S. Sutton;A. Barto
DOI：
10.1109/twc.2017.2770094
发表时间：
2014-09
期刊：
2014 52nd Annual Allerton Conference on Communication, Control, and Computing (Allerton)
影响因子：
--
作者：
R. Raman;K. Jagannathan
通讯作者：
R. Raman;K. Jagannathan
DOI：
10.1109/glocom.2016.7841870
发表时间：
2016-12
期刊：
2016 IEEE Global Communications Conference (GLOBECOM)
影响因子：
--
作者：
Jie Jia;Yansha Deng;Shuyu Ping;H. Aghvami;A. Nallanathan
通讯作者：
Jie Jia;Yansha Deng;Shuyu Ping;H. Aghvami;A. Nallanathan