QCell: Self-optimization of Softwarized 5G Networks through Deep Q-learning
QCell: Self-optimization of Softwarized 5G Networks through Deep Q-learning
复制标题
DOI：
10.1109/globecom46510.2021.9685171
复制
发表时间：
2021-12
期刊：
影响因子：
--
通讯作者：
Bernardo Casasole;Leonardo Bonati;Salvatore D’oro;S. Basagni;A. Capone;T. Melodia
中科院分区：
文献类型：
--
作者：
Bernardo Casasole;Leonardo Bonati;Salvatore D’oro;S. Basagni;A. Capone;T. Melodia
随着业务需求和移动的用户的空前增长，实时细粒度优化框架对蜂窝网络的未来至关重要。事实上，刚性和不灵活的基础设施无法适应5G网络的大量数据预测。网络软件化，即通过软件控制“一切”的方法赋予网络前所未有的灵活性，允许它运行优化和基于机器学习的框架，以灵活地适应当前的网络条件和业务需求。这项工作提出了QCell，一个基于深度Q网络的软件化蜂窝网络优化框架。QCell动态地将切片和调度资源分配给网络基站，以适应变化的干扰条件和业务模式。QCell在世界上最大的网络模拟器Colosseum上进行原型开发，并在各种网络条件和场景下进行测试。我们的实验结果表明，使用QCell显着提高用户的吞吐量（高达37.6%）和传输队列的大小（高达11.9%），减少服务延迟。
With the unprecedented rise in traffic demand and mobile subscribers, real-time fine-grained optimization frame-works are crucial for the future of cellular networks. Indeed, rigid and inflexible infrastructures are incapable of adapting to the massive amounts of data forecast for 5G networks. Network softwarization, i.e., the approach of controlling “everything” via software, endows the network with unprecedented flexibility, al-lowing it to run optimization and machine learning-based frame-works for flexible adaptation to current network conditions and traffic demand. This work presents QCell, a Deep Q-Network-based optimization framework for softwarized cellular networks. QCell dynamically allocates slicing and scheduling resources to the network base stations adapting to varying interference con-ditions and traffic patterns. QCell is prototyped on Colosseum, the world's largest network emulator, and tested in a variety of network conditions and scenarios. Our experimental results show that using QCell significantly improves user's throughput (up to 37.6%) and the size of transmission queues (up to 11.9%), decreasing service latency.