On a Novel Deep-Learning-Based Intelligent Partially Overlapping Channel Assignment in SDN-IoT
On a Novel Deep-Learning-Based Intelligent Partially Overlapping Channel Assignment in SDN-IoT
复制标题
DOI：
10.1109/mcom.2018.1701227
复制
发表时间：
2018-09
影响因子：
11.2
通讯作者：
Fengxiao Tang;Bomin Mao;Z. Fadlullah;N. Kato
中科院分区：
文献类型：
--
作者：
Fengxiao Tang;Bomin Mao;Z. Fadlullah;N. Kato
最近，SDN已经成为一种很有前途的技术，可以经济高效地提供物联网服务所需的规模和灵活性。在本文中，我们考虑面向物联网的无线SDN，即SDN-IoT，它有望智能地路由流量，并利用未充分利用的网络资源将物联网数据传输到云/互联网。然而，物联网设备的快速增加和随之而来的物联网数据流量的海量激增预计将给SDN-IoT带来巨大的压力。在本文中，我们将重点关注这一问题，并指出为每台SDN-IoT交换机分配合适的信道以避免潜在的网络拥塞的重要性。特别是，我们考虑了如何利用SDN-IoT中的PoC分配。然而，我们的研究表明，传统的固定PoC分配算法不适用于高动态的大规模SDN-IoT。因此，本文针对物联网数据流量动态变化的无线SDN-IoT，提出了一种新的基于深度学习的智能PoC分配方法。特别是，我们设想了两种基于深度学习的策略来预测未来的物联网流量负载和根据预测的流量负载自适应地分配PoC。基于计算机的仿真结果表明，通过在SDN-IoT控制器上实施所设想的深度学习方法，我们的方案实现了较高的流量负荷预测精度和信道分配过程的快速收敛。此外，与传统的PoC分配算法相比，我们的方案显著提高了网络性能。
Recently, SDN has emerged as a promising technology to cost-effectively provide the scale and flexibility necessary for IoT services. In this article, we consider the wireless SDN for IoT, referred to as SDN-IoT, which is anticipated to smartly route traffic and use underutilized network resources to deliver IoT data to the cloud/ Internet. However, the rapid increase of IoT devices and the subsequent massive surge of the IoT data traffic are expected to place a huge strain on the SDN-IoT. In this article, we focus on this issue and point out the importance of assigning suitable channels to each SDN-IoT switch to avoid potential network congestion. In particular, we consider how to exploit POC assignment in the SDN-IoT. However, our investigation reveals that the conventional fixed POC assignment algorithms are not viable for the highly dynamic large-scale SDN-IoT. Therefore, in this article, we propose a novel deep-learning-based intelligent POC assignment for the wireless SDN-IoT where the IoT data traffic dynamically changes. In particular, we envision two deep-learning-based strategies to predict the future IoT traffic load and to adaptively assign POCs according to predicted traffic load, respectively. Computer-based simulation results demonstrate that with the envisioned deep learning methods carried out at the SDN-IoT controller, our proposal achieves high accuracy of traffic load prediction and quick convergence of the channel assignment process. Additionally, in contrast with the conventional POC assignment algorithms, our proposal significantly improves the network performance.