Mobile edge computing (MEC) supporting localized context awareness creates a new technological frontier for 5G and beyond. Due to very asymmetric traffic related to MEC and the time division duplexing (TDD) system, we efficiently exploit the networking and computing functionalities for TDD orthogonal frequency division multiple access (TDD-OFDMA) technology supporting multiple services. The primary technical challenge of TDD-OFDMA systems lies in dynamic configuring based on the unknown characteristics of future traffic, i.e., the information lag. Therefore, a model-free online TDD configuration scheme is proposed based on context analysis and multi-armed bandit (MAB) optimization. The characteristics of future traffic are predicted by the context-aware MEC computing, so that TDD configuration is novelly modeled as a contextual MAB problem. Solving MAB by the contextual upper-confidence-bound, TDD configuration can be dynamically adjusted according to network traffic. To simultaneously reduce the energy consumption and makespan of mobile devices (MDs), a greedy resource allocation (GRA) embedded in the TDD configuration is further developed to select MDs and allocate resources. GRA algorithm decomposes the complex multi-factor coupling non-convex problem into a series of convex sub-problems, thereby asymptotically obtaining the selection and allocation with polynomial time complexity. Simulations justify significant performance gain in mobile networking and MEC.

中文翻译：

---

用于移动边缘计算的上下文感知 TDD 配置和资源分配

支持本地化上下文感知的移动边缘计算 (MEC) 为 5G 及其他领域创造了新的技术前沿。由于与 MEC 和时分双工 (TDD) 系统相关的非常不对称的流量，我们有效地利用了支持多业务的 TDD 正交频分多址 (TDD-OFDMA) 技术的网络和计算功能。TDD-OFDMA系统的主要技术挑战在于基于未来流量的未知特征，即信息滞后的动态配置。因此，提出了一种基于上下文分析和多臂老虎机（MAB）优化的无模型在线TDD配置方案。未来流量的特征是通过上下文感知 MEC 计算来预测的，因此 TDD 配置被新颖地建模为上下文 MAB 问题。通过上下文置信度上限解决 MAB，可以根据网络流量动态调整 TDD 配置。为了同时降低移动设备 (MD) 的能耗和完工时间，进一步开发了嵌入在 TDD 配置中的贪婪资源分配 (GRA) 来选择 MD 和分配资源。GRA算法将复杂的多因素耦合非凸问题分解为一系列凸子问题，从而渐近获得多项式时间复杂度的选择和分配。模拟证明了移动网络和 MEC 的显着性能提升是合理的。嵌入在 TDD 配置中的贪婪资源分配 (GRA) 被进一步开发以选择 MD 和分配资源。GRA算法将复杂的多因素耦合非凸问题分解为一系列凸子问题，从而渐近获得多项式时间复杂度的选择和分配。模拟证明了移动网络和 MEC 的显着性能提升是合理的。嵌入在 TDD 配置中的贪婪资源分配 (GRA) 被进一步开发以选择 MD 和分配资源。GRA算法将复杂的多因素耦合非凸问题分解为一系列凸子问题，从而渐近获得多项式时间复杂度的选择和分配。模拟证明了移动网络和 MEC 的显着性能提升是合理的。