Wireless network optimization has been becoming very challenging as the problem size and complexity increase tremendously, due to close couplings among network entities with heterogeneous service and resource requirements. By continuously interacting with the environment, DRL provides a mechanism for different network entities to build knowledge and make autonomous decisions to improve network performance. In this article, we first review typical DRL approaches and recent enhancements. We then discuss the applications of DRL for mobile edge computing (MEC), which can be used for user devices to offload computation workload to MEC servers. However, for the low-power user devices, for example, wireless sensors, MEC can be costly as data offloading also consumes high power in RF communications. To balance the energy consumption in local computation and data offloading, we propose a novel hybrid offloading model that exploits the complementary operations of active RF communications and low-power backscatter communications. To maximize the energy efficiency in MEC offloading, the DRL framework is customized to learn the optimal transmission scheduling and workload allocation in two communications technologies. Numerical results show that the hybrid offloading scheme can improve the energy efficiency over 20 percent compared to existing schemes.

中文翻译：

---

移动边缘计算中反向散射辅助数据卸载的深度强化学习


由于具有异构服务和资源需求的网络实体之间的紧密耦合，问题规模和复杂性急剧增加，无线网络优化变得非常具有挑战性。通过与环境不断交互，DRL为不同网络实体提供了一种构建知识并做出自主决策以提高网络性能的机制。在本文中，我们首先回顾典型的 DRL 方法和最近的增强功能。然后我们讨论 DRL 在移动边缘计算 (MEC) 中的应用，它可用于用户设备将计算工作负载卸载到 MEC 服务器。然而，对于低功耗用户设备（例如无线传感器），MEC 的成本可能很高，因为数据卸载也会在 RF 通信中消耗高功率。为了平衡本地计算和数据卸载中的能耗，我们提出了一种新颖的混合卸载模型，该模型利用有源射频通信和低功耗反向散射通信的互补操作。为了最大限度地提高 MEC 卸载的能源效率，DRL 框架经过定制，可以学习两种通信技术中的最佳传输调度和工作负载分配。数值结果表明，与现有方案相比，混合卸载方案可以提高能源效率20%以上。