射频供电物联网时延保障与性能增强机理研究

The contradiction between the energy limitations of Internet of Things (IoT) devices and the demand for low-latency services of tasks is a major challenge in the future Internet of Everything. RF-powered technology combined with edge computing technology provides the possibility to solve the above contradiction. This project focuses on the RF-powered IoT. From the three dimensions of computing, energy and communication, the task latency guarantee and enhancement technology are studied in the devices, charging points and the access points respectively. The research contents and objectives are as follow: Firstly, the theory of latency guarantee is developed. The algorithms of energy allocations and task migration are then proposed with objective to enhance the computation resources for the tasks. Secondly, in order to improve the RF-powered efficiency, the energy beamforming based on the cooperation of charging point and intelligent reflecting surface and the energy transform scheme in the devices are both designed. Finally, for the purpose of achieving network tradeoff between the number of access devices and the energy efficiency, the device scheduling and radio resource allocation schemes are studied under the latency constraints. The research results of this project will provide scientific guidance to ensure long-term and stable low latency demand for massive tasks in future IoT.

物联网设备能量受限与任务低时延服务需求之间的矛盾是实现万物互联过程中的一大挑战，射频充电技术结合边缘计算技术为上述矛盾的解决提供了可能性。本项目以射频供电物联网为研究对象，从计算-能量-通信三个维度出发，分别在设备、充电节点以及边缘接入点研究任务的时延保障和增强技术，具体研究内容和目标为：1) 研究基于网络演算的时延保障理论，提出基于设备端的能量分配和任务迁移算法，增强任务获得的计算资源；2) 研究充电节点联合智能反射面的能量波束设计方案和设备端的能量转化策略，增强射频能量的采集效率；3) 研究面向时延保障的、基于网络设备接入量与能效之间均衡的设备调度和无线资源分配策略，增强网络的通信能力。项目的研究成果将为未来物联网长时稳定地保障海量任务低时延需求提供科学指导。

本项目聚焦于射频供电物联网时延保障与性能增强机理，通过对端边协同任务时延保障理论、高能效边缘计算方法、边缘服务增强方法等多个方面开展深入研究，项目组在以下方面取得了重要成果：（1）考虑任务到达的突发性和无线信道状态的时变性，构建基于嵌入式马尔可夫理论的任务卸载排队模型，利用网络演算理论推导任务卸载和任务计算的时延违约概率，提出面向任务时延保障的端边协同机制，提升任务完成可靠性。（2）考虑物联网任务时延需求的差异性，构建了面向能耗优化的任务卸载模型，提出了卸载功率、计算频率、任务卸载比例联合优化策略，通过自适应调整节点的数据采集频率、能量采集和数据发送间隔，有效降低射频供电物联网总体能耗。（3）考虑用户兴趣偏好的动态性和无线资源的有限性，提出了群组兴趣感知的边缘缓存方法和面向时延保障的内容分发机制，针对物联网多边协同过程中由于弱链路导致分布式学习性能差的问题，提出了自适应联邦学习方法，提升了网络资源利用率。在项目执行期内，项目组在IEEE Transactions on Multimedia、IEEE Internet of Things Journal、通信学报、IEEE Globecom等高水平期刊和会议上共发表论文12篇，授权国家发明专利5项，获得IEEE WCSP最佳论文奖、IEEE Blockchain最佳论文奖、EAI MOBIMEDIA最佳论文奖。