基于稀疏恢复的虚拟网络底层重构及高阶信息感知研究

While the combination of the information transmission and processing technology and virtualization greatly improve the network performance, resource discovery and monitoring for large-scale substrate networks is still difficult to implement in the virtual network environment. On the other hand, the dimension growth of status and data in the network gives rise to the obstacle of information processing, especially in the case that the measurement set is incomplete. To overcome these disadvantages, this project develops sparse recovery theory and approaches in the virtual network environment. By using the incomplete physical-logical mixed data set, network estimation model and substrate network reconstruction algorithms based on the network deconvolution are proposed. Meanwhile, the measurement operator is constructed to adapt tensor measuring under the constraints of virtual networks structure. With the help of the operator, tensor sensing mechanisms are built for massive data recovery. Furthermore, the project focuses on the structural characteristics of the substrate networks and high-order data tensors. In order to improve sensing efficiency, the distributed high-dimensional information reconstruction algorithms are designed, and the corresponding virtual networks are built for data interaction. The main principle of the design stems from high-order data tensor decomposition and division. The research results of the project will be significant for efficient managements of resources, and establish the theoretical and technological foundation for data computations of large-scale networks in the context of network virtualization.

传输—处理结合的信息技术和网络虚拟化在极大提高网络性能的同时，也面临着大规模底层网络物理资源状态推断及监测困难、网络状态信息以及承载数据的高维化、高维网络信息测量的不完备等问题。围绕以上问题，本项目拟通过发展虚拟网络环境下稀疏恢复基本理论及方法，在非完备物理—逻辑混合信息限制下，提出基于网络解卷积的复杂底层网络重构模型和算法；同时，在构造适合于虚拟网络结构的测量算子的基础上，发展网络高阶张量感知机制。进一步地，本项目将针对超大规模底层网络和数据张量的结构特性，设计高效的分布式虚拟网络高维信息重构算法，并构建用于数据交互的虚拟网络结构。本项目的成果将对大规模网络的资源高效管理具有重要意义，为未来虚拟化背景下大规模网络数据计算建立相关的理论与技术基础。

基于虚拟化、集中/协作式信号处理、实时云计算等技术的新兴网络架构极大提高网络性能的同时，也存在大规模网络物理资源状态推断及监测困难、网络结构以及承载数据的高维化等问题以影响网络性能。本课题针对以上关键技术展开研究，提出基于稀疏恢复的虚拟网络结构重构以及网络环境中的多类型高阶信息处理、资源调度分配方案。本课题着眼于雾/云无线接入网络中存储内容、接入点和终端之间的虚拟关系，通过网络虚拟结构映射的动态预测和资源分配方法发挥该网络架构的优势，有效提升多媒体业务的服务质量和能量效率。与此同时，针对网络映射及网络承载内容，提出基于范数/伪范数差分结构广义稀疏信号感知理论及迭代重加权框架下的非凸优化方法，并且以此为基础发展了其在高维张量感知，矩阵填补、去噪修复、目标聚类识别等方面的应用，探索所提算法在工程中应用的前景。研究结果表明，本项目所提理论框架和算法优于现存主要方法。基于上述研究内容，本项目发表学术论文12篇（其中SCI检索的国际期刊论文9篇，EI检索学术论文3篇）并搭建相关演示系统。