认知无线网络的联合推送与缓存机制研究

Cognitive wireless networks significantly improve the overall spectrum efficiency by dynamically sharing spectrum holes, thereby holding the promise of supporting emerging mobile traffics and applications. Unfortunately, since the idle spectrum access opportunities appear randomly and dispersedly, a cognitive wireless network usually encounters a large traffic delay and therefore cannot effectively assure the Quality-of-Service. In this project, we aim at adopting a joint pushing and caching mechanism to overcome the above drawback of cognitive wireless networks. By extending the conventional physical layer spectrum sensing to the statistical learning of the user demands for content items, a base station or access point may predict what a user is interested in, and then exploit opportunistic spectrum access to push these content items, which can be then read from the buffer when requested by the user. By this means, we enable the opportunistic spectrum access, where the data transmissions are not real-time, to support a real-time content service virtually. As a result, a substantial spectrum efficiency gain can be achieved without causing any damage to the Quality-of-Service. In this project, we shall investigate the optimal policies of single user pushing, multicasting pushing, and layered pushing for delay tolerant users, as well as the optimal tradeoff and interactive mechanisms between the communications resources, storage resources, fairness metrics and average delay. Our aims are not only to conceive the typical methods that can trade scalable storage and computation resources for non-scalable communications resources from a theoretical perspective, but also to enhance the Quality-of-Service of emerging spectrum sharing networks such as LTE-U systems, from a more practical perspective.

认知无线网络可动态共享时频空洞，提升频谱的整体利用率，在频谱分配殆尽时支持新涌现的移动业务和应用。然而，由于空闲频谱接入机会是随机的、分散的，甚至在繁忙时段是稀疏的，因此认知无线网络的业务传输往往具有较大的延时，难以保障服务质量。本项目探索用联合推送与缓存机制克服认知无线网络的上述缺陷，通过将物理层频谱感知推广到对用户内容需求的统计和学习，预测其可能感兴趣的内容，并利用机会频谱接入推送。当用户提出内容需求时，从其缓存区中读出即可。由此，可通过非实时的机会接入通信可以支持虚拟实时的服务，即提升了频谱利用率，又确保了内容的服务质量。本项目拟探索单用户推送、多用户组播推送和延时容忍用户分层推送的最优策略，及通信资源、存储资源、公平性和平均延时之间的最佳折中关系和作用机理，从理论层面阐明利用可扩展的存储和计算资源，换取不可扩展的通信资源的典型方法，从应用层面提升LTE－U等频谱共享网络的服务质量。

本项目探索了认知无线网络中的联合推送与缓存机制，旨在利用缓存单元，在用户提出请求之前的“负时间轴”上发起传输，从而让认知无线网络有效地利用负时间轴上的空闲时频资源，缓解频谱繁忙时段的业务压力。课题组按照项目申请书和计划书开展研究工作，完成了预期的研究指标，在认知无线网络的单用户联合推送和缓存、面向多用户的组播推送、延时容忍业务的分层与编码推送等三个研究点上取得了成果。具体的，提出了基于需求时间预测的推送调度策略，可在探知的空闲时频资源下提升缓存内容的命中率，获得更高的有效吞吐量；给出了推送的能效优化策略，可针对典型信道进行功率和速率的自适应，从而优化联合推送与缓存系统的能量效率；分析了认知无线网络采用联合推送与缓存策略的代价，通过逆向应用排队理论中的Little公式，给出了一种缓存代价的计算方法，由此优化了缓存内容的删除时间，给出了缓存内容的在线动态选择方法；探索了认知无线网络中采用联合推送与缓存的损益机理，提出了一种可以激励推送并提升频谱和能量效率的定价机制，分析了编码缓存中用户的增益分配，为其定价提供了一种合作博弈的基础；针对用户偏好的差异，提出了一种在边缘缓存中进行二次压缩的方法，给出了相应的无损和有损压缩策略。基于本项研究，课题组在IEEE JSAC、TCOM、TWC和WCM等期刊上发表论文50篇，在IEEE ICC和Globecom等会议上发表论文45篇，申请中国发明专利13项，美欧日等境外专利6项。在本项目的执行过程中，已毕业博士生6名、硕士生10名，其中1名博士生获清华大学优秀博士学位论文，2名硕士生分获北京市和清华大学优秀硕士毕业生。在本项目执行过程中，积极开展国内外学术合作交流，先后派遣4名博士生赴境外知名学者团队进行联合培养，参加本领域主要国际学术会议十余人次。