CURT: A Real-Time Scheduling Algorithm for Coexistence of LTE and Wi-Fi in Unlicensed Spectrum
CURT: A Real-Time Scheduling Algorithm for Coexistence of LTE and Wi-Fi in Unlicensed Spectrum
复制标题
DOI：
10.1109/dyspan.2018.8610476
复制
发表时间：
2018-10
期刊：
影响因子：
--
通讯作者：
Yan Huang;Yongce Chen;Yiwei Thomas Hou;Wenjing Lou
中科院分区：
文献类型：
--
作者：
Yan Huang;Yongce Chen;Yiwei Thomas Hou;Wenjing Lou
载波侦听自适应传输（CSAT）是工业界解决未授权频带中LTE和Wi-Fi之间共存的主要方法。在CSAT下，一个关键问题是设计一个调度算法来分配无线资源在多个信道和大量的子信道。本文研究了这个调度问题，通过优化配方的目标是最大限度地减少LTE的不利影响的Wi-Fi用户。这是通过在信道和子信道级别上最优分配无线电资源以满足每个LTE用户的上行链路和下行链路速率要求来实现的。在LTE调度期间给出信道条件的特殊考虑。这里的一个主要挑战是在~1 ms的时间尺度上获得最优（或接近最优）的调度解决方案-对于该算法在该领域中是有用的，这是严格的定时要求。我们的主要贡献是开发的CURT，调度算法，可以获得接近最优的解决方案在~1毫秒。CURT利用了独特的结构的基本优化问题，并将其分解成大量的独立的子问题。这些子问题可以通过GPU多处理器并行高效地求解。通过在Nvidia GPU/CUDA平台上实现CURT，我们证明了CURT确实可以在约1 ms内提供接近最优的调度解决方案，并满足我们的所有设计目标。
Carrier-Sensing Adaptive Transmission (CSAT) is a major approach from industry to address coexistence between LTE and Wi-Fi in unlicensed bands. Under CSAT, a key problem is the design of a scheduling algorithm to allocate radio resources across multiple channels and a large number of sub-channels. This paper investigates this scheduling problem through an optimization formulation with the objective of minimizing LTE’s adverse impact on Wi-Fi users. This is achieved by optimal allocation of radio resources at channel and sub-channel levels to meet each LTE user’s uplink and downlink rate requirements. Special considerations of channel conditions are given during LTE scheduling. A major challenge here is to obtain an optimal (or near-optimal) scheduling solution on ~1 ms time scale — a stringent timing requirement for the algorithm to be useful in the field. Our main contribution is the development of CURT, a scheduling algorithm that can obtain near-optimal solution in ~1 ms. CURT exploits the unique structure of the underlying optimization problem and decomposes it into a large number of independent sub-problems. These sub-problems can be solved efficiently and in parallel by GPU multi-processors. By implementing CURT on Nvidia GPU/CUDA platform, we demonstrate that CURT can indeed deliver near-optimal scheduling solution in ~1 ms and meet all our design objectives.