| 摘要: | 隨著第五代行動通訊技術(5th generation mobile networks, 5G)商用網路推出，越來越多的物聯網應用被提出，其中許多遠端控制的應用如：工廠自動化、自動駕駛、遠程醫療手術等需要實時高精度的操作，對封包的成功率及延遲有嚴苛的要求。5G白皮書提出了三大場景：增強型行動寬頻通訊(Enhanced Mobile Broadband, eMBB)、超可靠低延遲通訊(Ultra-reliable and Low Latency Communications, URLLC)及大規模機器型通訊(Massive Machine Type Communications, mMTC)，其中URLLC便是針對需要實時高精度的物聯網場景而設計，在此種場景下，32bytes的封包傳輸需在1ms內完成，且成功率需達到1-10-5。在5G的標準中，無允諾上行(Grant-Free)及混合式自動重送請求(Hybrid Automatic Repeat request, HARQ)被提出來達成URLLC的要求，前者繞過設備上行時需和基地台獲取傳輸許可的步驟來降低延遲，後者則是透過多次重傳相同封包以提升成功率。
本論文利用機器學習及數學模型兩種方式，針對HARQ的上限次數做出動態調整，以最少的重傳次數達成URLLC的要求，以減輕設備因連續重傳造成的負擔結果表明在URLLC頻繁傳輸時，降低其HARQ上限能夠改善eMBB的成功率，在動態調整HARQ的部分，機器學習的方式會因探索的機制而略有些不穩定，數學模型的方法能夠有較穩定的結果，但使用上也會有額外的條件限制。
;With the launch of 5th generation mobile networks (5G) commercial networks, more and more IoT applications have been proposed. There are some of remote control applications such as factory automation, autonomous driving, and telemedicine surgery which requires real-time and high-precision operations. Based on those applications, it has strict requirements on the success rate and delay of the packet. The 5G white paper proposes three scenarios: Enhanced Mobile Broadband (eMBB), Ultra-reliable and Low Latency Communications (URLLC), and Massive Machine Type Communications (mMTC), where URLLC is designed for IoT scenarios that require real-time and high-precision. In this scenario, 32bytes packet transmission needs to be completed within 1ms, and the success rate needs to reach 1-10-5. In the 5G standard, Grant-Free and hybrid automatic repeat request (Hybrid Automatic Repeat request, HARQ) are proposed to meet the requirements of URLLC. To reduce the delay, Grant-Free upload data without constructing UL-grant. HARQ is to increase the success rate by retransmitting the same packet multiple times.
This paper proposed two methods: machine learning and mathematical models to achieve the target. The goal of this paper is to make a dynamic adjustment to the maximum amount of re-transmission of HARQ. Also to reduce the burden of equipment caused by continuous retransmissions. Simulation result show that reduce the number of retransmission can improve the performance of eMBB. In adjustment of HARQ repk, the mathematical method is more stable than machine learning, but there are stricter conditions for use. |
