為了因應未來資訊設備在傳輸上的多樣性，第五代行動通訊技術提供了對應的技術支援來滿足其需求。對於第四代行動通訊技術來說，在未來各種較為極端的服務需求迫使其在支援上有其困難性，而在不同服務需求的使用者同時存在於系統的情況下，要能妥善滿足各種服務在處理上將更為複雜。因此，在第四代行動通訊技術基礎上朝向各種服務上的支援為下一代技術發展重點之一。 3GPP R15提出了5G新無線電 (New Radio, NR)系統，其標準規範定義了更彈性以及更有效率之無線通訊技術，利如在訊框(Frame)結構上能夠依照需求情況進行時域與頻率無線資源的定義，以及在服務上定義了新的服務品質規範，而在協定以及物理層則透過Context的建構來針對其運作準則與限制加以定義。 在效能方面，通訊系統在服務需求的滿足(QoS fulfillment)、系統吞吐量(System throughput)以及系統服務容量(System capacity)依舊是無線技術必須探討的議題。 本論文研究係針對5G NR無線電接入網(Radio Access Network, RAN)當中無線資源管理(Radio Resource Management, RRM)相關議題進行討論，為了提供更好的系統效能，本研究提供了與RRM相關的規範技術在最佳化的問題定義，並且設計在機器學習架構上針對無線資源排程議題進行模擬與分析，完成與單一無線資源方法在效能上的比較。所設計的方法架構同時也適用於其他無線資源管理的議題，並且能夠進一步延伸所觀察的效能指標。;The fifth generation (5G) wireless system plays the role to provide the diverse services of future applications. It meets the bottleneck in the current communication system with the extremely service budget. And the coexistence of diverse requirement service takes more complexities of the system operation. 3GPP proposed the 5G New Radio (NR) technique specification taking more flexibility for the functionality and procedure to construct the new efficient radio access network (RAN) system. The flexible frame structure is defined as efficiently operations of the wireless communication. The new defined 5G quality of service (QoS) identifier (5QI) present the requirement of each kind of the application. And the context describing the operation criterion between the protocol and the physical stack is flexibly considered. For the performance aspect, the fulfillment of the service requirement, the system throughput, and the system capacity are still the issues for the utility of radio resource. Currently, there are many related works discuss the importance of radio resource utilization. In this article, we discuss the radio resource management (RRM) issue of the NR system. We arrange the RRM optimization direction for the performance target. And the machine learning architecture design for radio resource scheduling is proposed in the experiment part. The result of the performance comparison with the design and traditional scheme performs the advantage of the machine learning method. The proposed architecture is portable on the other RRM problem which has a similar characteristic. And it is suitable to extend the observation range.
