本計畫為兩年期之規劃，主要目的針對4G及5G網路物聯網上行資料隨機存取演算法之設計與比較。所探討的存取網路包括4G NB-IoT、及5G的URLLC與NOMA-SCMA，其中，4G NB-IoT為Grant base存取方式，而5G則是Grant free的存取方式。在演算法設計方面，將分別提出Rule based及Machine Learning based兩種設計方法。在傳輸性能比較上，將以防災應用(即時非週期性短時間大量訊務+非即時週期性訊務)、智慧工廠(即時週期性與非週期性訊務)、及醫療照護(即時非週期性訊務+週期性訊務)等三種物聯網訊務，進行模擬實驗比較，並檢視各演算法之詳細運作過程，從中分析不同存取網路、及不同演算法設計、在不同物聯網應用服務之適用性。兩年之主要計畫內容略述如下：第一年度：針對LTE NB-IoT及URLLC兩種網路之存取演算法進行設計及效能比較，在相同環境、假設條件、與預期目標下，分別提出Rule based及Machine Learning based共四種存取演算法，並進行在不同應用環境下之效能比較。第二年度：與OFDMA架構不同，SCMA採用非正交多工方式進行傳送，使得SCMA可以在同一個子載波上提供更多的CTU供UE傳送，能有更高的頻譜使用率，在演算法設計上，本研究將著重對CTU依演算法規畫進行分類，數量分配，並進行相關mapping rule的設計，以提升大量UE環境下之上行品質。除此之外，也將綜合分析兩年所提各種方法在不同物聯網服務之適用性。 ;This is a two-year proposal. The main focus of this proposal is to develop and compare the random access algorithms of uplink traffic for Internet of Things (IoT) in 4G and 5G networks. The target wireless networks include 4G NB-IoT, and 5G URLLC and NOMA-SCMA. Among them, 4G NB-IoT is grant based access while 5G is grant free access scheme. For the algorithm development, both of rule-based and machine learning-based approaches are applied. For the performance comparisons, three different traffic models, include applications of disaster prevention (aperiodic real time burst traffic and non-real time periodic data), intelligent factory (periodic and aperiodic real time data), and healthcare (real time aperiodic and periodic data), will be applied for detail examination during simulation process. And the applicability of different access networks and algorithms for different IoT services will be proposed. The main research contents of each year are provided as follows:The first year: we will develop the rule based and machine learning based algorithms for NB-IoT and URLLC access networks. Thus 4 algorithms will be proposed and compared in different application scenarios.The second year: Comparing to traditional OFDMA, SCMA applies non orthogonal multiplexing access, which provides more Contention Transmission Unit (CTU) for UE access. We will focus on the CTU numbers allocation of different classes and its mapping rule so that the uplink performance can be improved in huge UEs environment. And the applicability for different IoT services of the proposed algorithms in these two years will be completely analyzed.
