Chung Cheng University Institutional Repository
Not a member yet
889 research outputs found
Sort by
使用自定義舒適度考量進行電力成本優化;Cost Optimization of Electrical Usage with Customizable Comfort Considerations
No full text[[abstract]]對於大多數人來說，環境特徵如溫度的優化控制往往是較為一般性的;然而，某些特定的使用者可能找不到最佳的優化控制。基於大眾所知的預測平均投票（PMV）作為室內熱舒適指數和即時的人體回饋，本論文提出了使用者優化的控制策略，在功耗和使用者特定舒適度之間進行權衡。首先，PMV值的範圍根據個人的偏好進行動態修改。然後使用粒子群優化（PSO）來找到空調的設定，使得房間的熱環境在使用者可接受的熱舒適範圍內，實現了功率消耗和熱舒適度之間的權衡。與固定PMV設置控制方法相比，提出的控制方法可以更有效地將PMV值保持在使用者舒適範圍內。如果用戶喜歡溫暖的舒適環境，系統可以節省25％以上的功耗。如果用戶喜歡涼爽的熱舒適環境，系統雖然將耗費超過15％的功耗，但獲得超過68.3％的舒適度。
Optimized control of the environment features such as temperature is often generalized for the majority of people; however, a specific user might not find the controlled results are optimal for him/her. Based on the popular Predicted Mean Vote (PMV) as an indoor thermal comfort index and real-time human feedback, this work presents a user-optimized control strategy that trades off between power consumption and user-specific comfort. First the range of PMV values is dynamically revised according to an individual's preference. Then particle swarm optimization (PSO) is used to find a thermal setting for air-conditioners such that within the user acceptable range of thermal comfort a user-given tradeoff between the power consumption and thermal comfort is achieved. Compared with fixed PMV settings, the proposed control method can more effectively maintain the PMV value within the occupant comfort range. If the user likes a warmer thermal comfort environment, the system can save over 25% of power consumption. If the user likes a cooler thermal comfort environment, the system will spend more than 15% of power consumption, but gain more than 68.3% of comfort
網際網路多媒體子系統上基於NFV架構虛擬服務調用方法之設計與實作;Design and Implementation of a NFV based architecture for Virtual Service Invocation Function in the IP Multimedia Subsystem
No full text[[abstract]]IP多媒體子系統(IP Multimedia Subsystem；IMS)是一個可以提供使用者透過IP-based control plane的方式存取多媒體服務的系統，但是傳統IMS的服務調用的機制是透過呼叫會話控制功能(Call Session Control Function；S-CSCF)根據每個人購買的加值服務依序存取應用伺服器(Application Server；AS)，且S-CSCF判斷session會不會觸發加值服務的方式是依賴session的第一個request，但是有很多加值服務是收到第一個request之後的SIP訊息才會真的觸發服務，這個機制會造成S-CSCF的負擔過重，且通話建立的延遲時間會很長。本論文提出虛擬服務調用方法(virtual service invocation function；VSIF)架構來解決前述的問題，VSIF是部屬在IMS架構中且介於S-CSCF與AS之間，我們修改初始過濾規則(Initial Filter Criteria；iFC)的觸發表達式將多個iFCs合併成成一個，使得VSIF能夠接收S-CSCF的訊息且S-CSCF與VSIF只需要交換一次訊息。VSIF判斷會不會觸發服務是透過存在VSIF-DB裡的VSIF initial Filter Criteria (ViFC)來判斷，ViFC是由trigger Filter Criteria (tFC), execution Filter Criteria (eFC)與傳統的initial Filter Criteria (iFC)組成，利用這三種FC能大幅減少VSIF與AS的交換訊息個數，VSIF的設計不但能減少S-CSCF的負載與通話建立時間，網路營運商也不需要修改IMS原有架構即可使用，且經過我們利用Queuing Network分析，VSIF可減少S-CSCF的負荷與session建立的時間。我們將VSIF 設計得更具有擴展性，使它能夠佈署於網路功能虛擬化(Network Function Virtualization；NFV)平台，我們也透過Queuing Network的網路效能分析的方式設計出一個VSIF auto-scaling的演算法，並設計一套基於NFV架構的auto-scaling流程。此外，我們對VSIF與標準IMS作壓力測試實驗的比較，我們使用Clearwater部屬IMS，並且在OpenStack環境中進行測量與驗證。比較實驗結果指出VSIF相較於標準IMS有較高的通話建立成功率，且也能減少S-CSCF的CPU使用率，最後我們也實驗多個VSIF，以通話建立成功率與VSIF的CPU使用率來分析出VSIF在不同情況下的最佳部屬個數。
IP Multimedia Subsystem (IMS) is the service provide multimedia services to users through an IP-based control plane. The traditional IMS service invocation mechanism requires the Serving-Call Session Control Function (S-CSCF) invokes each Application Server (AS) sequentially to perform service subscription profile and the AS can be triggered only on initial SIP requests. However, many services need to be triggered only on the SIP subsequent request or response, which results in the heavy load of the S-CSCF and the long session set-up delay.This paper proposes a virtual service invocation function (VSIF) to fill these gap. The VSIF works as an overlay on the top of the current IMS infrastructure and it can be applied transparently to SIP dialogs between the S-CSCF and the AS. We change the trigger point expression to combine multiple iFCs to one. The VSIF can receive the message from S-CSCF and they only need to exchange message once. The VSIF is responsible for allowing the messages to be relayed to the AS only when the service really needs to be executed. The operation of VSIF is based on VSIF initial Filter Criteria (ViFC) which is stored in VSIF-DB and it consisted by trigger Filter Criteria (tFC), execution Filter Criteria (eFC) and original initial Filter Criteria (iFC). These FC can help the VSIF to have less exchanged messages with the AS. The VSIF not only reduce the load of S-CSCF and the call setup delay but also enables operators to deploy it without any change in the existing IMS core. After we analyzed by the Queuing Network, the result indicates that the VSIF achieves a shorter service invocation time and lower load of the S-CSCF. We design the VSIF with high scalability for deploying in the Network Functions Virtualization (NFV) platform. We also design a VSIF auto-scaling algorithm with Queuing Network and auto-scaling procedure to provide a cost-performance tradeoff. Moreover, we have a proof of concept evaluation to stress test on the VSIF and standard IMS. We measured and evaluated on top of an OpenStack cloud with Clearwater as our IMS environment. It presented the call success rate of the VSIF architecture is higher and the architecture can also reduce the CPU utilization of S-CSCF. Finally, we have an experiment to stress test on distinct numbers of VSIFs and analyzed the optimal number of VSIFs with the call success rate and CPU utilization of the VSIF separately
支援多藥材功效之中醫藥材配伍系統;Supporting Multiple Medicinal Effects in Medicinal Combination System for Traditional Chinese Medicine
No full text[[abstract]]本研究團隊先前研製一套基於中醫藥材知識本體的中醫藥材自動配伍系統。根據中醫方劑的君、臣、佐、使配伍原則，依據使用者指定的藥材功效，該系統可以自動推薦效力最佳的藥材。該系統共包含補氣類的35味藥材、28帖方劑、及65個標準化過的藥材功效。該系統的藥材功效標準化主要根據原始的藥材功效敘述，尚未考慮藥材的性、味及歸經。該系統開發一個強制多功效藥材配伍演算法。該系統依據君、臣、佐、使的順序推薦藥材。依據使用者指定的藥材功效，系統會強制性地加入順序在前的藥材功效，使用多目標最佳化演算法，計算效力最佳的藥材。本論文在該系統的基礎上做了下列擴充。第一，本論文擴增其他虛證常用的方劑，總共包含155味藥材及112帖方劑。第二，本論文藥材功效標準化增加考慮藥材的性、味及歸經，擴增至221個標準藥材功效。第三，本論文開發一個彈性多功效藥材配伍演算法，使用者可以指定任意多個藥材功效，再使用多目標最佳化演算法，計算效力最佳的藥材。根據在這112帖方劑的初步使用經驗，確認新演算法不但顯著提升指定多藥材功效的彈性，也提高推薦最佳效力藥材的精確度。
Our research team had developed an ontology-based medicinal combination system for traditional Chinese medicine. Based on the combination principles of sovereign, minister, assistant, and courier medicinals, this system can automatically recommend the most effective medicinals given a medicinal effect. This system contained only 35 medicinals, 28 formulas, and 65 standardized medicinal effects for qi deficiency syndrome. The standardization of medicinal effects in that system was based on the original descriptions of medicinal effects in the reference books and didn’t take into account the qi, flavor and meridian entry of medicinals. This system had developed a mandatory multiple medicinal effect combination algorithm. This system recommended medicinals following the order of sovereign, minister, assistant, and courier medicinals. Given a medicinal effect, this system will mandatorily take the medicinal effects of formerly recommended medicinals into account and use a multi-objective optimization algorithm to compute the most effective medicinals.Based on the formal system, this thesis makes the following three expansions. First, this thesis extends the system to contain 155 medicinals and 112 formulas for various deficiency syndromes. Second, this thesis extends the standardization of medicinal effects to take into account the qi, flavor and meridian entry of medicinals, and contains 221 standardized medicinal effects. Third, this thesis also develops a flexible multiple medicinal effect combination algorithm. The user can assign arbitrary number of medicinal effects and use a multi-objective optimization algorithm to compute the most effective medicinals. A preliminary experience on these 112 formulas confirms that the flexibility of this combination algorithm is significantly better and the precision of recommending the most effective medicinals is significantly higher
限制式白箱函式層級單元測試案例產生器;Constraint-Based Test Case Generation for White-Box Method-Level Unit Testing
No full text[[abstract]]本論文將測試案例產生問題定義成限制滿足問題，並實作一個限制式白箱函式層級單元測試案例產生器。一個測試案例包含一個測試輸入及其對應的預期輸出。針對測試輸入，這個工具先將程式碼轉換成測試模型，限制邏輯圖。限制邏輯圖以圖形的方式描述函式的實際行為或函式輸入及輸出的限制邏輯關係。在限制邏輯圖上，這個工具接著用圖形的方式進行函式行為的等價類別分割及測試覆蓋標準管理，以篩選數量少但覆蓋廣的代表性行為來測試。產生每個代表性行為的輸入可以被定義成一個限制滿足問題。這個限制滿足問題可以用限制邏輯程式來描述並求解。針對預期輸出，物件限制語言所定義的函式預期行為也可以被視為一個限制滿足問題。因為物件限制語言不能執行，這個工具會將物件限制語言的描述轉換成可執行的限制邏輯程式的描述，並依代表性行為的輸入解出對應的預期輸出。這個工具的特色是使用一致的技術來分別產生白箱函式層級單元測試案例的測試輸入及預期輸出。關鍵詞-單元測試、白箱測試、限制式測試測試案例產生、函式層級測試、限制邏輯程式
This thesis defines the test case generation problem as a constraint satisfaction problem, and implements a constraint-based test case generator for white-box method-level unit testing. A test case contains a test input and the corresponding expected output. For test inputs, this tool first transforms the source program into a constraint model called a constraint logic graph. A constraint logic graph describes graphically the actual behaviors of a method or the constraint logic relationships between the inputs and the actual outputs of the method. On the constraint logic graph, this tool then performs equivalence class partitioning of the actual behaviors and test coverage criteria management using graph-based approaches to generate a small-quantity and wide-coverage group of representative behaviors for testing. The problem of generating the input for a representative behavior can be defined as a constraint satisfaction problem. This constraint satisfaction problem can be described and solved by a constraint logic programming language. For expected outputs, the expected behaviors of a method defined by the Object Constraint Language can also be viewed as a constraint satisfaction problem. Because the Object Constraint Language is non-executable, this tool will transform the specification described by the Object Constraint Language into an executable specification described by a constraint logic programming language. Given an input, the executable specification can be solved to generate the corresponding expected output. One of the features of this tool is that it uses an uniformed constraint-based approach to generate both test inputs and expected outputs for white-box method-level unit testing.Keywords- unit testing, white-box testing, constraint-based testing,test case generation, method-level testing, constraint logic programmin
基於HADOOP實作動態分析資料表關聯調整資料放置方式;Implementing Correction Detection Between Tables to Adjust Data Placement Strategy Based on Hadoop
No full text[[abstract]]在面對巨量資料(Big Data)的分析時，關聯式資料庫已經無法負荷如此龐大的資訊量。Hadoop可以透過HDFS儲存大量資料，並且透過MapReduce分析巨量資料。由於HDFS的高容錯性和高擴展性，許多企業都開始將資料轉移到Hadoop中，利用雲端運算環境處理大量的資料。而Hadoop對於資料的儲存，是將資料以資料區塊作為區分，以隨機存放的方式，平均分散到叢集中，並沒有考慮資料的相關性。使得在執行分析任務時，有相關性的資料被分散到不同的節點中，增加了需要透過網路交換資料的成本，對於效能有很大的影響。本篇論文提出了在將關聯式資料庫轉移到Hadoop時，當資料從關聯式資料庫轉移到Hadoop時，將儲存的資料區塊從原先只是隨機選擇儲存地點，改進成透過檢視運算節點效能和資料表的參照關係，提供較佳的儲存策略，使得有相關性的資料區塊放置在一起，使得查詢資料時能夠降低資料的找尋成本，讓整體執行效能獲得提升。
Nowadays, in case of Big Data analysis, traditional database management system is no longer capable of processing such big data. Hadoop could save huge amount of data via HDFS and analyze huge amount of data via MapReduce. Owing to scalability and fault tolerance of HDFS, many enterprises now are transferring data to Hadoop, taking advantage of cloud computing for dealing huge amount of data. But for the method of storage of data of Hadoop is dividing data based on Block and store randomly and evenly into cluster. This method lacks the importance of correlation between data and makes relative data separated to different nodes when executing analysis, which adds up cost of network transmission and affects performance immensely. This paper address that while data are transferred from relational database to Hadoop, data block will be selected from efficacy of computing nodes and correlation of tables, instead of randomly choosing which provides better strategy of storage. More relative data is stored together, less search cost is spent when search performs and more efficiency is elevated for executing performance
基於深度學習之法律關鍵詞擷取;Keyword Extraction from Law Database Using Deep Learning
No full text[[abstract]]近年來，國際交流與貿易合作頻繁，國家間法律實務的互動需求也持續增加。 建構出標準翻譯法律關鍵詞典，作為翻譯的參考基礎，可以確保文件的一致性。因 法律數量之龐大，涵蓋領域甚廣，倘若以人工方式標註所有法律關鍵詞，將會是個 效率不彰的浩大工程。本研究基於深度學習建立一關鍵詞擷取系統，協助擷取法律 專家認定的關鍵詞。透過本研究所提出之分詞重組方法，產生候選法律關鍵詞集， 相較只使用分詞結果作為候選詞集，此法有效提升候選詞集涵蓋法律關鍵詞集之 涵蓋率，且相較傳統以字為單位之 n-gram 重組，可以大幅降低候選詞集數量，減 少後續分類運算量。亦針對法律領域文本特性，提出有效的鑑別特徵，並進行合適 的類神經網路模型訓練，根據實驗結果，本研究提出之架構可以精準的取得大量法 律關鍵詞，準確率達 92.6%，精確率達 89.2%，回現率達 88.2%，F-measure 達 88.2 %。
In recent years, international exchanges and trade cooperation become frequent. Legal practice among nations increases as well. To ensure consistency of legal document, constructing a standard translation dictionary as a translation reference is necessary. However, there are plenty of laws and regulations which cover a wide range of area. If we manually mark all legal keywords, it will be highly inefficient. In this thesis, we propose an automatic Chinese legal keyword extraction algorithm based on deep learning technology, specifically, Back Propagation Neural Network (BPNN). This system helps extract legal keywords from legal documents which legal expert would identify. This system consists of two parts: candidate keyword generation and keyword identification. First, the keyword candidate set is generated by using the word segmentation and combination method proposed in this study. Compared with word segmentation without combination, this method effectively improves the coverage of the actual legal keyword set. Furthermore, compared with the traditional n-gram combination of Chinese words, it can significantly reduce the number of keyword candidates and cost less time on following classification. To identify legal keywords based on BPNN, specific features are first defined based on the characteristics of legal keywords identified by experts to improve the performance of BPNN. A real world data set is then used to train the BPNN. Experimental results show the effectiveness of the proposed approach with the overall average accuracy, precision, recall and f-measure values of 92.6%, 89.2%, 88.2% and 88.2% respectively. All of these measures have been significantly improved as compared to our previous work
傳統製造業基層員工堅忍人格特質對作業環境與工作鑲嵌關係之調節效果研究;The Moderating Effect of Hardiness Personality on the Relationship between Work Contexts and Job Embeddedness among Traditional Manufacturing Employees
No full text[[abstract]]本研究主要探討傳統製造業基層員工之堅忍人格特質、作業環境與工作鑲嵌的相互關?性，以及堅忍人格特質對作業環境與工作鑲嵌之調節效果。其主要目的是傳統製造業基層員工本身所具備的堅忍人格特質是否會因為作業環境感受之差?，進而提高工作鑲嵌程?，使傳統製造業人?資源單位在招募的過程中，?能準確聘僱合適人才，以達到減少人?成本損失及時間?費之成效。本研究採用採用問卷調查的方式，均為自陳式?表，受試者皆為傳統製造業基層員工，共收集247 份有效樣本，研究結果如下:（1）作業環境感受與工作鑲嵌無顯著負向關係。（2）作業環境感受與堅忍人格特質無顯著正向關係。（3）堅忍人特質與工作鑲嵌存在顯著正向相關性。（4）堅忍人格特質的調節效果?顯著於作業環境感受與工作鑲嵌的關係。
The study aims to explore the relationships among traditional manufacturing employee’s hardiness personality, work contexts and job embeddedness as well as the moderating effect of hardiness personality on the relationship between work contexts and job embeddedness. Its main purpose is the traditional manufacturing employees themselves have the hardiness personality whether it will feel the differences in work contexts, and thus improve the degree of job embeddedness. Further, the traditional manufacturing human resources units in the recruitment process, more accurate hire the appropriate people to reduce the costs and time wasted in order to achieve the effectiveness.This study used survey methods, both self-report scales, subjects were traditional manufacturing employees in Taiwan and researcher collected 247 valid questionnaires.The results included:1. Work contexts is not negatively related to job embeddedness.2. Work contexts is not positively related to hardiness personality.3. Hardiness personality is positively related to job embeddedness.4. Hardiness personality does not moderate the relationship between work contexts and job embeddedness
以LoRa為基礎之無線網狀網路通訊模組的設計與分析;Design and Analysis of a LoRa-based Wireless Mesh Networking Module
No full text[[abstract]]物聯網使數以萬計的節點透過網際網路分享資訊，這些節點資訊透過有效的利用可以達到許多應用，例如城市內的交通控制、環境監控等。目前常見的通訊技術應用範圍大部分是應用於個人網路(PAN)或是小區域的網路(LAN)，如Wi-Fi、ZigBee和BLE等。這些無線技術傳輸距離較短，比較不適合應付大範圍的節點運作。本研究探討新興無線技術LoRa，以低速率、長距離且低耗電的特性，適合應用於長距離的無線傳輸。本研究整合Mesh Network架構，開發出一個基於LoRa的無線網狀網路通訊模組Bonora，其特點為不受限於類似LoRaWAN的星狀網路設計，而可以透過多跳來傳輸資料，提升傳輸成功率。針對室內或是建築物的阻隔所造成通訊不良的影響，Bonora以多跳方式將感測資料回傳，提高封包傳輸成功率。我們把19個節點佈建在中正大學校園600公尺乘800公尺範圍內，實驗結果顯示Bonora能透過多跳方式提高節點的傳輸成功率，所有節點的平均資料傳輸成功率達88.5%；在本實驗架構下若是採用LoRaWAN之星狀網路架構平均傳輸成功率僅剩58.7%。最後，本研究以訊號強度(Received Signal Strength Indicator, RSSI)與訊雜比(Signal-to-noise ratio, SNR)歸納出推估傳輸成功率(Packet Delivery Ratio, PDR)的模型。使用者透過此模型可以了解當前節點的通訊狀況，用以調整網路中節點，提高傳輸成功率。
LoRa, an emerging wireless technology, is examined for long range communication performance in a large area. Although LoRa shows good performance for long-distance transmission in the countryside, its radio signals can be attenuated and interfered with by buildings, trees and other signal sources. In urban areas, our observation shows that LoRa requires a dense deployment of LoRa gateways to ensure that indoor LoRa devices can transfer data back to remote servers. Mesh networking is a solution for increasing the communication range and packet delivery ratio without the need to install additional LoRa gateways. This study presents our design of a LoRa mesh-networking module, the Bonora, for IoT applications (e.g.,collecting data from multiple, widely distributed sensors on a university campus). We deployed 19 LoRa mesh-networking devices distributed in an 800m by 600m area on campus to serve as a gateway and to collect data at one-minute intervals. The results show our LoRa mesh networking module achieved a 88.49% packet delivery ratio (PDR), while the LoRaWAN in star-network topology was only 58.7% under the same conditions
於IaaS中利用虛擬安全設備建置聯合防禦系統之設計與實現;Design and Implementation of Collaborative Defense System Based on Virtual Security Devices in Infrastructure as a Service
No full text[[abstract]]隨著通訊網路技術以及數據應用的快速發展之下，雲端技術也廣泛的應用在生活之中，例如：Google Drive、Amazon Web Service、VMforce…等等，依雲端提供的服務不同可分為雲端軟體服務(Software as a Service，SaaS)、雲端平台服務(Platform as a Service，PaaS)、雲端設備服務(Infrastructure as a Service，IaaS)。許多SaaS與PaaS業者選擇使用IaaS租用大量虛擬機以建置所需要的服務，而如何確保IaaS中虛擬機的安全性便是一個重大的挑戰。本論文提出利用軟體定義網路(Software-Defined Networking，SDN)與虛擬安全設備建置出安全的雲端環境，並藉由虛擬安全設備之間與外部實體防火牆的訊息交換，能夠先行配置對應的安全政策，再利用外部實體防火牆將惡意流量阻擋在雲端主機外，維持雲端服務主機的資源可用性。在此架構中，IaaS用戶可依不同安全需求自行增加封包比對規則，使整體雲端設備運用更為彈性。本論文將運用虛擬安全設備之間與外部實體防火牆的聯合防禦機制，保護Cloud Data Center中的虛擬機免於遭受攻擊，且能將惡意流量阻擋在雲端主機外部。考量到雲端服務提供給不同租戶(Tenant)使用，安全策略無法一體適用，因此利用租戶自訂的封包比對規則來達到租戶不同的安全需求。本論文研究重點著重於雲端架構、封包導向、虛擬設備與實體防火牆之間的聯防機制之設計，使雲端設備服務中的虛擬機獲得更高的安全等級。
With the rapid development of network technology and big data, cloud computing has been widely utilized in life. E.g., Google Drive, Amazon Web Service, VMforce. According to the service of the cloud provision, the service can be divided into Software as a Service, Platform as a service and Infrastructure as a service. Many SaaS and PaaS provider use IaaS to rent a large number of VM to build their service. In addition, how to secure VMs in IaaS is a big challenge. This paper proposes the use of Software-Defined Networking and virtual security devices to build a safe cloud environment. Through the virtual security devices and the physical firewall message exchange, the security rule can be applied in advance to block the malicious traffic outside of the cloud servers. In this architecture, the tenants of IaaS can add their packet matching rules according to the security requirements, and make the cloud more flexible.This paper uses the collaborative defense of virtual security devices and physical firewall to avoid VMs in Cloud Data Center from attack, and the proposed architecture can block the malicious traffic outside of the cloud server. Considering the cloud service is provided for different tenants, the security policy cannot apply uniformly. Therefore, this paper uses the flexible of packet matching rules to achieve the requirements of tenants. This paper focus on the design of cloud architecture, collaboration between virtual security devices and physical firewall, and flow control. The architecture will make the VMs in IaaS to obtain a higher security level
波束分群5G上鏈系統之多用戶檢測技術設計;On the Design of Uplink Multi-user Detection for Beam-group Division 5G Systems
No full text[[abstract]]為了達到5G系統高傳輸率及高系統容量的需求，5G系統將採用高頻段的無線資源和巨量天線陣列。當基地台使用巨量天線同時服務大量的用戶，由於天線域(antenna domain)的通道呈現高度頻率選擇性，會使系統效能大幅減低。此外為了獲得通道資訊，系統亦必須大量的進行通道估測。由於波束域(beam domain)訊號處理具有解決通道的頻率選擇性，並且能夠利用高頻段的通道稀疏特性，以有效降低用戶通道估測的系統複雜度，因此非常適用於5G系統。此外5G系統若採用高階MU-MIMO(multi-user multiple input multiple output)技術，系統的複雜度異常龐大。為了解決複雜度的問題，本論文提出一個新的波束分群5G系統(beam-group division 5G System)，本系統的優點主要是藉由將干擾嚴重的用戶形成同一用戶群進行檢測，來提升系統效能並更進一步降低系統的複雜度。本篇論文主要研究波束分群5G系統上行鏈路的多用戶檢測技術。首先提出利用通道稀疏性及最小均方誤差(minimum mean-squared error, MMSE)為準則設計的檢測器，來降低用戶通道估測和多用戶檢測的系統複雜度。其次，提出利用用戶通道正交的特性對用戶進行選擇，此舉能提升系統效能。最後，提出一用戶分群方法在結合上述兩種設計後，能在提升系統的效能下更進一步降低系統的複雜度。本篇論文經過模擬後，證實所提出的檢測器在系統效能只損失2.7%的情況下，能夠降低61%的計算複雜度。此外使用所提出的用戶選擇方法，系統能提升25%的效能。最後，在配合所提出的用戶分群方法後能讓系統效能在提升12%情況下，系統計算複雜度能進一步降低64%。
In order to gain high transmission rate and system capacity in the future 5G systems, 5G systems will use high-frequency radio resource and massive antenna arrays. When base station uses massive antennas to serve a lot of users simultaneously, the performance of the system will be degraded due to a high frequency selectivity for channels on antenna domain. In addition, huge complexity on channel estimation is inevitable because the larger the number of channel is, the higher the computation complexity. Recently, the beam domain signal processing, has been proposed to reduce the frequency selectivity and complexity of channel estimation by utilizing the channel spasity characteristics of high-frequency bands. In addition, the complexity of high-order MU-MIMO is very high if it is employed to increase system capacity. To solve these problems, we proposed a new beam-group division 5G System. The concept of this system is to group the users with serious interference to a user-group for detection to improve the performance and to reduce the complexity of the system.This paper investigates the issue of multi-user detection of the uplink beam-group division 5G system. First, a detector is proposed based on channel sparse characteristics and minimum mean square error (MMSE) criterion to reduce the complexity of user channel estimation and multi-user detection. Second, a user selection method is proposed based on channel orthogonality to improve system performance. Finally, a new user-grouping method in combination with the above two methods is proposed to further reduce the system complexity and improve system performance .Our numerical results show that first the proposed detector can reduce the computation complexity by 61% only with 2.7% degradation on system performance. Second, by using the proposed user selection rule, the system can increase the performance 25%. Finally, the proposed user-grouping algorithm along the above two methods can reduce overall computation complexity by 64% and increase system performance 12%