與Wi-Fi公平有效共存於免執照頻段之LAA 多工接取層自適應參數設定研究

Adaptive MAC Parameter Setting for Fair and Efficient Coexistence of LAA and Wi-Fi in Unlicensed Band

方敬勻(Jing-Yun Fang)

指導教授 ： 張時中

國立臺灣大學/電機資訊學院/電信工程學研究所/碩士(2018年)

https://doi.org/10.6342/NTU201803980

摘要

為了處理LTE無線網路快速增加的資料流量，3GPP提出了Licensed Assisted Access (LAA) 技術使用載波聚合(Carrier Aggregation)技術來整合執照頻譜(licensed band)以及 5GHz 免執照頻譜(unlicensed band)頻譜資源，增加可用頻譜以及系統容量。然而使用免執照頻譜時，LTE-LAA將無可避免的與免執照頻譜的大宗Wi-Fi產生共存問題。
LTE-LAA與Wi-Fi採用先聽後說Listen-Before-Talk機制來接取頻譜資源，Listen-Before-Talk (LBT)機制會在傳輸前會依照Contention window(CW)和back-off stage(m)設定等待數個時間幀(time slot)來分享頻譜資源。然而，LTE-LAA除了LBT外還有獨特的CCA機制，如果通到空閒CCA 時間則LTE-LAA會直接開始傳輸。除此之外，LTE-LAA是中央式管理且有執照頻譜來協調免執照頻譜。因此，LTE-LAA比Wi-Fi有更高的頻譜使用效率(spectrum efficiency)。然而，增加LTE-LAA的接取將會佔據Wi-Fi的頻譜資源造成不公平。如何客觀的訂定比例性公平(Proportional Fairness)指標並且在公平以及效率中取捨是我們在LAA與Wi-Fi共存網路中的核心問題。
為了調整共存網路中的公平以及效率關係，我們研究對3GPP在LAA接取機制中所訂定的三個參數 CCA, CW, m進行調整，來改變LAA在與Wi-Fi共存下的資源配置。為此，我們需要理解三個參數設定對公平及效率的影響來做合適的配置。然而，除了接取機制以外，LAA與Wi-Fi的流量(traffic load), 地理位置 以及偵測閥值(detect threshold)都會影響LAA及Wi-Fi的資源分配。
在單一運營商擁有的LAA和Wi-Fi間，由於無法完全掌握共存網路中共存的所有裝置與通道的狀態，需要為LAA設計有適應環境調整參數設定的功能，以有效共存。
在本研究當中，我們以3GPP TR36.889所設定的LAA與Wi-Fi共存環境進行研究，考慮室內環境中，使用cat.4 LBT的LTE-LAA與Wi-Fi在5GHz免執照頻譜的下行接取情形。主要的研究問題以及相應的挑戰為:
P1.如何基於LTE-LAA與Wi-Fi網路接取的比例公平(proportionally fair)及效率(efficiency)訂定合理的效能指標?
C1.訂定除測量吞吐量(throughput)與理論傳輸速率(theoretic link data rate)外考慮流量的比例性公平指標及效能指標公平與效率間的取捨。
P2.公平效率的效能指標與LAA 先聽後說(Listen before talk)機制參數(CCA, CW, m)之間的關係為何?
C2.LAA-LBT中CW, m以及新的CCA參數設定與網路效能間關係複雜，目前沒有可供快速、精確估計的數學模型。
P3.如何適應(adapt)網路環境設定出最佳的LAA LBT參數(CCA, CW and m)?
C3.設計可根據回報的網路效能指標找出最佳LAA LBT參數配置的演算法。
在本篇研究當中，我們基於Jain’s fairness index設計我們的比例性公平標準,計算公平性時使用測量的吞吐量做為分子、取基站流量和理論傳輸速率的較小者最為分母，因此達到考慮真實基站流量需求和避免無效資源分配。而後將我們的比例公平指數及網路總吞吐量加權相加得到效能指標(Performance index) 。我們設定不同的效能指標權重對參數配置與效能指標的關係進行分析，讓其他研究者可以依照結果設置權重達到想要的公平和效率分配。
為了得到參數設置與公平跟效率的關係，我們延伸Wi-Fi LBT 馬可夫鍊(Markov-Chain)模型，加入CCA機制相關的狀態(state)及狀態轉換流程(transition flow)發展出LAA與Wi-Fi共存網路的數學模型。此模型可以快速的估算出網路效能並結得出參數與效能的關係，並得到CCA機制的特性，因此，我們借由此模型設計了依序輸入CCA, CW, m進行分析得到最佳參數的數值(numerical)最佳化方法。但是此模型主要用於LBT機制的分析，在進行真實網路模擬時發現效能也會被LBT外因子影響，所以，我們基於reinforcement learning原理設計了我們的LAA LBT參數適應演算法(Adaptive LAA LBT Parameters Algorithm)，此演算法可以藉由目前環境回報的效能指數對LTE-LAA LBT參數進行動態的調整，從而在複雜環境中找到最佳配置。即使環境發生改變，此演算法也可以持續的對參數進行動態修正。最後，我們在現有LTE-LAA 協定上(protocol)加入參數最佳化模組。
我們開發了Wi-Fi與LAA共存網路的模擬器進行模擬，同時將我們的參數最佳化模組設計實做於模擬平台，進行概念驗證與效能評估。結果顯示當使用數值最佳化方法於狹小室內時，在最佳化效率時可以達到10%的網路總吞吐量改善，而在最佳化公平性時則都可以得到0.95以上的公平指數。而使用適應(adaptive)最佳化於寬闊的室內時，可以在不影響系統公平性的狀況下提升20%的網路總吞吐量，同時在最佳化公平性時也可較標準配置得到0.1-0.2的公平指標改善。

關鍵字

免執照長期演進技術 ； 需執照協助之存取 ； 公平 ； 效率 ； 公平效率取捨 ； 長期演進技術及WiFi共存 ； 先聽後說機制 ； 協定設計 ； 自適應參數調整 ； 強化學習演算法 ； 馬可夫鍊

並列摘要

To cope with rapid traffic growth of LTE wireless access networks, 3GPP has proposed Licensed Assisted Access (LAA) technology to operate in 5GHz unlicensed band. LTE-LAA aggregates licensed band and unlicensed band traffic through Carrier Aggregation (CA) technique. LTE-LAA technology can increase available spectrum resource and improve system capacity. However, LTE-LAA would inevitably coexist with Wi-Fi access networks, which have been widely operated in unlicensed band.
Both LTE-LAA and Wi-Fi MAC layer adopt Listen-Before-Talk (LBT) mechanism. To share spectrum resource, LBT mechanism will wait for several time slot before transmit data based on contention window (CW) and back-off stage (m) setting. However, LTE-LAA has unique CCA mechanism except LBT mechanism, which will instantly start transmission if station finds channel idle for CCA period. Furthermore, LTE-LAA can coordinate unlicensed band via licensed band LTE signal with centralized system. Therefore, LTE-LAA has more spectrum efficiency than Wi-Fi. However, increasing LTE-LAA channel access will occupy Wi-Fi spectrum resource and cause unfairness of channel accesses. Trade-off between spectrum efficiency and fairness of access is an important problem in coexistence network.
To strike a balance between fairness and efficiency, we study LAA MAC parameters (CCA, CW and m) which are specified by 3GPP to adjust unlicensed spectrum resource access between LAA and Wi-Fi. To make proper configuration, we have to analyze the relationship among three LAA LBT parameter setting, efficiency and fairness. However, except for channel access mechanism, traffic load, position and detection threshold of LAA and Wi-Fi would also influence network resource allocation. Because LAA and Wi-Fi stations which own by one operator can’t absolutely got the states and channel information of all devices, we need to design optimal LAA parameter setting algorithm which can adapt to current environment for effective coexistence.
In our research, we follow 3GPP specification (TR 36.889) LAA cat.4 LBT, and consider downlink access of unlicensed band channels in indoor environment. There are three research problems and their respective challenges are as follows:
P1. What is a meaningful performance index for proportionally fair and efficient spectrum accesses by LAA and Wi-Fi network?
C1. The definition of proportional fairness should include traffic load besides measured throughput and theoretic link data rate and its trade-off with efficiency in overall performance.
P2. Relationship between performance indices of fairness and efficiency and LAA LBT (CCA , CW , m) parameters?
C2. CCA which is new mechanism in LAA-LBT, CW and m have complex relationship to performance and currently have no numerical model for rapid and precise evaluation
P3. How to set optimal LAA LBT parameters (CCA, CW and m) adapt to environment?
C3. Design algorithm to find optimal parameters based on performance index feed-back.
In this research, we firstly give the performance a brief definition by weighting summing proportional fairness and system total throughput. We design proportional fairness criteria based on Jain’s fairness index. We use measured throughput as numerator and apply lesser between traffic load and theoretic link rate as denominator for considering traffic demand and avoiding ineffective resource allocation. Then, we simulate the different weighting of fairness and efficiency to analyze the relationship between LAA parameters setting and performance index. Network designer can set proper weighting to achieve desired fairness and efficiency based on our analysis.
Following, to clarify the relationship among LAA parameter setting, fairness and efficiency. We extend the Wi-Fi LBT Markov-chain model with new states and transition flows to establish LAA model with adjustable CCA. MC model can rapidly evaluate network performance and obtain property of CCA mechanism so we exploit MC model to design numerical optimization method which sequentially searches CCA, CW, m to find LAA optimal parameter setting.
However, the MC model can only handle behavior of LBT mechanism and we find other factors also influence coexistence network performance a lot in network simulation. Thus, we design an Adaptive LAA LBT Parameter Algorithm (ALLPA) based on reinforcement learning. This adaptive optimization algorithm can dynamically update configuration based on the environment performance feed-back. As a result, ALLPA can update optimal parameters in complex and changing environment.
We develop a network simulator to simulate coexistence network and implement our optimization modules into the experiment platform to prove the concept and evaluate algorithmic performance. As a result, when applying numerical optimization in a 20mx20m indoor environment, we can improve 10% system throughput and achieve fairness index values higher than 0.95. When applying adaptive optimization in a 100mx100m indoor environment, we can obtain 20% system throughput improvement without degrading fairness index and improve fairness index values by 0.1 – 0.2 as compared to fairness index value achieved under standard LAA parameter setting.

並列關鍵字

LAA ； Wi-Fi ； fairness ； efficiency ； LAA-WiFi coexistence ； listen-before-talk ； protocol design ； reinforcement learning ； markov-chain ； trade-off between fairness and efficiency ； LBT parameters adjustment ； adaptive algorithm

參考文獻

[3GPP RP-140808] "Review of Regulatory Requirements for Unlicensed Spectrum" 3GPP Internet draft, RP-140808, Oct. 2014

[3GPP RP-141664] Verizon, "Study of Licensed Assisted Access Using LTE, " 3GPP Internet draft, RP-131680, Sep. 2014.

[3GPP TR 36.889] "Feasibility Study on Licensed-Assisted Access to Unlicensed Spectrum." 3GPP Internet draft, TR 36.889, Jul. 2015

[ACP13] E. Almeida, A. M. Cavalcante, R. C. Paiva. "Enabling LTE/WiFi coexistence by LTE blank subframe allocation." Communications (ICC), 2013 IEEE International Conference on, Jun 2013.

[ADK14] N. L. Van Adrichem, C. Doerr, F. A. Kuipers. "Opennetmon: Network monitoring in openflow software-defined networks. " In Network Operations and Management Symposium (NOMS), pp. 1-8, May 2014.