不完美通道狀態資訊下基於隨機最佳化之通訊品質保護與使用者配對問題研究

Abstract

現今第三代及第四代LTE系統正面臨使用者日增的需求 ， 為了解決此問題 ， 於下世代通訊網路中支援裝置之間的通訊設計被廣泛討論 ， 另外 ， 使用非正交多工接取技術以利用能量域的多訊號複用 ， 更可以有效提升頻譜效率 。 在本篇論文的第一部分 ， 我們在Rayleigh衰落通道下 ， 透過隨機最佳化 ， 針對如何分配同個資源給裝置間通訊使用者以及行動通訊使用者問題進行深入探討 ， 尤其在行動通訊使用者必須有較優先的保護下 ， 無線通道的隨機特性會導致裝置間通訊使用者以及行動通訊使用者共存干擾問題 。 不同於現今文獻只考慮長期的通道增益衰落 ，我們著重在通道隨機特性上 ， 決定裝置間通訊使用者的傳輸功率 ， 以最佳化總資料傳輸率並透過一臨界值保護行動通訊使用者通訊品質 。 首先 ， 我們介紹一個將目標函式 ， 限制函式以及隨機變數轉換成等效確定性函式的新技術 ， 因為此為非線性及非凸函式問題 ， 我們更嚴格的定義傳輸功率之上限限制 ， 以減少複雜度以及搜尋最佳解的不確定性 ， 接著 ， 我們提出兩種演算法 ， 一為將此非線性問題做線性規劃 ， 二為根據差分計算的次經驗法則直接解此非線性問題 。 模擬結果顯示 ， 在同樣的限制下 ， 我們所提出的演算法可以保護行動通訊使用者品質 ， 雖然在通道的不確定性下 ， 但裝置間通訊使用者可以在總資料傳輸率有更傑出表現 。 在論文的第二部分 ， 雖然很多文獻利用頻譜效率證明非正交多重接取技術勝過於正交多工接取 ， 但如何有效設計非正交多重接取排程演算法 ， 是我們著重的重點 。 考量實際子頻段應用以及其限制 ， 即在不同子頻段下 ， 同個使用者需使用相同的傳輸模式 ， 我們將排程建構成功率最佳化問題 。 為了解決此限制 ， 首先 ， 我們提出隨機分群方法將使用者分群 ， 在SINR限制下 ， 利用回傳的子頻段CQI以及功率分配因子來計算每個群的大小以及使用者距離 ， 接著 ， 將排程分成使用者分配以及功率分配兩個子問題 ， 並利用有母數分析方法 、 整數線性及非線性規劃來解決此問題 。 模擬結果顯示 ， 我們所提出的隨機分群方法 ， 比起使用k-means分群演算法 ， 有45%的效能增益 。

The current 3G/4G-LTE system is facing a problem in satisfying the relentless increasing demand from cellular users. To mitigate the problem, the 5G network is designed to support the Device-to-Device (D2D) communication. Moreover, the spectrum is expected to be used more effectively when signals from multiple users are multiplexed in the power domain using Non-Orthogonal Multiple Access (NOMA). In Part I of this dissertation, we aim to investigate the problem of D2D-mode users sharing the same radio resource blocks with the cellular-mode user under Rayleigh channel fading through stochastic optimization. The stochastic nature of the wireless channel causes the coexistence problem between cellular-mode and D2D-mode users a nontrivial task, especially when protection of cellular-mode users is strictly required. While related work has investigated the interference management problem in different scenarios, most approaches have considered only the long-term channel gain without explicitly addressing the randomness of the channel. Our objective is to determine the transmission power of all D2D-mode users for optimizing their sum data rates while ensuring the outage probability of signal sent by the cellular-mode user to stay below the desired protection threshold. We first introduce a new technique to transform both the objective and constraint functions involving random variables into equivalent yet deterministic forms. Since the formulated problem is non-linear and non-convex, we further tighten the upper bound of the transmission power constraint to reduce the complexity and uncertainty of searching for the optimal solution. To solve the formulated problem, we propose two different algorithms: the first algorithm reformulates and solves the problem as a linear programming (LP) problem while the second algorithm directly solves the non-linear problem based on the meta-heuristic of differential evolution (DE). Simulation results demonstrate that by using the proposed algorithms, the outage probability of the cellular-mode user can be maintained below the desired threshold despite the uncertainty of channel conditions, while the sum rate of D2D-mode users outperforms baseline methods under the same constraint. In Part II, our concentration is the scheduling method used for NOMA. While many endeavors have focussed on showing that NOMA has a practical advantage in terms of spectrum efficiency over OFDMA, there still needs more research on the design of efficient NOMA scheduling algorithms. We formulate the scheduling for NOMA as an optimization problem of energy efficiency in which we take a practical requirement for subband scenario into consideration. In particular, a NOMA receiver is required to use the same transmission schemes on different allocated subbands. The constraint, in fact, raises a challenge for designing an optimal scheduling algorithm for the subband scenario. To address such a special constraint, our approach is to classify users into different groups first. In the case of two NOMA users, there will be at most two clusters and we propose to use the probabilistic clustering method to determine the membership probability of each user. Particularly, we take into account not only the feedback subband CQIs but also the required power split factor to meet the SINR constraint as two main factors in calculating the size of each cluster and distance to them. The novel design thus allows us to classify users into a correct cluster that fits the special NOMA constraint. After having user clustered, we divide the scheduling problem into two sub-problems: user assignment and power allocation. Since both of the sub-problems appear in the fractional form, we introduce a parametric approach to transform them into non-fractional form and apply popular algorithms such as integer linear programming and non-linear programming algorithm like interior point method to solve the problem. Evaluation results show that our proposed algorithm based on the probabilistic clustering method has much better performance gain than the baseline method which uses the k-means clustering algorithm, with up to 45 percent gain.