針對雲端無線存取網路所設計之聯合分群與功率分配機制

Abstract

雲端無線存取網路(Cloud radio access network, C-RAN) 藉由分散式遠端無線射頻單元(RRU) 可有效提升傳輸範圍。此網路架構能利用 分散式佈建與集中式的運算處理的優點，將分散式遠端無線射頻單元 盡可能佈建於使用者旁邊，使其能用較少的能源消耗去對抗通道效應 的衰減，而集中式的處理能根據整體網路的狀況進行資源分配，以提 升整體網路的頻譜及能源效益。然而，集中式的資源處理將導致此 架構的運算複雜度是相當可觀的，因此利用分群(clusterization) 的機制，將分散式遠端無線射頻單元及使用者分入數個群集，以降低整體 的運算複雜度。此外，本研究希望藉由聯合分散式遠端無線射頻單元 的分群與能源分配機制能有效的進行干擾抑制與減少能源消耗，進而 提升整個網路效能。此最佳化問題是個非凸優化問題，為了不將問題 進行相關轉換，本研究提出隨機性聯合分群與能源分配機制，進而取 得群集與能量分配的分配政策。此外，為了降低隨機性聯合分群與能 源分配機制的運算複雜度，進一步提出確定性聯合分群與能源分配機 制。此演算法先將問題轉換成凸優化問題並利用微分資訊去取得群集 與能量分配政策。再者，分群與能源分配利用聯合方式進行求解，使 的演算法產生較高的複雜度，因此提出確定性分離式分群與能源分配 機制，將此聯合的最佳化問題拆成分群與能源分配子問題進行求解， 以更進一步降低複雜度。考慮當雲端無線存取網路存在大規模的分散 式遠端無線射頻單元與使用者，將造成演算法複雜度提升，為了降低 此種情形的複雜度，提出漸近式的聯合分群與能源分配機制，利用隨 機矩陣定理與凸優化特性，進一步減少確定性聯合分群與能源分配機 制的運算複雜度。最後，本研究將針對所提出的演算法進行複雜度分 析之外，亦與現有的方法進行效能分析，所提出的演算法皆有較佳的 效能。

In this paper, the cloud radio access network (C-RAN) is considered to extend the transmission coverage via the distributed deployment of large scale remote radio units (RRUs). The benefits of C-RAN system deployment is distributed network with centralized management solution. The RRUs are more closer to the user equipments (UEs), power consumption can be reduced to overcome pathloss attenuation and enhance spectral and energy efficiency (EE). However, this type of structure can induce considerable computational loadings due to the centralized management mechanisms. To reduce the complexity incurred in the C-RAN architecture, the clusterization technique is designed to categorize those RRUs into several groups. For the purpose of enhancing EE as well as the consideration of computational complexity, the joint clusterization and power allocation schemes are proposed to obtain the better tradeoff under the quality-of-service (QoS) requirement for each UE. The optimization problem is modeled as mixed combinatorial problem. To obtain the near-optimal solution without any problem transformation, the stochastic joint clusterization and power allocation (S-JCPA) scheme is proposed to jointly solve the problem with cross-entropy (CE) algorithm. In addition, reducing the complexity incurs from stochastic process, the deterministic joint clusterization and power allocation (D-JCPA) schemes are also proposed. By converting the problem into convex problem and using derivation information to find the clusterization and power allocation policies. Moreover, this joint optimization problem can be divided into two sub-problems, each of which is solved by iterative algorithm. Therefore, the iterative times in D-JCPA scheme can be reduced. In addition, as large number of RRUs are deployed in the C-RAN system, the large random theorem with convex properties can be applied to derive asymptotic form for performing D-JCPA scheme. Finally, the complexity of the proposed method has been analyzed. Simulation results show that the proposed algorithms can provide better performance gain than the existing method.