應用於雲端無線接取網路之邊際快取管理的深度學習與最小成本流量演算法

在本篇論文當中，我們探討如何設計在雲端無線接取網路的邊際快取管理。我們設定基頻單元可以預測行動使用者要求內容傳遞的分布情況並且決定哪些內容會被無線寬頻頭端快取。當無線寬頻頭端所快取的內容有服務到使用者時，整個網路會得到獎勵，獎勵的多寡取決於使用者的評價，所以我們想要設計出一個好的快取策略使得獲取獎勵最大化。我們利用一種機器學習的方法叫做回響狀態網路，使得基頻單元能夠預測使用者要求內容的分布。接著，設計邊際快取放置矩陣可以轉換成最小流量問題，而且我們進一步提出能讓任意兩個距離相近的無線寬頻頭端共同決定快取擺放的演算法，這樣一來能使邊際快取內容多樣化。在模擬結果中，顯示了提出的快取演算法能有效降低回程網路的資源消耗。

關鍵字

雲端無線接取網路；快取管理；深度學習

並列摘要

In this thesis, the problem of edge cache management is studied for cloud radio access networks (cloud-RANs). In studied model, the baseband units can predict the content request distribution and make the decision of which contents to cache at remote radio heads (RRHs). The problem is formulated as an optimization problem to maximize the reward which the network get if an RRH provides service to mobile users. To solve this problem, we firstly use the machine learning framework of echo state networks (ESNs) such that the BBUs can predict each user’s content request distribution. Then, we model the design of cache placement matrix as a minimum cost flow problem. Furthermore, we propose a refined caching algorithm with additional cooperation between any two RRHs which are close to each other. Simulation results show that the proposed algorithm can significantly reduce resource consumption in back-haul network.