Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/73494
Title: | 以圖論與深度強化學習網路的動態資源分配並將使用者分群來實現基地台干擾協調 Graph based and Deep Q-Network based Adaptive Resource Allocation with User Grouping on ICIC |
Authors: | Chien-Hao Lee 李建皜 |
Advisor: | 魏宏宇(Hung-Yu Wei) |
Keyword: | 基地台干擾協調,圖論,深度強化學習網路,資源分配,功率控制,使用者分群, Inter-Cell Interference Coordination (ICIC),Graph Theory,Deep Q-Network (DQN),Resource Allocation,Power Control,User Grouping, |
Publication Year : | 2018 |
Degree: | 碩士 |
Abstract: | 基地台干擾協調,旨在於減輕細胞邊緣使用者所受到的干擾影響。過去有許多論文,提出了各種基地台干擾協調的方法,我們在這些論文之中發現了兩個主要缺陷。其一是部分論文僅針對系統整體總吞吐量進行最佳化,對於細胞邊緣使用者的表現並無顯著的提升。另一是部分論文沒有考慮到基地台干擾協調的設定必須維持一段較長的期間,這一點使得這些論文提出的方法難以在真實系統中實作。我們參考一些近年來相關的論文,設計了一個圖論的基地台干擾協調方法,以及一個深度強化學習網路的基地台干擾協調方法。這兩種方法分別包含一個資源分配最佳化演算法與一個使用者分群演算法。在圖論的基地台干擾協調方法中,我們設計了一個廣度優先搜尋資源分配最佳化演算法,以及一個加強細胞邊緣使用者分群演算法。在深度強化學習網路的基地台干擾協調方法中,我們設計了一個深度強化學習網路,以及一個平衡使用者類別分群演算法。根據模擬結果,使用加強細胞邊緣使用者分群演算法,能夠有效提升廣度優先搜尋資源分配最佳化演算法的表現;而採取平衡使用者類別分群演算法,則能夠使深度強化學習網路有更佳的結果。在使用者數量少時,深度強化學習網路的基地台干擾協調方法擁有最佳的表現;而在使用者數量多時,圖論的基地台干擾協調方法表現更為出色。整體而言,不論是在使用者數量少,或是使用者數量多的情況下,本篇論文提出的兩種方法都能夠顯著提升基地台干擾協調的效果。 Inter-Cell Interference Coordination (ICIC) is aimed at mitigate interference at cell-edge users. A variety of ICIC methods are proposed in previous literature. However, we find two main drawbacks in those works. Some of them can maximize the overall system throughput but do not have enough enhancement of the performance at cell-edge users. Some of them do not consider that ICIC configuration should be effective in a longer period, which is the issue of real-world implementation. To overcome these two drawbacks, we propose several algorithms for a centralized dynamic ICIC scheme. Inspired by some recent works, we design a graph based ICIC scheme and a Deep Q-Network (DQN) based ICIC scheme. Each of them includes a resource optimization algorithm and a user grouping algorithm. The graph based ICIC scheme consists of Breadth-First Search Resource Optimization Algorithm (bfsOpt) and Edge-Enhanced User Grouping (EEUG). The DQN based ICIC scheme is made up of a DQN and Type-Balanced User Grouping (TBUG). Simulation results show that EEUG is the optimal user grouping algorithm for bfsOpt and TBUG is the optimal user grouping algorithm for the DQN. The DQN based ICIC scheme has the best performance when the number of users is low, while the graph based ICIC scheme has better performance when the number of users is high. Overall, the two proposed schemes outperforms the benchmarks in both sparse and dense user distribution. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/73494 |
DOI: | 10.6342/NTU201801335 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 電機工程學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-107-1.pdf Restricted Access | 1.9 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.