基于深度强化学习的无线资源分配算法  

Wireless Resource Allocation Algorithm Based on Deep Reinforcement Learning

作　　者：张珍凤 李芳 ZHANG Zhenfeng;LI Fang(Shanxi Institute of Energy,Taiyuan 030600,China)

机构地区：[1]山西能源学院,山西太原030600

出　　处：《现代信息科技》2023年第17期8-14,共7页Modern Information Technology

基　　金：山西省基础研究计划(自由探索类)项目(202303021211285);晋中市科技重点研发计划(工业)项目(Y211019)。

摘　　要：为了在有限的无线资源条件下提供更高的信息传输速率,第五代移动通信(5G)引入多种高效的频谱复用技术,如终端直通技术(Device-to-Device,D2D)和非正交多址技术(Non-orthogonal Multiple Access Technology,NOMA)等。针对D2D网络,提出一种无监督的基于深度强化学习(Deep Reinforcement Learning,DRL)的信道和功率分配算法,解决了D2D用户信息传输速率最大化的问题。文章将该问题分解为信道分配和功率分配两个子问题,并分别用深度强化学习算法获得较优的信道和功率分配策略。实验仿真结果表明,基于DRL的资源分配算法相比传统的优化算法,具有较低的时间复杂度以及更好的实验性能,更加适用于动态无线网络中的资源管理。In order to provide higher information transmission rates under limited wireless resource conditions,a variety of efficient spectrum reuse technologies are introduced in the 5th Generation Mobile Communication(5G),such as Device-to-Device(D2D)and Non-orthogonal Multiple Access Technology(NOMA),etc.An unsupervised channel and power distribution algorithm based on Deep Reinforcement Learning(DRL)is proposed for D2D networks,which can solve the problem of maximizing the information transmission rate of D2D users.In this paper,the problem is divided into two sub-problems,channel allocation and power,and the better channel allocation strategy and power allocation strategy are obtained by deep reinforcement learning algorithm.Experimental simulation results show that the resource allocation algorithm based on DRL has lower time complexity and better experimental performance than the traditional optimization algorithm,and is more suitable for resource management in dynamic wireless networks.

关 键 词：D2D网络 资源分配 深度强化学习 

分 类 号：TN929.5[电子电信—通信与信息系统] TP18[电子电信—信息与通信工程]