eLAA中基于强化学习的动态帧配置算法

A Q-learning based dynamic frame configuration algorithm in eLAA networks

增强许可辅助访问(eLAA)/MulteFire能够在免授权频谱上同时支持上行(Uplink,UL)下行(Downlink,DL)传输。在免授权频段上,eLAA/MulteFire采用LBT(Listen Before Talk,LBT)机制接入信道,但该机制并不能避免与隐藏节点之间的传输冲突,从而导致时延加长或WiFi丢包。合理灵活的帧配置将减少这种传输冲突,降低WiFi传输时延,提高信道的接入概率。本文提出一种基于Q学习(Q-learning,QL)的动态上下行帧配置(DFC)机制。在这个机制中,基站被视为一个智能体,并将吞吐时间和公平性的不同组合定义为智能体状态,不同帧配置定义为智能体行为。智能体基于获取到的临近基站的帧配置和WiFi的平均传输时长,学习得到最优的帧配置。仿真结果表明,提出的基于QL的帧配置方法在确定一定公平性的前提下,极大提高了免授权频段的信道接入概率和吞吐量,降低了传输时延。

Enhanced License Assisted Access(eLAA)/MulteFire can support both Uplink(UL)and Downlink(DL)transmission on the unlicensed spectrum.In the unlicensed frequency band,eLAA/MulteFire uses the Listen Before Talk(LBT)mechanism to access the channel.But this mechanism cannot avoid the transmission collision of the eLAA and hidden WiFi Access Points(WAPs),and further lead to increase the transmission delay and loss of WiFi packets.Reasonable and flexible frame configuration will minimize the transmission conflict,reduce the WiFi transmission delay,and improve the access probability of the channel.Therefore,this paper proposes a Q-learning(QL)based UL/DL Dynamic Frame Configuration(DFC)mechanism.In the mechanism,base station is regarded as an agent,and different combinations of throughput time and fairness are defined as agent state.The different frame configurations are defined as agent behavior.The agent learns the optimal frame configuration based on the acquired frame configuration of the adjacent base BS and the average transmission time of WiFi.Simulation results show that the QL based frame configuration method can greatly improve the access probability and throughput of the unlicensed spectrum and reduce the transmission delay under the premise of certain fairness.