针对现有的码率自适应(adaptive bitrate,ABR)算法存在控制规则简单,不能有效提升用户体验质量(quality of experience,QoE),提出一种基于元学习的LABR(reinforcement learning based ABR)算法。采用策略梯度训练策略网络,利用元学习(me...针对现有的码率自适应(adaptive bitrate,ABR)算法存在控制规则简单,不能有效提升用户体验质量(quality of experience,QoE),提出一种基于元学习的LABR(reinforcement learning based ABR)算法。采用策略梯度训练策略网络,利用元学习(meta-learning)方法学习基线(baseline)函数来减少因网络吞吐量差异产生的方差,进一步提高模型的准确性和鲁棒性;通过在策略函数中加入熵损失方法提高累计期望奖励值。实验结果表明,LABR算法具有泛化性与鲁棒性,能有效提高用户的视频体验质量。展开更多
为解决多客户端的带宽资源分配问题,提高用户体验质量(quality of experience,QoE),建立多客户端视频流的体验质量优化框架。针对已有视频流算法在多客户端领域的缺陷,基于模型预测控制算法提出一个多客户端带宽动态调度算法,根据每个...为解决多客户端的带宽资源分配问题,提高用户体验质量(quality of experience,QoE),建立多客户端视频流的体验质量优化框架。针对已有视频流算法在多客户端领域的缺陷,基于模型预测控制算法提出一个多客户端带宽动态调度算法,根据每个客户端的带宽预测情况对它们进行动态资源分配,通过提高带宽利用率进而提升总体用户QoE。在HSDPA网络带宽轨迹上的仿真结果表明,相比各客户端平均带宽分配方式,优化方法在总体用户体验质量上提升42.6%以上,相比最新的Minerva方案提升了7.8%。展开更多
随着互联网技术的快速发展以及智能设备的普及,基于HTTP的动态自适应流媒体(Dynamic Adaptive Streaming over HTTP,DASH)业务发展迅速.但在带宽受限网络中,大规模用户的视频请求,将会加重网络负载,严重影响网络带宽资源的有效利用,同...随着互联网技术的快速发展以及智能设备的普及,基于HTTP的动态自适应流媒体(Dynamic Adaptive Streaming over HTTP,DASH)业务发展迅速.但在带宽受限网络中,大规模用户的视频请求,将会加重网络负载,严重影响网络带宽资源的有效利用,同时用户码率调节缺乏全局协调控制机制,容易造成网络拥塞.针对软件定义网络中的DASH视频传输业务,将视频业务提供商长期平均收益最大化作为优化目标,设计并实现了基于神经元动态规划的DASH视频路由和用户码率调节联合决策算法.最后,通过在Mininet平台上建立SDN(Software-Defined Networking)网络环境并进行对比实验,我们验证了本文提出的联合决策算法能够提高网络带宽资源利用率,最大化DASH视频业务提供商长期平均收益.展开更多
码率自适应(Adaptive BitRate,ABR)算法是视频客户端提高用户体验质量(Quality of Experience,QoE)的一种有效途径.针对现有ABR算法存在频繁缓冲、视频卡顿、画质较低和网络吞吐量预测不准确等问题,本文提出一种基于深度强化学习的码率...码率自适应(Adaptive BitRate,ABR)算法是视频客户端提高用户体验质量(Quality of Experience,QoE)的一种有效途径.针对现有ABR算法存在频繁缓冲、视频卡顿、画质较低和网络吞吐量预测不准确等问题,本文提出一种基于深度强化学习的码率自适应(Deep Reinforcement Learning based ABR,DRLA)算法.DRLA用实际网络带宽数据训练神经网络,通过收集客户端缓冲区占用率和网络吞吐量向视频服务器请求最佳码率的视频.首先,DRLA用基线函数方法优化损失函数L,用熵随机探索方法防止损失函数局部收敛;其次利用约束条件限制新旧策略的散度更新幅度提高算法的鲁棒性;最后通过置信域(trust region)优化找到最优策略,使得QoE达到最优.与现有ABR算法对比的实验结果表明:DRLA减少了训练时间,能进一步提高算法的鲁棒性和用户的QoE,并在实际环境下验证了算法的有效性.展开更多
Dynamic adaptive streaming over HTTP(DASH)can adaptively select the appropriate video bitrate for mobile users.Mobile edge computing(MEC)scenario is of great benefit to improve the performance of mobile networks by pr...Dynamic adaptive streaming over HTTP(DASH)can adaptively select the appropriate video bitrate for mobile users.Mobile edge computing(MEC)scenario is of great benefit to improve the performance of mobile networks by providing computing and storage capabilities.And the utilization of spectrum resources can be improved by multicast transmission,but the performance of the multicast transmission will be directly affected by the selected grouping algorithm and resource allocation algorithm.In order to improve the quality of experience(QoE)of video users in the 5G MEC scenario,this paper proposes a QoE-driven DASH multicast scheme,which mainly covers the grouping algorithm and the adaptive bitrate(ABR)algorithm.First of all,we take the optimized target QoE as the basis for grouping and propose an adaptive grouping algorithm that can dynamically adjust the grouping results.Besides,we design a joint optimization ABR algorithm based on the prediction of QoE,which comprehensively considers the process of resource allocation and bitrate decision-making based on the prediction of QoE of video segments in a certain forward-looking field of view.The simulation results show that the proposed DASH multicast scheme performs well in QoE and fairness.展开更多
文摘针对现有的码率自适应(adaptive bitrate,ABR)算法存在控制规则简单,不能有效提升用户体验质量(quality of experience,QoE),提出一种基于元学习的LABR(reinforcement learning based ABR)算法。采用策略梯度训练策略网络,利用元学习(meta-learning)方法学习基线(baseline)函数来减少因网络吞吐量差异产生的方差,进一步提高模型的准确性和鲁棒性;通过在策略函数中加入熵损失方法提高累计期望奖励值。实验结果表明,LABR算法具有泛化性与鲁棒性,能有效提高用户的视频体验质量。
文摘为解决多客户端的带宽资源分配问题,提高用户体验质量(quality of experience,QoE),建立多客户端视频流的体验质量优化框架。针对已有视频流算法在多客户端领域的缺陷,基于模型预测控制算法提出一个多客户端带宽动态调度算法,根据每个客户端的带宽预测情况对它们进行动态资源分配,通过提高带宽利用率进而提升总体用户QoE。在HSDPA网络带宽轨迹上的仿真结果表明,相比各客户端平均带宽分配方式,优化方法在总体用户体验质量上提升42.6%以上,相比最新的Minerva方案提升了7.8%。
文摘码率自适应(Adaptive BitRate,ABR)算法是视频客户端提高用户体验质量(Quality of Experience,QoE)的一种有效途径.针对现有ABR算法存在频繁缓冲、视频卡顿、画质较低和网络吞吐量预测不准确等问题,本文提出一种基于深度强化学习的码率自适应(Deep Reinforcement Learning based ABR,DRLA)算法.DRLA用实际网络带宽数据训练神经网络,通过收集客户端缓冲区占用率和网络吞吐量向视频服务器请求最佳码率的视频.首先,DRLA用基线函数方法优化损失函数L,用熵随机探索方法防止损失函数局部收敛;其次利用约束条件限制新旧策略的散度更新幅度提高算法的鲁棒性;最后通过置信域(trust region)优化找到最优策略,使得QoE达到最优.与现有ABR算法对比的实验结果表明:DRLA减少了训练时间,能进一步提高算法的鲁棒性和用户的QoE,并在实际环境下验证了算法的有效性.
基金the National Key R&D Program of China under Grant 2020YFA0711400the National Science Foundation of China under Grant 61673360the CETC Joint Advanced Research Foundation under Grant 6141B08080101.
文摘Dynamic adaptive streaming over HTTP(DASH)can adaptively select the appropriate video bitrate for mobile users.Mobile edge computing(MEC)scenario is of great benefit to improve the performance of mobile networks by providing computing and storage capabilities.And the utilization of spectrum resources can be improved by multicast transmission,but the performance of the multicast transmission will be directly affected by the selected grouping algorithm and resource allocation algorithm.In order to improve the quality of experience(QoE)of video users in the 5G MEC scenario,this paper proposes a QoE-driven DASH multicast scheme,which mainly covers the grouping algorithm and the adaptive bitrate(ABR)algorithm.First of all,we take the optimized target QoE as the basis for grouping and propose an adaptive grouping algorithm that can dynamically adjust the grouping results.Besides,we design a joint optimization ABR algorithm based on the prediction of QoE,which comprehensively considers the process of resource allocation and bitrate decision-making based on the prediction of QoE of video segments in a certain forward-looking field of view.The simulation results show that the proposed DASH multicast scheme performs well in QoE and fairness.