Recently, fog-computing-based radio access networks (F-RANs) have been conceptualized to provide high quality of experience (QoE) for adaptive bit rate (ABR) streaming, where additional computing capacity is supplemented on fog nodes to facilitate complicated cross-layer optimization (i.e., joint bit rate selection and radio resource allocation). However, finding an optimal global solution with acceptable complexity is still infeasible by the conventional optimization methods. In this work, we propose an artificial intelligence (AI) aided joint bit rate selection and radio resource allocation scheme referred to as iABR, which provides a new vision for handling the over-complicated optimization in F-RANs. Based on multi-agent hierarchy deep reinforcement learning, the proposed iABR can dynamically allocate radio resource and select bit rate in a multiuser scenario, by perceiving the network environment and clients' player information. Moreover, long short-term memory (LSTM) is employed by the iABR algorithm, which enables accurate prediction of the change of channel quality by learning the history of the wireless channel. Hence, iABR is able to adjust the action policy in advance to accommodate the future channel quality for avoiding bit rate fluctuation. According to the experimental results, the iABR exhibits higher QoE in terms of high average bit rate, low rebuffering ratio, and average bit rate variance. Last but not least, the QoE performance of all the clients are fairly guaranteed by the iABR algorithm, enhancing the practicality of AI-driven F-RANs for multimedia service delivery.

中文翻译：

---

F-RAN上的自适应视频流的人工智能辅助联合比特率选择和无线电资源分配

最近，基于雾计算的无线电接入网络（F-RAN）已被概念化，可为自适应比特率（ABR）流提供高质量的体验（QoE），其中雾节点上增加了额外的计算能力，以简化复杂的交叉层优化（即联合比特率选择和无线电资源分配）。然而，通过传统的优化方法仍然难以找到具有可接受的复杂度的最优全局解决方案。在这项工作中，我们提出了一种称为iABR的人工智能（AI）辅助联合比特率选择和无线电资源分配方案，它为处理F-RAN中过于复杂的优化提供了新的视野。基于多主体层次的深度强化学习，通过感知网络环境和客户端的播放器信息，提出的iABR可以在多用户情况下动态分配无线电资源并选择比特率。此外，iABR算法采用了长期短期记忆（LSTM），通过学习无线信道的历史记录，可以准确预测信道质量的变化。因此，iABR能够预先调整操作策略以适应未来的信道质量，从而避免比特率波动。根据实验结果，iABR在高平均比特率，低重新缓冲率和平均比特率变化方面表现出更高的QoE。最后但并非最不重要的一点是，iABR算法可以相当保证所有客户端的QoE性能，从而增强了AI驱动的F-RAN用于多媒体服务交付的实用性。