DRL-Based Channel and Latency Aware Radio Resource Allocation for 5G Service-Oriented RoF-MmWave RAN
DRL-Based Channel and Latency Aware Radio Resource Allocation for 5G Service-Oriented RoF-MmWave RAN
复制标题
DOI：
10.1109/jlt.2021.3093760
复制
发表时间：
2021-09-15
影响因子：
4.7
通讯作者：
Chang, Gee-Kung
中科院分区：
文献类型：
--
作者：
Shen, Shuyi;Zhang, Ticao;Chang, Gee-Kung
提出并评估了一种基于深度强化学习（DRL）的信道感知和时延感知无线电资源分配算法。该方案旨在优化5G时代面向服务的多用户毫米波（mmWave）无线接入网（RAN）的上行调度。在DRL系统中，通过多种统计模型实现多个应用流程，并设计系统的关键功能模块，以反映面向服务的ran的运行和需求。特别是，通过实验收集了系统中使用的毫米波信道特性，并通过具有动态信道变化的光纤无线电（RoF）毫米波试验台进行了验证。结果表明，所提出的DRL算法可以自适应信道变化，与传统的单规则方案相比，平均奖励提高至少12%，同时提供了误码率和延迟性能的联合改善。
A channel and latency aware radio resource allocation algorithm based on deep reinforcement learning (DRL) is proposed and evaluated. The proposed scheme aims to optimize the uplink scheduling for service-oriented multi-user millimeter wave (mmWave) radio access networks (RAN) in the 5G era. In the DRL system, multiple application flows are implemented with various statistical models and the key function modules of the system are designed to reflect the operation and requirements of service-oriented RANs. In particular, the mmWave channel characteristics utilized in the system are collected experimentally and verified via a radio-over-fiber (RoF)-mmWave testbed with dynamic channel variations. Results show that the proposed DRL algorithm can operate adaptively to channel variations and achieve at least 12% average reward improvement compared to conventional single-rule schemes, providing joint improvement of bit error rate and latency performance.