基于深度强化学习的细粒度5G RAN切片功能复用映射与路由

DRL-Assisted Fine-Grained Function Placement and Routing with Reuse Scheme for 5G Network Slicing

在无线接入网(RAN)中,5G网络功能细粒度分割机制可以有效地缓解基带功能集中处理和传输带宽所带来的压力。但这就需要高效的资源管理方法,才能极大地避免细粒度功能单元(FU)部署所导致的网络资源和计算资源的浪费。主要研究了基于功能复用(FR)策略的细粒度功能部署和路由算法。首先使用混合整数线性规划(MILP)来解决上述问题,以此生成的最优解作为基准。在时间复杂度方面,混合整数线性规划模型无法在限定时间内解决大规模问题。由此,提出了一种深度强化学习(DRL)辅助的资源优化分配方法,用于高效地解决基于功能复用机制的FU基带功能部署和路由问题。从MILP仿真结果中可以看出,功能复用机制大大提升了资源利用率,验证其有效性;从DRL的仿真结果可见,DRL方法性能明显优于启发式算法,可以使资源成本显著降低,而且在时间复杂度上,也远远优于MILP方法,具有更高的可扩展性。

The fine-grained split is an effective way to balance the baseband processing centralization and optical bandwidth saving in radio access network(RAN).However,without efficient resource management measures,fine-grained unit(FU)deployment under fine-grained segmentation can result in a great waste of resources.This paper proposes a deep-reinforcement-learning-assisted resource allocation method for baseband function placement and routing with function reuse(FR)in 5G fine-grained unit architecture.Two novel mixed integer linear programming(MILP)formulations with or without the FR scheme are presented to generate the optimal solution.With respect to the time complexity,the MILP models cannot solve large-scale problems in an acceptable execution time.Therefore,the DRL-assisted method is proposed for the issue in the FU-based RAN.The simulation results show that MILP with FR outperforms MILP without FR.So the validity of FR is proved.The proposed DRL-based method with the FR scheme can achieve better performance with the minimized resource cost than the MILP method without the FR scheme and the heuristic method.Also,addressing the large-scale network scenario,the DRL-based method can achieve higher scalability than MILP.