一种基于预留-重用联合的C-V2X通信Q学习型半持续调度算法

A Reservation and Reuse Combined Q-learning Semi Persistent Scheduling for C-V2X Communication

第3代合作伙伴计划(3GPP)模式4提供了一种直连通信的接入方式,以支持车联网C-V2X应用。然而V2X网络动态变化的业务负载和车辆移动性导致信道质量不稳定,加剧了传输过程中分组碰撞问题。为了满足高可靠低时延的V2X通信需求,该文针对动态负载环境下的高效分布式资源分配策略,提出一种预留-重用联合的Q学习型半持续调度(RRC-QSPS)算法。该文首先对模式4中现有的半持续调度(SPS)算法分组碰撞问题进行理论建模,分析了影响碰撞概率的关键参数,继而提出了动态业务环境下车辆智能体的强化Q学习模型,建立包括预留-重用联合的动作空间与Q目标函数,并通过-贪心算法求解,智能决策动态负载环境下无线资源的预留与重用。仿真对比结果表明,相比于已有的Lookahead-SPS优化算法,RRC-QSPS算法在高速高负载场景下分组接收率提高了7%,数据包更新时延降低了10%。

The 3rd Generation Partnership Project(3GPP) mode 4 provides a direct communication mode to support the C-V2X applications. However, the dynamic traffic load and vehicle mobility lead to uncertainty of channel quality with serious packet collision problem. In order to meet the demand of ultra reliability and low latency of V2X communication. A Reservation-Reuse Combined Q-learning Semi Persistent Scheduling(RRCQSPS) algorithm is proposed for efficient distributed resource allocation in dynamic load environment. Firstly,the theoretical model of the collision probability of Semi Persistent Scheduling(SPS) algorithm is built. Then,the Q-learning model of vehicle agent in dynamic load environment is proposed with the reservation-reuse combined action and Q function. By using ε-greedy method, the optimal reservation and reuse of wireless resources in dynamic load environment can be solved. The simulation results show that compared with the existing Lookahead-SPS optimization algorithms, the packet reception ratio of RRC-QSPS is improved by 7%and the update packet delay is reduced by 10% in high speed and high load scenarios.