基于Q强化学习的AVB数据转发时延模型研究

为了提高音视频桥接技术(Audio/Video Bridge,AVB)数据时延分析结果的正确性,以AVB交换机处转发的数据为研究对象,在经典的AVB协议模型中融入Q强化学习理论,构造时延数学模型.在完成AVB协议模型和Q强化学习理论研究基础上,提出更加符合实际的AVB交换机数据转发策略,结合车载以太网数据转发时延影响因...

Research on AVB Data Forwarding Delay Model Based on Q Learning

A delay mathematical model was constructed, which combined the classical AVB (Audio/Video Bridge) protocol model and Q Learning with the objective of AVB switch's data to improve the compactness of AVB data delay analysis results. Based on the study of the AVB protocol model and Q reinforcement learning theory, a more practical forwarding strategy was proposed for AVB switch data. The delay of each frame and the average delay of all types of traffic were analyzed based on factors influencing the data forwarding delay of automotive Ethernet. To verify the correctness of the established mathematical model, a simulation test bench of automotive Ethernet was built by combining Intrepid's RAD_Galaxy and the samples of the surround view system, and the experimental results were compared and analyzed based on Wireshark analysis. The simulation of the surround view system and the testing results of simulation test bench indicate the following: under a given condition, the average delay of class A flow is 41.36 μs, and the maximum deviation is 3.17%; the average delay of class BE flow is 103.53 μs, and the maximum deviation is 2.01%; the average delay of class BE flow is 141.99 μs, and the maximum deviation is 3.14%. At the same time, under different loads of BE flow, the delay performances of all types of data are also different. When the link load increases, the average delay of A and B data frames also increases, whereas the average delay of BE data frames does not change significantly. However, when the load exceeds the bandwidth, the average delay of the BE data frames becomes very large. That is, the protocol guarantees the transmission quality of data frames A and B. In addition, the maximum deviation of the delay analysis based on this mathematical model of the simulation test bench is 3.11%, which verifies the reliability of the mathematical model. Therefore, the model can effectively evaluate and guide improvements in network performance.