| With the continuous development of wireless communication technology and the large-scale use of smart terminals,many video playback software have been proposed.Benefiting from the popularity of various short video software such as Douyin,Weishi,and Kuaishou,as well as the popularity of various live broadcast platforms,the video traffic data on the terminal side has exploded.Transmission of video data generally requires a large network bandwidth,however,with the advent of the big data era,wireless spectrum resources are becoming more and more tense.In addition,wireless channels have inherent temporal and spatial changes and are vulnerable to interference.Therefore,ensuring the quality of experience(QoE)in the video transmission process has become a challenging task now.Due to the drastic changes in throughput during video transmission and the large number of network states,traditional adaptive algorithms are difficult to match the dynamically changing network environment.Based on the current development of streaming media technology,in view of the current problems faced by video transmission,the main content of the thesis has the following three points:(1)Aiming at the problem that the throughput changes dramatically and the user's QoE decreases in the single path transmission process,an adaptive control method of video bit rate based on deep reinforcement learning is proposed.First,by modeling the dynamic changes of the buffer model,the video bit rate set is obtained,and then the video quality,the buffer occupancy,and the rebuffering event are set as the QoE metrics,and used as the reward function of deep Q-Network algorithm to guide the optimization direction of the algorithm.The simulation results show that compared with the ordinary DQN algorithm,the proposed algorithm increases the buffer occupancy by 20%,and the video rebuffering time is reduced by 25%,which is effective improve the user's QoE.(2)Aiming at the problem that the shunt decision can not adapt to the change of network condition and lead to the inconsistent arrival time of packets in the process of multi-path transmission,an adaptive shunt decision method based on data-driven and cybernetics for heterogeneous link data transmission is proposed.First,the sending end decides the video data distribution ratio of each subpath according to the A3 C algorithm,and transmits it to the receiving end in parallel through different network connections.Then the dynamic change model of the receiver buffer is improved to obtain the set of video bit rate that meets the requirements.Finally,the rate control module is designed based on the proportional difference controller by using the control theory to realize the self-adaptive video rate.The simulation results show that the proposed algorithm increases the network throughput by23% under different distribution ratios,and increases the buffer occupancy by 11% when the requested bit rate is high,and the video rebuffering time decreases by 58%.(3)Aiming at the parallel transmission scenario in the heterogeneous wireless network environment,a set of multi-terminal collaborative video transmission control system is built.Through the functional interaction between each module of the system,the video multi-path transmission function is realized,and the adaptive bit rate decision and adaptive shunting decision are realized by using different characteristic parameters and deep reinforcement learning algorithm.The test results show that the system can effectively aggregate the transmission performance of multiple terminals.Compared with single-terminal transmission,the throughput can be increased by up to 4 times,and the rebuffering time can be reduced by 82%,Thus,the smooth playback of high-definition video can be achieved effectively,and the user's QoE can be improved. |
PDF Full Text Request