Test And Optimization Of 5G Key Applications Performance Based On Open Communication Platform

Open wireless networks are of great importance to the convergence of the 5th generation mobile networks(5G)and vertical industry.It uses white box equipment to build flexible and reconfigurable networks based on software defined radio,software defined network and network function virtualization.Although there are many preliminary performance tests one open wireless communication platforms,most of the experiments focus on traditional applications such as file downloading and video transmission and there is lack of research on emerging 5G applications.Therefore,this paper designs a 5G service demonstration platform based on open wireless communication solution and evaluates and optimizes the performance of wireless virtual reality(VR)and video object detection services.The main contributions are as follows:Facing the problems of current open communication experimental platform,such as the lack of intelligence,the difficulty of performing joint optimization of multi-layer parameters and the lack of tests on 5G key applications,a novel demonstration platform is designed and established.In addition to open communication equipment and application servers,the platform also integrates an intelligent controller that locates between the application layer and the wireless network layer.The controller can collect wireless environment status and application performance via open interfaces provided by open base stations and applications,and it can configure various parameters in different layers as well.Moreover,the controller incorporates an endogenous intelligence module which facilitates the use of artificial intelligence(AI)models.With AI,user intention input can be transformed into proper parameters configurations and multi-layer parameters can be intelligently optimized.Based on the platform,wireless VR and video object detection services are tested and the impacts of network and application parameters on key performance metrics are evaluated and analyzed.Facing the difficulty in deriving the explicit relationship between application performance metrics and multi-layer parameters,and meanwhile in order to avoid the operation complexity caused by frequent interaction with real environments,this paper proposes a multi-layer parameter optimization approach based on offline deep reinforcement learning(DRL).The input state of the DRL model includes terminal status,such as signal noise ratio and retransmission rate,while each output action corresponds to a possible parameter configuration.To balance the use of radio resources and the overall performance of all users,the reward function takes into account the number of occupied radio resource blocks,user throughput and dedicated metrics of 5G applications.By extracting useful information from the multi-dimensional input information,the DRL model can make proper parameter configuration decisions in an online fashion.Based on the comparison with traditional DRL and random parameter configuration,the performance gain of the proposal is illustrated.Furthermore,by comparing the performance of joint parameter optimization with single layer parameter optimization,the advantage of the former setting is verified.In summary,this paper builds a demonstration platform for 5G key applications based on open communication equipment,and the performance of VR and video object detection services are tested on the platform.Meanwhile,a DRL based approach to the joint optimization of multi-layer parameters are developed,which shows an effective way to meet the differentiated requirements of 5G applications.