Research On Video Live System Based On Perception And Prediction

With the popularity of wireless communication technologies,spectrum resources are becoming increasingly tight.How to improve the efficiency of spectrum utilization has become a concern of both academic and industrial circles.As a new and popular multimedia form in recent years,live video has been loved by young people.However,the live video system needs to work in a high-rate,low-latency,low-interference wireless environment,which poses a serious contradiction with the problem of tight spectrum resources.Spectrum resources is prone to spectrum congestion,and spectrum sensing and dynamic spectrum access are an effective way to solve spectrum congestion problems.By using cognitive radio or deep learning,dynamic sensing and access to the spectrum can be achieved to alleviate spectrum congestion.This thesis takes video live broadcast system as the research scene,and deeply studies the dynamic spectrum access technology based on spectrum sensing based on cognitive radio technology and spectrum prediction based on deep learning.The main work of the thesis is as follows:Firstly,in order to solve the problem of spectrum congestion using dynamic spectrum access technology in live video scene,this thesis designs a video live system based on USRP RIO software defined radio platform.Firstly,according to the environmental conditions of indoor communication,the framework and components of the live video system are designed.The system is based on a centralized network and includes information fusion centers,femto cells and users.Then,the functional and physical implementation of each component is explained.Finally,the workflow of the live video system is illustrated with reference to the system flow chart.Secondly,in order to study the use of cognitive radio technology to solve the spectrum congestion problem,this thesis combines the physical layer interference coordination with the application layer interference coordination,and uniformly allocates the spectrum resources through the system control module,so that the unlicensed users can use the idle spectrum resources.Firstly,the difference between the system composition and the system flow of the live video system based on double-layer interference coordination and the basic system is clarified.Then,the core strategy of the system control module is described in detail: the fusion strategy and the scheduling strategy.Through the core policy,the system can select idle spectrum resources according to the perceived spectrum,and can automatically adjust resources according to QoE.So the unlicensed users can use idle spectrum resources without affecting authorized users.Finally,through experimental results and data analysis,it is verified that in the live video system based on double-layer interference coordination,unauthorized users can efficiently use idle spectrum resources.At last,in order to solve the problem of spectrum congestion using depth learning-based spectrum prediction,this thesis uses long short-term memory neural network for frequency prediction and uses the prediction result to guide the video live system for resource scheduling,so that unauthorized users can find idle spetrum resources in the wireless environment.First,the system composition and system flow of the live video system based on deep learning prediction are described.Secondly,the characteristics of the slot-by-slot scene and time series data are illustrated.Then,the difference between long short-term memory networks and traditional cyclic neural networks is described.Finally,through experimental results and analysis,it is verified that the long short-term memory networks successfully predict the trend of spectrum data,and reduce the system delay and load of the live video system.