Deep Learning-based Resource Allocation For Broadband Multimedia Service In 5G C-RAN

Broadband multimedia services have the characteristics of strong business effectiveness,a variety of services and a variety of scenarios.With the increase of users' requirements for business service experience,multimedia services have higher network bandwidth requirements.However,the growth of network bandwidth lags behind the increased bandwidth requirements.It is a challenge for providing better multimedia services to users in existing networks.In order to effectively improve the service quality of broadband multimedia services,5G have proposed the further enhanced multimedia broadcast multicast service(FeMBMS),which uses cellular networks and local point-to-multipoint transmissions to process a large number of concurrent multimedia services.Under the dense networking requirements,C-RAN has become an important network architecture for 5G networks to carry broadband multimedia services.Different from the previous multimedia services transmitted by dedicated facilities or resources,5G networks will arrange for multimedia services and other network services with very different characteristics to share wireless network infrastructure,and allocate resources as needed to make full use of network resources.Effective use of 5G C-RAN network resources to meet the broadband multimedia service transmission needs in a timely manner.It is necessary to solve the problem of unable to accurately obtain real-time multimedia broadcast traffic demand during the allocation of network resources.The resource allocation for multimedia services has insufficient energy efficiency considerations and resource allocation methods.It is inefficient to deal with large-scale resource optimization problems and cannot meet real-time requirements.Solving the above problems is the key to ensuring the quality of multimedia services and improving the comprehensive benefits of 5G networks.This paper conducts research on this issue and proposes a deep learning-based broadband multimedia service resource allocation method to effectively solve the above problems.Aiming at the problem that the dynamic change of user needs may bring uncertainty to resource allocation,a user traffic demand prediction method based on Long Short Term Memory(LSTM)is proposed.The user traffic demand includes the number of users And the farthest user.This method first analyzes the influencing factors of multimedia user traffic demand,secondly models user traffic data based on the LSTM model,predicts user traffic demand,and finally performs simulation experiments based on actual traffic data of the operator network in a certain city.It shows that compared with traditional traffic prediction models such as Holt-Winters and ARIMA,the LSTM model can more accurately predict the number of users and the worst user distance.When the time span is 7 days,the prediction results are the best.Aiming at the problems of incompatible unicast and multicast resource allocation and insufficient consideration of energy efficiency,a method for allocating resources for broadband multimedia services based on deep reinforcement learning is proposed.This paper first constructs a mathematical model for resource allocation of broadband multimedia services in a 5G CRAN network,and defines the action space,state space,and reward rules of deep reinforcement learning based on it,and provides a resource allocation algorithm flow based on the deep reinforcement learning framework.The simulation experiment results show that compared with the traditional resource allocation method,the energy efficiency index value of the proposed method under different numbers of multicast users is about 6%lower on average,and the transmit power index is about 20%lower,indicating that the method has excellent resource allocation Performance.