A Deep Reinforcement Learning Based Mechanism For Cell Outage Compensation In 5G UDN

As one of the key technologies of 5G,UDN has a more complex structure and a large number of deployed nodes.If a failure occurs and can't be alleviated its effect in time,it will lead to a significant drop in network performance.The self-healing function of SON can detect the network and recover failures autonomously.As a key factor of self-healing,cell outage compensation can automatically adjust network parameters to eliminate the impact of failures.Therefore,COC problem in UDN is very important.However,the existing works mainly focus on COC in 4G scenarios and user needs have not been considered in detail.Besides,there are few researches on COC in 5G scenarios.DRL provides better solutions to complex problems,with significant results in multiple areas,including self-organizing networks.But there are fewer applications for cell outage compensation in UDN.Therefore,this topic focuses on the research of COC.For two representative UDN scenarios,Ultra-Dense HetNets and Massive IoT Environments,this paper proposes a framework based on DRL to solve COC problems in UDN,which improves network performance and reduces network operation cost.Aiming at the high bandwidth users in Ultra-Dense HetNets,this paper proposes a COC method for user QoS guarantee in UDN.First,according to the characteristics of Ultra-Dense HetNets,a COC model for user QoS guarantee is designed.Among them,the user connection relationship and BS transmission power allocation are taken as decision variables,the sum of transmission rates of all users is taken as the optimization goal,and the demand of mobile users is taken as the constraint.On this basis,an Ultra-Dense HetNets COC framework based on DQN is proposed.The K-means++clustering algorithm is used to determine the user connection relationship,and DQN algorithm is used to allocate the BS transmission power to the compensation users to maximize the sum of transmission rates of all users.Finally,simulation results show that the proposed method has faster training and convergence speed,guarantees the user QoS,and maximizes the sum of transmission rates of all users.Aiming at the ubiquitous users with low bandwidth demand in Massive IoT Environments,this paper proposes a COC method for network coverage optimization in UDN.First,according to the characteristics of Massive IoT Environments,a COC model for network coverage optimization is designed.Among them,the user connection relationship,BS antenna downtilt and BS transmission power allocation are taken as decision variables,and maximizing the BS connectivity while meeting the demand of each compensation user is taken as the optimization goal.On this basis,a Massive IoT Environments COC framework based on DDQN is proposed.The K-means++clustering algorithm is used to determine the user connection relationship,and DQN algorithm is used to adjust BS antenna downtilt and allocate the BS transmission power to the compensation users to maximize the BS connectivity and meet the demand of each compensation user.Finally,compared with DQN algorithm,simulation results show that the proposed method has faster convergence speed,and the result is closer to the optimal solution which shows that the over estimation problem of the DQN algorithm is solved.Besides,it realizes the goal of maximizing the BS connectivity and meeting the demand of each compensation user.In a word,this paper proposes and implements a DRL based mechanism for COC in 5G UDN.According to characteristics of different UDN scenarios,we propose the corresponding models and frameworks,which provides a new way to solve the problem of COC in UDN scenarios.