Intelligent Scheduling Research Based On Reinforcement Learning In The Spectrum Aggregation

With the development of the society, people have a higher demand requirement for communication. The emergence and popularity of intelligent terminal bring new challenges to the wireless communication system. In order to meet the need of the mobile data requirements, a new generation of mobile communication technology LTE-Advanced(LTE- A) obtains greater spectrum bandwidth by using carrier aggregation technology. In greater spectral bandwidth, how to maximize the use of spectrum resources is the one of the problems needed to be solved. This paper introduces two kinds of reinforcement learning algorithm to improve the utilization of the communication system in the carrier aggregation.This paper introduces the background of the carrier aggregation and reinforcement learning, then focus on single intelligent entity of reinforcement learning system model,reinforcement learning process and multiple intelligent entities joint reinforcement learning model. This paper derives the multi-agent reinforcement learning formulas,which is the theoretical basis for the allocation of resources in the carrier aggregation.This paper proposes using reinforcement learning to allocate resource inside the cell,and between cells and cells.When allocating resource between cells and cells, this paper uses the gradient reinforcement learning algorithm to optimize the resource allocation between cells and cells. Gradient reinforcement learning algorithm is combined with the current load of each cell and carrier characteristics, to choose the appropriate resources for each cells.The simulation result shows that, compares with the traditional fixed allocations and dynamic allocations, using the gradient distribution of reinforcement learning algorithm is better in using of resources, improving the spectrum efficiency, reducing the user satisfaction. Also, reward in gradient reinforcement learning algorithm can be reset according to the need of the system.When allocating resource inside the cell, this paper uses Q-learning algorithm, to optimize the resource allocation inside the cell. This paper uses Q-learning algorithm to learn user business model. According to the user business, user position and the characteristics of carrier resource block, Q-learning algorithm set different rewards for different allocations. The simulation result shows that, compared with the traditionalgreedy algorithm, Q learning allocation is better in making the users get higher utility value, reducing the user’s business blocking rate and improving overall system resource utilization.