Joint Power Channel Allocation Algorithm Based On DRQN

This paper mainly studies the power channel resource allocation algorithm under the integrated information network.At present,5G mobile communication system is in the initial stage of large-scale commercial use,and many key technologies still need to be enhanced and perfected,which cannot meet the communication requirements in some specific scenarios.It will be the main research direction of the next generation of mobile communication technology to make human beings have the integrated information network which is closer to the goal of "anyone in any place and anyone in any form of communication".The users in the integrated information network are complex,highly dynamic and sensitive to time delay,so it is necessary to design a new wireless resource allocation method to allocate power and channel resources for the users in the integrated information network.Deep reinforcement learning algorithms have made remarkable achievements in dealing with high-dimensional resource allocation problems and are expected to become the supporting technology in the field of resource allocation in integrated information networks.In this paper,a single cluster of the integrated information network will be modeled,and the deep loop Q network algorithm of deep reinforcement learning algorithm will be used to solve the problem of resource allocation in this scenario.Firstly,according to the characteristics of high dynamic and complexity of integrated information network,this paper expands the concept of cell in traditional mobile network and puts forward the concept of "cluster".In order to study the problem of resource allocation within a single cluster,a three-dimensional Manhattan grid model is proposed.This model can be designed as a centralized and distributed architecture according to the needs of operators,and can simulate the communication behavior of users in a single cluster.Then,on the basis of the reinforcement learning theory,this paper introduces a deep cyclic Q network algorithm which can deal with the resource allocation problem with partial observation Markov process better.Secondly,we design a centralized power channel joint allocation algorithm in a three-dimensional Manhattan grid.We assume that the perceptual data of the perceptual nodes in the environment are all noisy and model the problem as a partially observational Markov process.Then,the reward function is designed according to the minimum SNR threshold of the user,and the sensing data of the sensing nodes at different locations are used as the environmental state vectors of the network.The small-batch parameter learning method is used to iterate repeatedly to learn the optimal power channel joint allocation strategy for the user.Finally,we design a distributed power channel joint allocation algorithm in a three-dimensional Manhattan grid.We assume that users in the model have some autonomy and have communication links with two different types of quality-of-service requirements: user-to-user and user-to-infrastructure.Then,we abstract the problem into a multi-objective joint optimization problem with the constraints of the two kinds of link interference.We train the deep loop Q network with the offline training mode,learn the optimal power channel joint allocation strategy for users,and compare the network performance with the online test mode.