Research On Allocation Policy Of Wireless Cellular Network Resources Based On Reinforcement Learning

With the increasing number of wireless data transmission and mobile devices in 5G communication systems,cellular networks require more and more base stations to meet the demands of higher system capacity.However,due to severe co-channel interference,limited radio resources cannot support a wide range of base stations to transmit data efficiently at the same time.This is a prominent problem in large dense networks.Therefore,designing a reasonable resource allocation and interference management strategy is crucial for improving the system data capacity.Due to the non-convexity and large-scale nature of optimization problems,the mathematical model of the system required by traditional centralized algorithms is difficult to accurately describe,and the computational complexity of that is unacceptable in practice.In this paper,the problem of dynamic power allocation in downlink cellular networks based on multi-agent Deep Reinforcement Learning(DRL)is studied,in which the downlink between each base station and user is modeled as a learned agent to learn an optimal allocation policy to maximize the overall rate of the system.The power allocation problem is transformed into a multi-agent Markov decision process to be solved by a distributed deep reinforcement learning algorithm.This paper focuses on the scalability of the reward function and state space in the learning process to adapt to changes in network size in the 5G environment,such as the number of serving base stations or access users,and changes in cell coverage areas.Furthermore,this paper evaluates the impact of different learned hyperparameters on the performance of the algorithm.Finally,the effectiveness of the deep reinforcement learning algorithm and its superiority compared with the traditional centralized algorithm are verified by numerical results in different scenarios.This paper also studies how to allocate power to mobile users.On the basis that the user maintains a fixed location but the channel keeps changing,the system model is further expanded,considering the mobility of the user and the time delay of the allocation strategy reaching the user end,and the model is closer to reality;for the scenario of mobile users,this paper proposes a centralized training and distributed execution learning framework with time delay,using deep deterministic policy gradient algorithm and the deep Q-learning algorithm to determine the allocation strategy for users in real time.In this paper,a preprocessing method that simplifies the learned state space is used,and a scalable form of the reward function is designed.Through simulation experiments,it is verified that the performance of the deep reinforcement learning algorithm proposed in this paper exceeds the traditional optimization algorithm with an ideal channel model under the premise of considering the delay;this paper also simulates different test environments,and the performance of the deep reinforcement learning algorithm is better than traditional optimization algorithms.