Dqn-Based Resource Allocation for Power Management in Cellular Network

In recent years, power allocation in cellular networks has become increasingly crucial with the rapid development of mobile communications. This report presents a comparison of different power allocation approaches, including Deep Q-Network (DQN), Policy Gradient, Fractional Programming (FP), and Weighted Minimum Mean Square Error (WMMSE). The study first constructs a grid-based multi-cell downlink cellular network environment that incorporates practical features such as path loss, channel fading, and interference models. Through experiments, the study evaluates these methods in terms of performance, computational efficiency, and stability. The results show that DQN achieves a balanced performance with an average reward of 1.52 and runtime of 316-349 seconds, while WMMSE achieves slightly higher performance (1.58) but requires significantly more computational resources (850 seconds). The findings show that deep reinforcement learning approaches, particularly DQN, offer promising solutions for power allocation in cellular networks, combining strong performance with reasonable computational efficiency. This study aims to explore the impact of different methods in Power Management in Cellular Networks.