3D Aerial Base Station Position Planning based on Deep Q-Network for Capacity Enhancement

When the existing traditional terrestrial base station is insufficient to meet the sudden traffic demand or is not available, deploying the aerial base station(aerial-BS) is a fast and effective solution for achieving network capacity enhancement. How to plan the best 3D location of the aerial-BS according to the user's business needs and service scenarios is a key issue to be solved. At present, the conventional optimization algorithms that solve this problem have high time complexity and it is difficult to utilize experience. However, applying the deep reinforcement learning model can quickly get an optimal solution by historical experience feedback training. Therefore, it is suitable for solving the optimal 3D location planning problem of the aerial-BS. In this paper, firstly, aiming at the maximum spectral efficiency of the system, considering the effects of line-of-sight and non-line-of-sight path loss, a mathematical optimization model for the location planning of the aerial-BS is proposed. For this model, the model definition and training process of deep Q-Network are constructed, and through the large-scale pre-learning experience of different user layouts in the training process to gain experience, improve the timeliness of the training process. The simulation results show that the proposed method can achieve the spectral efficiency of more than 91% of the theoretical maximum spectral efficiency, which has lower time complexity than traditional genetic algorithms (such as hill climbing algorithm and simulated annealing algorithm).