Deep Learning for Power Control and Allocation of Satellite Earth Link

With the vigorous development of low orbit satellite internet, the advantage of wide area coverage in satellite communication can effectively compensate for the limited coverage capacity of ground cellular networks. Therefore, studying how ground users can access the low Earth orbit satellite internet and how to allocate resources after access is an important part of the implementation of the space-air-ground integrated network. On the other hand, since entering the AI era, deep learning has always provided rich possibilities for solving various problems. Therefore, achieving the allocation of satellite power between satellite to ground links based on deep learning has important research significance. This paper discusses a model for users to access satellite networks nearby, and establishes an optimization problem for satellite power allocation with the goal of maximizing satellite network capacity. Since the downlink channel fading of low orbit satellites mainly comes from path loss, the geographic location information of satellites and ground stations can be used as input features for neural networks to learn power allocation methods. This paper constructs a forward propagation neural network and trains it in a supervised manner. Through simulation, it can be proven that the neural network proposed in this paper can effectively allocate satellite downlink power at a lower algorithm complexity.