Abstract:Due to the advantages of beamforming technology in multi-antenna wireless communication systems, it has been widely studied and applied. However, its effectiveness in high-mobility scenarios may significantly deteriorate due to severe channel aging. In this paper, we propose an improved Lévy induced stochastic differential equation network (LDE-Net) based on the Euler-Maruyama method for channel prediction. The proposed scheme captures the stochastic and rapid changes of wireless channels, by dividing the neural network into fine-grained stages while maintaining a moderate model size by parameter sharing among stages. In order to verify its performance in practice, we conduct channel measurements in Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) scenarios on our campus, respectively. In both scenarios, LDE-Net performs better than the conventional and widely-used autoregressive model, long short term memory, and recurrent reural network schemes. In the LoS scenario, its throughput is 19.615% on average higher than all other schemes at various speeds, and 10.64% for NLoS.

Euler-Maruyama Method Based Channel Prediction: An LDE-Net Implementation and Field Evaluation