Deep Learning for Movable Antenna Precoding in 2D MISO Communication System

The rapid advancement of wireless communication systems has spurred the exploration of innovative methods to significantly enhance signal transmission efficiency. Movable Antenna (MA) technology has emerged as a promising approach to improving the spatial characteristics of wireless channels. In flexible precoding systems, the positions of antennas at the trans-mitter can be dynamically adjusted to adapt to changing channel conditions. However, optimizing the positions of a large number of movable antennas in massive MIMO systems introduces substantial complexity, making real-time channel adaptation a critical challenge. By incorporating mobility into the antenna array, it becomes possible to design a more efficient and robust intelligent communication system. This paper proposes a CNN-based Flexible Precoding (CBFP) framework that leverages deep learning (DL) techniques to optimize both antenna positions and beamforming weights. Furthermore, we present the theoretical foundations and develop a two-dimensional (2D) antenna position search algorithm to identify the optimal configuration, while en-suring compliance with the minimum antenna spacing constraint. Extensive experimental results demonstrate that the proposed method significantly outperforms other optimization techniques, yielding notable improvements in both spectral efficiency and computational efficiency.