Adaptive Massive MIMO Hybrid Precoding Based on Meta Learning

Abstract

Massive multiple-input multiple-output (MIMO) mmWave communication has emerged as a pivotal technology in the fifth generation (5G) mobile communication system, with hybrid precoding technology garnering significant attention. In this paper, a hybrid precoding scheme based on convolutional neural network (CNN) is proposed. By utilizing an analog precoder with a partial connection structure, the network size is reduced and convergence is facilitated. Compared to traditional algorithms, this approach reduces computational complexity and time overhead. However, similar to other deep learning algorithms, it lacks robustness as the trained network may not be adaptable to new environments. To address this limitation, this paper further proposes a hybrid precoding scheme based on meta-learning, enabling the CNN model to quickly adapt to changing environments.