Channel State Information Prediction Using Autoencoder and Transfer Learning

Abstract

Accurate channel state information (CSI) is essential for optimizing wireless communication performance. Traditional models, such as parameter-based and autoregressive approaches, suffer from noise interference and adaptability issues. In response, this paper proposes a novel deep learning framework combining a deep convolutional autoencoder with CNN-BiLSTM to enhance CSI prediction. The autoencoder denoises and refines CSI data, while the CNN-BiLSTM extracts both local and global temporal features. To address time-varying channel dynamics, transfer learning is employed, enabling the model to adapt to new environments with minimal data. Comparative analysis with traditional and deep learning methods demonstrates that the proposed approach significantly improves prediction accuracy, optimizes resource allocation, and enhances overall communication quality.