Deep Learning Based Lightweight Neural Network for Massive MIMO CSI Feedback

In Massive multiple-input multiple-output (MIMO) systems, channel state information (CSI) is usually estimated and fed back to the BS from the user equipment (UE), and the quality of CSI received by the BS has a large impact on the performance of the large-scale MIMO system. To address the problem that existing DL-based CSI feedback methods have difficulty in balancing model complexity and feedback performance, in this work, we propose CFNet, an efficient and lightweight model based on MetaFormer, which employs convolution to replace the attention module in the traditional Transformer model. CFNet combines the lightweight of convolutional neural networks with the high performance of the Vision Transformer (VIT) model. CFNet combines the lightweight of convolutional neural network and the high performance of Vision Transformer (VIT) model. Experimental results show that CFNet can maintain high feedback accuracy with very low computational complexity under different compression ratios.