MetaFormer-Based Lightweight Neural Network for Massive MIMO CSI Feedback

Channel state information (CSI) is usually estimated by the user equipment (UE) and fed back to the base station (BS). The quality of the CSI received by the BS significantly impacts the performance of large-scale multiple-input-multiple-output (MIMO) system. The CSI feedback consumes a large amount of uplink bandwidth resources, especially in massive MIMO system. Traditional CSI feedback methods cause feedback performance bottlenecks due to their computational error and complexity limitations. In recent years, deep learning (DL)-based CSI feedback methods have made significant progress. However, most of the existing DL-based methods improve CSI feedback performance at the cost of higher computational complexity. In this letter, we introduce the MetaFormer generic lightweight architecture to CSI feedback and design a lightweight feedback model PFNet based on it, which employs pooling operations instead of the attention mechanism in the traditional Transformer architecture, thus significantly reducing the complexity. Experimental results show that PFNet outperforms current lightweight SOTA networks in most scenarios.