Enhancing XL-MIMO Hybrid-Field Channel Estimation With Modular Staged Dictionary Learning

Extremely large-scale multiple-input multiple-output (XL-MIMO) is a promising key technology for 6G, offering significant performance improvements. However, the deployment of ultra-large antenna arrays introduces hybrid-field channel characteristics, where far-field (FF) and near-field (NF) components coexist. This creates major challenges for channel estimation, including high pilot overhead and increased complexity. To address these issues, we propose a modular staged dictionary learning enhanced hybrid-field approximate message passing (MSDL-HF-AMP) scheme. The proposed method uses HF-AMP as the core framework and connects the FF-AMP and NF-AMP modules in sequence through a residual transfer mechanism, enabling adaptive separation and estimation of FF and NF components. At the same time, the MSDL strategy builds a deep neural network based on HF-AMP and employs a three-stage training process to gradually optimize the angular-domain and polar-domain sparse dictionaries, effectively adapting to hybrid-field channel characteristics. Simulation results show that the scheme outperforms the existing schemes in terms of estimation accuracy and robustness.