Deep Learning for Efficient CSI Feedback in Massive MIMO: Adapting to New Environments and Small Datasets

Channel State Information (CSI) feedback, powered by Deep Learning (DL) methodologies, exhibits significant promise in enhancing spectrum efficiency within massive MIMO systems. However, DL-based approaches typically necessitate substantial CSI datasets for each specific scenario, and managing multiple learned models demands considerable storage and updating bandwidth. To overcome this costly barrier, we develop a solution for efficient training and deployment enhancement of DL-based CSI feedback, which involves a lightweight translation model to cope with new CSI environments and introduces a novel dataset augmentation based on domain knowledge. Specifically, we first develop a deep unfolding CSI feedback network, SPTM2-ISTANet+, which incorporates spherical normalization to mitigate the challenge of path loss variation. Additionally, SPTM2-ISTANet+ integrates a trainable measurement matrix and residual CSI recovery blocks to enhance efficiency and accuracy. Employing SPTM2-ISTANet+ as a foundational feedback model, we introduce an adaptive CSI feedback architecture termed CSI-TransNet. CSI-TransNet features a scenario-adaptive plug-in module for CSI translation, composed of a sparsity aligning function and a compact DL module, facilitating the reuse of pretrained models in unencountered environments. To accommodate the small datasets, we propose a lightweight and general augmentation strategy based on domain knowledge. Test results demonstrate the efficacy and efficiency of the proposed solution for accurate CSI feedback given limited measurements for unseen CSI environments.