AdapCsiNet: Environment-Adaptive CSI Feedback via Scene Graph-Aided Deep Learning

TL;DR

This work tackles the generalization gap in DL-based CSI feedback across diverse propagation environments and the overhead of online training. It introduces AdapCsiNet, which embeds environment information as a scene graph into a hypernetwork that generates environment-aware parameters for the CSI reconstruction NN, enabling dynamic adaptation without online CSI collection. A two-step training process first establishes a robust baseline reconstruction and then refines it with environment-conditioned parameters, yielding up to $46.4\%$ improvement in $NMSE$ (2.7 dB) and performance comparable to online training but with no additional runtime burden. The approach demonstrates that explicit environmental priors can substantially enhance CSI feedback performance, offering practical benefits for multi-environment deployments in next-generation wireless systems.

Abstract

Accurate channel state information (CSI) is critical for realizing the full potential of multiple-antenna wireless communication systems. While deep learning (DL)-based CSI feedback methods have shown promise in reducing feedback overhead, their generalization capability across varying propagation environments remains limited due to their data-driven nature. Existing solutions based on online training improve adaptability but impose significant overhead in terms of data collection and computational resources. In this work, we propose AdapCsiNet, an environment-adaptive DL-based CSI feedback framework that eliminates the need for online training. By integrating environmental information -- represented as a scene graph -- into a hypernetwork-guided CSI reconstruction process, AdapCsiNet dynamically adapts to diverse channel conditions. A two-step training strategy is introduced to ensure baseline reconstruction performance and effective environment-aware adaptation. Simulation results demonstrate that AdapCsiNet achieves up to 46.4% improvement in CSI reconstruction accuracy and matches the performance of online learning methods without incurring additional runtime overhead.

Figures (7)

• Figure 1: One-sided CSI reconstruction framework.
• Figure 2: Illustration of the AdapCsiNet framework.
• Figure 3: Detailed NN architecture. (a) CSI reconstruction NN. (b) Hypernetwork structure.
• Figure 4: Visualization of the scene graph. (a) Discrete scene graph matrix $100 \times 100$ where 1 (orange) denotes internal walls, 0 (blank areas) represents free space, and 2 (black) indicates spatial boundaries where UEs are distributed. (b) Ray tracing path visualization of a UE in the scene.
• Figure 5: Comparison of NMSE under different CRs for the general feedback NN mix2024train and the proposed AdapCsiNet.