AirFM-DDA: Air-Interface Foundation Model in the Delay-Doppler-Angle Domain for AI-Native 6G

Researchers from academia and industry have introduced AirFM-DDA, a novel air-interface foundation model designed for AI-native 6G physical layer tasks. Unlike existing models that operate on channel state information (CSI) in the space-time-frequency (STF) domain—where multipath components are inherently superimposed and structurally entangled—AirFM-DDA reparameterizes CSI into the delay-Doppler-angle (DDA) domain. This transformation explicitly resolves multipath components along physically meaningful axes, enabling more universal channel representation learning. The model employs a window-based attention mechanism augmented with frame-structure-aware positional encoding (FS-PE), which aligns with locally clustered multipath dependencies and avoids the quadratic-complexity overhead of global attention. This design reduces training and inference costs by nearly an order of magnitude while maintaining or improving performance.

In extensive experiments, AirFM-DDA demonstrates superior zero-shot generalization across unseen scenarios and datasets, consistently outperforming baseline models on channel prediction and estimation tasks. The model maintains robustness under challenging conditions including high mobility, large delay spreads, severe noise, and extreme aliasing. By decoupling the complex wireless channel into structured components and using efficient attention, AirFM-DDA represents a significant step toward practical, scalable foundation models for 6G networks. The work was submitted to arXiv on April 19, 2026, and is available under arXiv:2605.00020.