Hardware-Aware Dynamic TransDiff: Generative Channel Tracking and Coding for High-Mobility UAV-RIS 6G Networks

Abstract

Uncrewed aerial vehicles (UAVs) that use reconfigurable intelligent surfaces (RIS) offer exceptional spatial flexibility for 6G networks. However, their real-world application faces significant challenges due to fast-changing channel conditions and limitations in hardware performance. Most generative channel estimation models incorrectly assume perfect, zero-power phase shifts and static conditions, which leads to significant failures when dealing with strong Doppler effects and real-world 1/2-bit RLC circuit limitations. To address this issue, this letter proposes a dynamic Transformer-Diffusion (TransDiff) framework that takes into account hardware considerations. By smoothly combining a generative Vision Transformer (ViT) with temporal Kalman tracking, the proposed method successfully captures the quick changes in both spatial and temporal channel behavior. Furthermore, a generative coding mechanism is integrated to physically penalize the diffusion reverse process using exact RLC impedance mismatches. Numerical evaluations confirm that at a high mobility of 120 km/h, the proposed framework achieves an unprecedented normalized mean square error (NMSE) of -21.87 dB, yielding an effective pilot overhead reduction of 29.3% and maximizing the energy efficiency to 19.15 bits/J, substantially outperforming state-of-the-art CNN-based denoisers.