Transformer-Based Prediction and Synthesis of Generative Channel for Orientation-Aware Beamforming

As 5G and emerging 6G systems increasingly leverage high-frequency bands for high-speed communications, beamforming becomes essential for reducing interference and enhancing coverage. However, effective beamforming must contend with the dynamic movements of user equipment (UE), which can cause rapid fluctuations in communication channels. While recent studies have explored learning-based approaches to address this challenge, many remain limited by their reliance on site-specific characteristics, restricting their generalizability across diverse environments. To enhance the robustness, we propose a generative-predictive transformer framework that predicts the optimal transmitter (TX)-receiver (RX) beam index pair based on recent channel states and variations in UE orientation. By learning generalized patterns from past channel behavior, the proposed system can adapt effectively to diverse deployment scenarios. We evaluate the performance of our model across a wide range of dynamic conditions, demonstrating its robustness and effectiveness in maintaining reliable beam alignment.