Maximizing UAV Data Efficiency in NextG Networks: A Transformer-Based mmWave Beamforming Approach

Avi Deb Raha, Apurba Adhikary, Mrityunjoy Gain, Yu Qiao, Hyeonsu Kim, Jisu Yoon, Choong Seon Hong

http://doi.org/10.5626/JOK.2025.52.2.170

Beamforming is essential in the rapidly evolving field of next generation (NextG) wireless communication, particularly when leveraging terahertz and millimeter-wave (mmWave) frequency bands to achieve ultra-high data speeds. However, these frequency bands present challenges, particularly concerning the costs associated with beam training, which can hinder Ultra-Reliable Low-Latency Communication (URLLC) in high-mobility applications, such as drone and Unmanned Aerial Vehicle (UAV) communications. This paper proposes a contextual information-based mmWave beamforming approach for UAVs and formulates an optimization problem aimed at maximizing data rates in high-mobility UAV scenarios. To predict optimal beams while ensuring URLLC, we have developed a lightweight transformerThe self-attention mechanism of the transformer allows the model to focus selectively on the most important features of the contextual information. This lightweight transformer model effectively predicts the best beams, thereby enhancing the data rates of UAVs. Simulation results demonstrate the design's effectiveness, as the lightweight transformer model significantly outperforms baseline methods, achieving up to 17.8% higher Top-1 beam accuracies and reducing average power loss by as much as 96.79%. Improvements range from 12.49% to 96.79% relative to baseline methods.