This research transforms wireless communication by applying large language models to predict optimal millimeter wave beam patterns, improving reliability and performance of next-gen networks.

• Transforms the beam prediction challenge into a time series forecasting task
• Converts beam pattern data into text-based representations with trainable tokenizers
• Employs the prompt-as-prefix technique for contextual enrichment
• Demonstrates more robust and accurate beam predictions than traditional methods

This innovation addresses a critical challenge in 5G/6G networks by enhancing signal quality and reducing connection disruptions in millimeter wave communications, which are essential for high-bandwidth applications but struggle with environmental obstacles.

Beam Prediction based on Large Language Models