SA-TBP: A Speed-Adaptive Temporal Beam Prediction Framework With AI for mmWave Communications

This paper presents a speed-adaptive temporal beam prediction (SA-TBP) framework based on artificial intelligence (AI) for millimeter-wave (mmWave) communication systems. The framework dynamically adjusts the historical and prediction windows according to the speed of the user equipment (UE). The model is built on an encoder-decoder LSTM architecture, with an attention mechanism integrated into the decoder. The speed-adaptive module allows flexible adjustment of both the historical measurement and future prediction windows based on the UE speed. Experimental results show that the proposed model achieves a Top-1 prediction accuracy of 92.48%, effectively reducing power consumption and resource overhead caused by frequent beam measurements, while enhancing beam selection efficiency.