Optimal Beam Prediction Using Physical Space Information for Multiple APs on 60 GHz Channels

Abstract

Millimeter-wave (mmWave) bands are promising for growing mobile traffic. However, their strong sensitivity to line-of-sight (LOS) conditions makes reliable communication challenging in dynamic environments. Therefore, accurate beam selection is essential to maintain link quality (LQ). Conventional beam selection relies on exhaustive beam sweeping, which incurs latency in narrow-beam systems and often leads to suboptimal beam choices during user equipment (UE) mobility. To overcome this issue, we propose a long short-term memory (LSTM)-based method that predicts future optimal beams using only physical space information-UE position, direction, and velocity. This approach eliminates beam sweeping and is inherently scalable, independent of the number of beams or access points (APs). Real-world indoor pedestrian experiments with two 60-GHz-band APs demonstrate that the proposed method significantly reduces the probability of the received signal strength indicator (RSSI) falling below -80 dBm by 64.2% and 41.2% compared with conventional beam sweeping-based selection and LSTM-based prediction using RSSI, respectively. Furthermore, the method remains robust in both LOS and challenging non-line-of-sight (NLOS) environments, where optimal beams frequently change non-sequentially.