The use of millimeter-wave (mmWave) bands is essential for increasing the capacity of mobile networks. However, the link quality (LQ) in mmWave bands is highly sensitive to signal blockage caused by obstacles, making it necessary to predict future LQ and adaptively control wireless communication to maintain stable connectivity. Deep neural network (DNN)-based LQ prediction using physical space information, such as the positions of user equipment (UE) and surrounding obstacles, enables proactive control by anticipating user movement and blockage events. However, the number of blocking obstacles is generally unknown in advance, and conventional supervised learning approaches require separate models for each environment, limiting their scalability. To address this limitation, we propose a common LQ prediction model that can flexibly adapt to environments with varying numbers of blocking obstacles. As an evaluation scenario, we considered an indoor 5G environment where one UE holder and several pedestrians walk, causing two types of LQ degradation: self-blocking due to UE holder and blocking due to pedestrians. Experimental results from indoor commercial 5G experiments under three pedestrian-density scenarios demonstrate that the proposed single model can accurately predict future LQ in all cases. We also analyze the contribution of physical space information, showing that position, direction, and velocity are important for the UE holder, whereas position alone is sufficient for pedestrians.
Nagata et al. (2026) studied this question.