Digital-Twin Aided Spatial Beam Prediction

Millimeter-wave communication systems are vital for next-generation wireless networks, offering high data rates but requiring precise beamforming due to their highly directional propagation. This paper focuses on leveraging a digital-twin to address challenges in spatial beam prediction by augmenting real-world datasets with synthetic data. The digital-twin is constructed by replicating the geometry of the testbed environment in Blender and simulating the wireless channel characteristics using DeepMIMO. Combining real-world power measurements from wide beams with synthetic data, we aim to predict the optimal narrow beam for communication at 28 GHz using a convolutional neural network. The performance is evaluated by comparing the mean absolute error and top-k accuracy of models trained solely on real-world data versus those augmented with synthetic data. Our results demonstrate that the inclusion of synthetic data reduces the reliance on extensive real-world measurements while improving beam prediction accuracy. These findings highlight the potential of digital-twin technology to enhance scalability and efficiency in millimeter-wave communication systems.