Wireless Communication Data Replica with Autoencoder and Digital Twin

In this paper, we propose a digital twin data refinement algorithm leveraging autoencoder for real environment data augmentation. Data augmentation by digital twin for machine learning-based indoor positioning is already a well-known method. However, as accurate replication of real-world indoor environments is still challenging, the differences between real-world data and digital twin data are inevitable. To address this problem, we leverage an autoencoder to refine the digital twin data to decrease those differences. Our experimental results show that the root mean square error (RMSE) between received signal strength indicator (RSSI) data generated in digital twin environment and real-world RSSI data is decreased from 7.1619 dBm to 0.0053 dBm after conducting the proposed refinement algorithm.