Deep Learning-Based Radio Estimation Using a Semi-Automatically Created Indoor Building Information

The adoption of wireless communication in various fields has created the needs for easy wireless network construction and optimal wireless system selection, considering the coexistence of multiple wireless systems. In this paper, we propose two main contributions: a database that describes indoor building information and a wireless estimation method that uses deep learning. The first contribution involves digitizing indoor building information and creating the database for the building information, while the second utilizes the database to operate a deep learning model called RadioResUNet, which predicts the radio signal strength (RSS) of a 2.4 GHz Wi-Fi system in complex indoor environments with multiple walls and obstacles. By leveraging the building information database, the deep learning-based radio estimation method enables the estimation of RSS for a given building configuration and the position of the access point (AP). This method also allows for radio prediction in a desired area in advance, even when updating the indoor building information in the database. It can save time and expense while ensuring ongoing wireless connectivity support by comprehending the wireless situation of the new indoor building configuration,