Frequency extension of radio propagation model using fine-tuning

抄録

Research and development of wireless emulation technology have been conducted for large-scale evaluation and verification of wireless communication systems in a virtual space. Emulation for various scenarios requires accurate and fast modeling techniques for radio propagation characteristics. The authors have proposed modeling methods using machine learning. However, when the amount of measurement data is slight, such as when new frequencies are implemented, the modeling accuracy is an important issue due to insufficient learning. This paper clarifies the relationship between the amount of data and the modeling accuracy. Moreover, we propose a fine-tuning method for modeling the propagation characteristics in a new frequency by pre-training in a frequency with large training data. Finally, through the evaluation using the measurement data in various areas, we demonstrate the effectiveness of the proposed method.