Two-step path loss prediction by artificial neural network for wireless service area planning

抄録

In recent years, wireless network systems are utilized in various industry fields and the wireless service area planning became one of the important tasks to realize efficient and high-quality wireless communication service. The machine learning technology attracts the interests of researchers to improve the efficiency of the area planning task because the radio propagation loss in unknown locations can be predicted by the training data without explicit algorithms. Our previous work showed that the path loss (PL) characteristics become complicated in the high PL region, and it can degrade the entire prediction accuracy. In this paper, we propose the two-step PL prediction method by the artificial neural network (ANN) to solve the issue. Firstly, the area is classified into several zones according to the PL range. And then the PL is predicted by ANNs that were trained for respective zones. Our proposal was evaluated by the ray-tracing simulation data, and the result showed that it improved the root mean square error (RMSE) of PL prediction from 7.9 dB to 4.1 dB. The method is expected to be utilized for the wireless service area planning in various environments.