Deep-learning path loss prediction model using side-view images

This paper proposes a path loss prediction model based on a convolutional neural network utilizing side-view images to consider over-rooftop propagation, in addition to the top-view images around the receiving station of the conventional model in the urban macrocell environment. The building profile between the transmitting and receiving stations was used for side-view images. In addition, the scalar parameter of frequency was added to the fully connected neural network part as a proposed method to consider frequency characteristics. The model was learned and validated using the measured data, and the estimation error was compared with the conventional model to evaluate its validity. Our findings showed that the RMS error of 12.1 dB using the conventional model was improved to 4.4 dB by the proposed model.