| （英） |
Recently, deep neural network (DNN) have been utilized in various fields as image recognition, speech recognition, and more.In this work, we describe DNN based path loss prediction in an urban macrocell (UMa) environment.Some models using convolutional neural network (CNN) as the DNN for path loss prediction have been proposed.Most of these models input the building information around the transmitter (Tx) and the receiver (Rx) to the CNN as the image seen from above.However, in the UMa non-line-of-sight (NLoS) environment, propagation over rooftops between Tx and Rx becomes dominant.Therefore, in order to consider the over-rooftop propagation, we define the side-view image on the straight line as new input image to the CNN.We then propose a new prediction model using not only side-view image but also top-view image used for input to the conventional model, and verify the model using measurement data.The estimation result showed that the RMS error of the proposed model was less than 6 dB, while the conventional model was about 9 dB. |