Three-Dimensional Radio Spectrum Map Prediction Based on Fully Connected Neural Network

Aiming at the problems that the existing radio spectrum maps only consider the two-dimensional map environment hardly meet the engineering requirements, and the simulation time of ray tracing is too long, this paper proposes a three-dimensional (3D) radio spectrum map construction method. First, the ray tracing method is used to obtain the data set, use the data set to train the fully connected neural network, obtain the preliminary model, and set the correction function, so that the model can be modified by a few measured points. Receiving point coordinates and Reference Signal Receiving Power (RSRP) can be obtained only by inputting coordinate values and house map data, and 3D radio spectrum map can be obtained by converting RSRP into thermal values and drawing it into 3D thermal map. The paper reports a remarkable achievement in RSRP prediction accuracy, achieving up to 95% accuracy within 5dB. Furthermore, the proposed method is noted to be significantly faster—up to 60 times—than traditional ray tracing simulations.