Prediction of indoor radio propagation using deep neural network

Abstract

Radio propagation prediction is important for improving the stability and reliability of wireless communication systems. However, propagation loss prediction models by multiple regression based on measured data have problems in setting appropriate parameters and function types. This paper evaluates a radio propagation model that applies machine learning for indoor environments. The proposed neural network has the input parameters of the environmental characteristics such as floor area, ceiling height, and materials, and a dropout layer that disables the network units randomly is inserted to improve generalization performance. The RMSE (root mean squared error) by the proposed model is less than 5 dB for test data and the average accuracy of the proposed model is better than the conventional model.