An Improved Recurrent Neural Network for Radio Propagation Loss Prediction

Abstract

Prediction of the radio propagation loss using a numeric parabolic equation method is often accepted for its accuracy, but the large computational time is a hindrance in applications requiring real-time situation awareness. A modified Elman recurrent neural network is proposed and developed to resolve this problem. In this paper, the three dimensional parabolic equation models is used to provide the sample set of the neural network, and improved BP algorithm is used for the training and study of network. Then the Elman network model established is used to predict propagation loss in rest region. In contrast to other prediction models, the results show that Elman neural network that dramatically improves the computation speed with a better precision is reliable and practical.