Investigating Predictive Capabilities of RBFNN, MLPNN and GRNN Models for LTE Cellular Network Radio Signal Power Datasets

Abstract

Efficient radio frequency signal coverage planning with well configured transmitters and receivers’ communication channels, is the heart of any cost-effective cellular network design, deployment and operation. It ensures that both network quality and coverage are simultaneously make best use of (i.e. maximized). This work aim to appraise the adaptive learning and predictive capacity of three neural network models on spatial radio signal power datasets obtained from commercial LTE cellular networks. The neural network models are radial basis function neural network (RBFNN), multilayer perceptron neural network (MLPNN) trained with Bayesian regulation algorithms and general regression neural network (GRNN) models. Largely, it is established from the results that ANN prediction methods can tolerate and adapt to measurement errors of attenuating LTE radio signals. Performance comparisons reveal that all the neural network models can predict the propagated LTE radio signals with considerable errors. Specifically, RBFNN delivered the overall best performance with the smallest mean absolute percentage error, root mean square error, mean absolute error and standard deviation values. The GRNN model also gave better prediction results with marginal errors compared to the MLPNN. Thus, the predictive abilities of RBFNN and GRNN models can be explored as a useful tool to successfully plan or fine-tune mobile radio signal coverage area. Keywords: Neural networks; Signal power; attenuating radio signals; radial basis function multilayer perceptron, general regression neural network, Adaptive signal prediction