Radial basis function neural network path loss prediction model for LTE networks in multitransmitter signal propagation environments

Path loss prediction models occupy a central role in wireless signal propagation because of the continuous need to achieve reliable and high quality of service for subscribers satisfaction. However, the adoption of deterministic and empirical models for pathloss characterization presents a highly contending trade-off between simplicity and accuracy. On the one hand, empirical models are relatively simple to apply but are mostly inaccurate and inconsistent. Deterministic models are more accurate but quite complex to develop, time-consuming, and possess nonadaptable characteristics. Toward this end, this paper proposes to address the problems associated with the existing models (empirical and deterministic) through the introduction of machine learning algorithms to path loss predictions. The contribution of this paper is in threefold. First, experimental data were collected in multitransmitter scenarios via drive test in six base transceiver stations, and the pathloss of the received signal level was derived and analyzed. Two machine learning-based path loss prediction models were then developed using the measured data as input variables. The developed path loss prediction models are the radial basis function neural network (RBFNN) and the multilayer perception neural network (MLPNN). Further to this, the MLPNN and the RBFNN models were compared with the measured path loss, and the RBFNN appears to be more accurate with lower values of root mean squared errors (RMSEs) in comparison with the MLPNN. Finally, the proposed machine language-based path loss prediction models (MLPNN and RBFNN) were compared against five existing empirical models, and again, the RBFNN shows the most accurate results.