ANN-based model for multiband path loss prediction in built-up environments

Path loss propagation models are critically needed for optimum planning and deployment of wireless communication networks. However, the complexity exhibited by the propagated signals makes the prediction of the received losses difficult in built-up environments. There is however a new paradigm shift towards the application of computational methodologies, such as the Artificial Neural Networks (ANN), for multi-band path loss prediction. In this paper, we have developed a new ANN-based model for path loss prediction. The model was developed using large scale path loss data collected across 485 base stations in 6 urban cities of Nigeria, West Africa. The data collection, which spanned a period of 9 years, were taken over open areas, sub urban and urban environments, and the bands considered were 89.3 MHz, 103.5 MHz, 203.25MHz, 429.25 MHz, 529.25MHz, 615.25MHz, 629.25MHz, 900MHz, 1800 MHz and 2100MHz. The developed model was validated using independent path loss data across different frequencies, environments; and, distances and the results were compared with the popular empirical models such as Hata, COST 231 and Egli models, and to previously published ANN-based multi-frequency models. The global performance of 4.81 dB was obtained in terms of RMSE value with an R-value of 0.96, thus outperforming the existing ANN-based path loss models that were developed for multiple frequencies. Based on these findings, the proposed model can be deployed across all categories since the average RMSE values are all within the acceptable thresholds. Furthermore, the model is multi-frequency, thus will be suitable for multiple and complex environments, and usable for both short- and long-range wireless communication networks