Extended Weighted ABG: A Robust Non-Linear ABG-Based Approach for Optimal Combination of ABG Path-Loss Propagation Models

Abstract

This paper proposes a robust non-linear generalized path-loss propagation model, the Extended Weighted ABG (EWABG), which efficiently allows generating a path-loss propagation model by combining several available path-loss datasets (from measurements campaigns) and other previously proposed state-of-the-art 5G path-loss propagation models. The EWABG model works by integrating individual path-loss models into one single model in the least-squares sense, allowing to extend knowledge from frequencies and distances covered by path-loss datasets or path-loss propagation models. The proposed EWABG model is the first non-linear extension of the common ABG-based approach, which surpasses the non-uniformity problem between the low and high 5G frequencies (as most measurements campaigns have taken place in low frequencies). The EWABG also addresses the problem of removing outlier measurements, a step not included in previous propagation path-loss models. In this case, we have compared the most recent techniques for avoiding outliers, and we have adopted the Theil-Sen method, due to its strong robustness demonstrated in the experiments carried out. In addition, the proposed model specifically considers non-linear attenuation by atmospheric gases, in order to improve its estimations. The good performance of the proposed EWABG model has been tested and compared against recent 5G propagation path-loss models including the ABG and WABG models. The exhaustive experimentation carried out includes the 5G non-line-of-sight environment in different 5G scenarios, UMiSC, UMiOS and UMa. The proposed EWABG obtains the best accuracy, specially in noisy environments with outliers, reporting negligible increment error rates (with respect to the non-outliers situation), lower than 1%, compared to the ABG and WABG.