A Hybrid Stacked Learning Framework for Accurate Path Loss, LOS Probability, and Coverage Prediction in 28 GHz 5G mmWave Networks

Abstract

Purpose: The aim of the research is to design a consistent and dynamic mechanism for estimating key performance indicators of millimetre wave wireless communication systems such as path loss, probability of line-of-sight (LOS) channels, maximum coverage area, and spectral efficiency.  In 28 GHz mmWave networks with respect to Indoor, UMi, UMa, and RMa deployment environments. The objectives of the research are to study the propagation characteristics of 28 GHz mmWave communication systems across different deployment environments, including Indoor, UMi, UMa, and RMa. The research also aims to analyse and estimate the probability of Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) channel conditions for different mmWave deployment environments. Furthermore, it investigates the impact of environmental factors, including distance, blockage, antenna height, and user density, on mmWave network performance. In addition, the study conducts a comparative analysis of the deployment environments under consideration based on the estimated KPIs and identifies the most efficient scenario for reliable mmWave communication. Finally, the proposed estimation mechanism is validated using standard channel models and simulation tools for next-generation wireless communication systems. Design / Methodology / Approach: The method adopted in the research was the design of a hybrid stacked learning algorithm, a hierarchical ensemble machine learning approach that involved the use of RF, SVR, and DNN as base models, while the meta-model was XGBoost. Research Limitation: UMa and RMa scenarios exhibit higher attenuation due to environmental and distance effects. The analysis focuses only on four deployment environments: Indoor, Urban Microcell (UMi), Urban Macrocell (UMa), and Rural Macrocell (RMa) scenarios. Computational complexity and training overhead associated with hybrid stacked learning algorithms may increase for large-scale network deployments. Findings: The hybrid model proposed here significantly improves predictive performance compared to other models, achieving lower R², RMSE, and MAE. Scenario analysis showed that the Indoor and UMi scenarios have high spectral efficiency and low path loss, whereas the UMa and RMa scenarios exhibit higher attenuation due to environmental and distance effects. Practical Implication: The model presented in this paper presents useful applications in network design, optimisation, and resource management for future wireless networks. Social Implication: Improved communication performance and efficient utilisation of available bandwidth in future wireless networks. The development of high-speed and reliable next-generation wireless communication systems improves overall connectivity for society. Originality/Value: The proposed model incorporates the dynamic influence of environmental and propagation variations on the characteristics of 28 GHz mmWaves, thereby improving estimates of network performance.