A Systematic Technical Review and Comparative Analytical Study of Extreme Value Theory-Based Algorithms for Ultra-Reliable Wireless Communications

Abstract

The rapid evolution of 5G and emerging 6G wireless networks has intensified the demand for ultra-reliable low-latency communication (URLLC), where stringent reliability requirements are often governed by rare but severe events such as deep fading, extreme interference, and simultaneous adverse channel conditions. Conventional statistical models, which primarily characterize average or central channel behavior, may provide limited accuracy when estimating the probability of such extreme events. This study presents a comprehensive review and comparative analysis of Extreme Value Theory (EVT)-based approaches for modeling and managing rare wireless-channel events in URLLC systems. The analysis examines Generalized Pareto Distribution (GPD)-based tail modeling, EVT-driven predictive interference management, EVT-enriched radio maps, GAN-assisted GPD parameter estimation, and Multivariate Extreme Value Theory (MEVT) for spatial-diversity-based rate selection. The reviewed results demonstrate that GPD-based models can closely approximate empirical extreme-channel behavior, with reported deviations of less than 10% in the evaluated conditions. EVT-based predictive interference management has also demonstrated substantial reliability improvements, including a reported reduction in outage rates of up to 100-fold and approximately 15% lower radio-resource usage compared with conventional approaches. Furthermore, EVT-enriched radio maps improve the prediction of spatial regions satisfying stringent outage requirements, while GAN-based GPD parameter estimation offers potential advantages under limited-data conditions. The MEVT-based analysis demonstrates that explicitly modeling joint tail dependence in spatial diversity systems can improve achievable transmission rates, reaching approximately 3.05 bps/Hz with six antennas compared with 1.95 bps/Hz for the conventional approach, representing an improvement of approximately 56% under the evaluated conditions. The findings indicate that EVT provides a valuable complementary statistical framework for addressing the rare-event reliability challenges of 5G and 6G URLLC networks. The study concludes by identifying the integration of EVT with generative artificial intelligence, adaptive learning, and higher-dimensional MEVT as promising directions for future ultra-reliable and data-efficient wireless communication systems.