Evaluación de ruido e interferencia en localización de nodos IoT mediante aprendizaje automático

Abstract

CONDITIONING OF PROJECT PUBLICATION. Achieving precise indoor localization remains quite a challenging task for modern wireless networks, primarily because the environment itself often imposes severe conditions that limit performance. For this reason, this study proposes and validates a hybrid strategy, specifically designed to minimize positional error in complex scenarios. The research works under the hypothesis that, through intelligent processing, it is possible to effectively compensate for the limitations or shortcomings often seen in traditional geometric methods. To carry out this development, three-dimensional scenarios were modeled using ray tracing techniques in MATLAB, adhering strictly to the IEEE 802.15.4a standard to properly characterize Ultra-Wideband (UWB) channels. Essentially, large amounts of synthetic data were generated to feed the regression models, combining Time Difference and Angle of Arrival (TDoA/AoA) algorithms with Random Forest techniques. The results clearly evidenced that the fusion of sensory data helped to stabilize the measurements. Furthermore, thanks to the intervention of artificial intelligence, the Mean Squared Error (MSE) was drastically reduced from 1.12 meters to just 0.19 meters, demonstrating that the system was able to effectively overcome the effects caused by dynamic obstacles. It is concluded that the integration of predictive correction in navigation systems not only enables operations in highly saturated zones but also significantly optimizes the reliability of technological infrastructure. Keywords: Ultra-Wideband, Ray Tracing, Hybrid Algorithms, Indoor Positioning