Modelo de detección de espectro para redes móviles de radio cognitiva mediante el uso de técnicas de Inteligencia artificial

Abstract

Traditional fixed-allocation models for the radio spectrum have become a source of inefficiency for wireless networks, where congested bands coexist with others that are barely used. Conventional detection methods, based on energy thresholds, are insufficient for today's dynamic and noisy environments, severely limiting resource optimization. In response to this problem, this work develops and validates a spectral detection model that applies artificial intelligence to achieve greater accuracy. For this purpose, an experimental system with Software-Defined Radio (SDR) was implemented to capture a dataset in a controlled environment. With this information, two classification models, Logistic Regression and Random Forest, were trained and compared to determine their ability to correctly identify channel availability. The results obtained demonstrate the effectiveness of this approach. The Random Forest model exhibited exceptional performance, achieving a 98% detection probability with a false alarm probability of only 2%. It is crucial to highlight that this reliability was maintained even in low signal-to-noise ratio scenarios, a known weakness of energy detectors. This research concludes that artificial intelligence offers a robust and practical solution for dynamic spectrum management, establishing a validated foundation for the design of future communication networks, which will necessarily be more efficient and adaptive.