Censado de espectro utilizando inteligencia artificial para redes móviles LTE y 5G.

Abstract

The project originated from the need for a tool to monitor and manage the radio spectrum. The study used state-of-the-art spectrum sensing techniques employing artificial intelligence to develop an effective system. The methodological approach was predominantly applied, combining the theoretical knowledge acquired with the practical implementation of the system. The process was divided into sections, ranging from signal reception by central frequencies based on information provided by the ARCOTEL on spectrum allocation for mobile networks to mobile identification networks using convolutional neural networks (CNNs), focusing on LTE and 5G technologies. The results obtained, analyzed through graphs generated by executing code in MATLAB, reveal that the accuracy of spectrum sensing depends on the power of the captured signals. This dependence links to the inherent limitations of the hardware used (USRP B210). Likewise, it is observed that identification is influenced by the complexity of the CNNs employed, with more complex networks achieving accuracy. It notes that CNNs, although a powerful tool, do not achieve 100 % accuracy. There is evident confusion between noise, 5G signals, and signals of unknown origin. It is highlighted that user-trained CNNs demonstrate higher accuracy comparing pre-trained ones. Although, the required training time can be significant, depending on the complexity of the network and the dataset used. In conclusion, implementing a spectrum sensing system with artificial intelligence represents a noteworthy advancement. It is necessary to recognize its limitations and continue researching to improve accuracy and robustness.