Spectrum Sensing in Cognitive Radio for Internet of Things using Deep Learning Models

Abstract

An intelligent wireless communication system capable of learning from its surroundings is called cognitive radio. It permits Secondary Users to reuse the radio resources that are made available to them while avoiding damaging interference with licensed users. The key component of cognitive radio technology is spectrum sensing. The literature has become interested in the use of machine learning approaches for spectrum sensing. Enhancement of performance through further spectrum availability forecast and comprehension of the first activities of the primary user. Supporting many SUs at once will boost the speed of spectrum sensing and data transfer over the available, limited spectrum. The performance of current spectrum sensing techniques is constrained by severe channel conditions.With the advent of deep learning in various fields, we are trying to analyze the performance of spectrum sensing (SS) wherein deep learning method is adopted for data fusion. In this research, a data-driven deep learning model is suggested to automatically classify the received raw signal data, which is regarded as time-series data. In this paper, we have provided a comparative analysis of various deep learning models, including ResNet, VGG, LSTM, and MLP. The performance comparison was provided using various sample lengths and SNR values, both low and high. The ResNet model’s performance has produced the highest detection probabilities in both low and high SNR ratios. The simulation result of the proposed methodology would improve the system’s accuracy with a reduction in losses that occurred during the false alarm of prediction as well as an improvement in the probability of detection. Therefore, simulation results of the suggested methodology would lead to an improvement in the system’s accuracy, a decrease in losses from false alarms of prediction, as well as an increase in the likelihood of detection. Better spectrum sensing would be achieved through deep learning’s analysis of PU statistics