Investigation and implementation of different deep learning techniques to increase the spectrum sensing performance in cognitive radio

Abstract

The cognitive radio network for the next-generation of wireless newlinecommunication networks is made possible in large part by spectrum sensing. newlineEnergy detectors, matching filters, and cyclo-stationary are just a few of the newlineapproaches that have been suggested throughout the years. Nevertheless, there newlineare a number of problems with these approaches. Because cyclo-stationary newlinedetector are incredibly problematical and similar strainers require a prior newlineunderstanding of the major user signals, energy detectors operate poorly when newlinethe signal-to-noise ratio (SNR) is changing. Furthermore, the effectiveness of newlinethese approaches detection exclusively relies upon the accuracy of the newlinesensing because they rely on limits based on specific signal-noise assumptions newlinemade by the model. So one of the top problems for wireless scientists is still newlinedeveloping trustworthy and smart spectrum detection technology. Machine newlineLearning (ML) and Deep Learning (DL) approaches have lately received newlineattention in the construction of extremely accurate spectrum sensing models. newlineThe computational difficulty and high rate of misclassification of multi-layer newlinealgorithms for learning nevertheless make them unsuitable for processing newlinetime-series information. The study suggests a hybrid approach that combines newlineLong Short Term Memory (LSTM) and Extreme Learning Machines (ELM), newlinelearning time-dependent characteristics from spectral information while also newlineutilising additional external behaviour statistics like energy, distance, and newlineduty-cycle duration to improve detecting performance. On the basis of newlinespectrum data for several radio innovations collected with a Raspberry Pi newlineModel B+ and a demonstration test bed for GNU-radio, the suggested method newlineis verified. newline