Intelligent Spectrum Sensing And Resource Allocation In 5G Cognitive Wireless Networks

As smart user devices,self-driving cars,and the Internet of Things(IoT)rise in popularity,spectrum scarcity has emerged as one of the most significant challenges for the fifth-generation(5G)wireless communication networks.With the growing demand for increased network capacity and more data transmissions,cognitive radio(CR)technology is ever more relevant today.Traditional static spectrum allocation is no longer a feasible option.Through dynamic spectrum sharing(DSS),CR users can tap into unused licensed spectrum bands,which can improve spectrum utilization efficiency and fuel scarce spectrum applications.Motivated by the superior performance of artificial intelligence(AI)in various research paradigms,this thesis aims to improve spectrum utilization efficiency in 5G CR networks(CRNs)by developing intelligent spectrum sensing and resource allocation schemes.Therefore,besides using a classical approach,this thesis mainly focuses on the AI approaches to optimize the spectrum efficiency in 5G CRNs.Firstly,we propose an improved method based on the generalized likelihood ratio test(GLRT)framework and estimator-correlator to detect primary user(PU)and address the noise uncertainty challenges in CRNs.In this context,three robust schemes are proposed in a multiple-input multiple-output(MIMO)based cognitive wireless sensor networks(WSNs).The PU detection problem in these schemes is formulated as a hypothesis-testing problem based on the improved GLRT and estimator-correlator methods with few observation samples and in the critical low signal-to-noise ratio(SNR)conditions.These schemes perform efficient spectrum sensing with known and unknown noise distributions and channel gain statistics.Then,we focus on the AI approach and propose three schemes.(?)A machine learning(ML)based joint spectrum sensing and allocation scheme is proposed for IoT devices in cellular CRNs,called as CR-IoTNet.It allows secondary user(SU)IoT devices to detect the existence of PUs in a large-scale,dynamic 5G network of multiple PU base stations.Through multi-dimensional feature learning,CR-IoTNet efficiently identifies the unoccupied channels in the multiband PU spectrum and allocates them to the SU-IoT devices requesting spectrum access.The simulation results show that our proposed framework achieved 95.11%accuracy,with 92.67%true positive rate and 96.34%true negative rate in terms of spectrum sensing and optimal allocation of vacant channels.(ii)A deep learning(DL)based PU activity-aware spectrum sensing scheme(dubbed as PU-Net)is proposed that allows SUs to detect and classify the PU transmitted signal patterns efficiently.PU-Net is implemented in a 5G smart city CRN,where unmanned aerial vehicles(UAVs)serve as aerial PU base stations to provide adaptive wireless coverage for IoT users in the primary and secondary systems.Unlike traditional schemes,PU-Net is a data-driven approach that does not require a priori statistics about the primary network(such as signal-noise distribution or prior probability).It learns the PU activity patterns by exploiting the inherent multilevel spatial-temporal features from input signals.The simulation results validate that PU-Net outperforms other benchmark detectors,achieving 99.74%accuracy,99.78%recall,and 99.70%precision in predicting the spectrum occupancy.Moreover,compared to existing DL methods,it has an average 85.60%performance gain in prediction accuracy at low SNR(-25 dB to-20 dB).Besides,when evaluated on real-world signals,it achieves 0.72%gain in sensing accuracy at-20 dB SNR.(iii)A congestion-aware hybrid learning scheme is proposed for the Internet of Vehicles(IoVs)in a cluster-based CR-assisted IoV network(named as CRAV-Net).The proposed scheme jointly optimizes spectrum sensing and resource allocation in a dynamic,high-mobility environment of 5G-based IoV networks.CRAV-Net leverages DL and ML techniques to improve spectrum efficiency and achieve optimum utilization of vehicular network resources.The proposed DL-based spectrum sensing model dynamically learns the multiscale spatial and temporal graphical features from input spectrograms through layer-by-layer propagation and efficiently predicts the spectrum occupancy.Then,to assign the detected vacant channels,an ML-based network node selection mechanism is proposed for SU vehicles to access the spectrum using multi-dimensional feature learning in order to improve the spectrum efficiency of IoV networks.Simulation results demonstrate that CRAVNet achieves an overall accuracy of 99.74%in terms of spectrum sensing using the custom dataset.Further,it outperforms state-of-the-art DL methods by 12.60%at-25 dB SNR while reducing the inference time by 33.33%.In addition,it achieves a 0.81%performance gain in spectrum sensing accuracy when evaluated on realworld signals.Furthermore,it achieves 98.45%mean accuracy in the context of optimal network node allocation with 0.63%and 18.32%performance gains in terms of accuracy and allocation time,respectively.