Research On Spectrum Sensing Technology Based On Transfer Learning And IRS In Cognitive Radio Networks

The emergence of new applications and technologies has driven the demand for wireless spectrum,and improving spectrum utilization has become a major goal for researchers in the context of increasingly constrained modern wireless communication resources.Among them,both deep learning and IRS are widely studied as cutting-edge research methods today.In this paper,single-node spectrum sensing based on deep learning and multi-node spectrum sensing based on IRS are considered as the research basis,and two schemes that can improve spectrum sensing capability and efficiency,and verify them through engineering practice and simulation analysis are proposed.The main work accomplished in the paper is as follows.(1)In this paper,firstly,single-node and multi-node spectrum sensing in cognitive radio are studied and researched,and the advantages and shortcomings in the deep learning-based approach and in the IRS-based cognitive radio spectrum sensing are found.Subsequently,improvement studies,engineering experiments and simulation validation are conducted for them,respectively.(2)The efficiency of deep learning-based cognitive radio spectrum sensing is improved by fine-tuning methods.For the convolutional neural network-based model used for spectrum sensing,firstly,the performance of the model based on simulation data and the model based on real data are analyzed and compared.Due to the limitation of the number of real-time real data in practical applications,fine-tuning is applied to the spectrum-aware model,and a fine-tuning method based on real data is proposed to solve the problem of long training time for large real data or training without generalization for small amount of data.Finally,a comparative analysis is carried out for the fine-tuning of different layers of convolutional neural networks.The simulation results show that the fine-tuning method of using a small amount of real data for the convolutional neural network trained based on simulated data can obtain a higher detection probability without any feature analysis of the received data,which can greatly reduce the time and expense of online training.(3)The spectrum sensing efficiency of IRS-based cognitive radio networks is enhanced by designing a threshold adaptive method.This paper proposes a threshold adaptive algorithm for collaborative spectrum sensing in IRS-enhanced cognitive radio networks,where the signal-to-noise ratio of reflected signals varies too much and the traditional uniform threshold design algorithm has great limitations.The threshold adaptive design can make sub-users with different SNR set different energy detection thresholds according to their SNRs,thus improving the spectrum sensing performance of each sub-user and enhancing the performance of collaborative spectrum sensing.Simulation verifies the convergence and effectiveness of the algorithm.In summary,this topic investigates the spectrum detection methods for single-node and multi-node in cognitive radio,identifies the problems in the newly proposed methods,and conducts in-depth research and improvement for their problems,respectively.