Research On Spectrum Sensing Algorithm Of Cognitive Radio Based On Graph Signal Processing

As a dynamic spectrum allocation strategy,cognitive radio(CR)provides a new way to solve the shortage and underutilization of spectrum resources.Spectrum sensing is the precondition and basis of CR implementation.Its task is to detect whether the channel of primary-user is occupied in a narrow sense.In a broad sense,it also includes the recognition of signal parameters such as signal modulation mode and waveform.Generally,there is a non-collaborative relationship between the secondary-user and the primary-user.Therefore,how to design a low complexity spectrum sensing algorithm under the conditions of low signal-to-noise ratio(SNR),lack of prior information of the primary-user and no training samples has been a classic topic in CR signal processing.Graph signal processing(GSP),as a new signal processing technology,can be used to transform the signal from traditional time and frequency forms to the graph domain for processing.It has been preliminarily applied in signal detection,modulation recognition and other aspects.However,there are still some problems such as high computational complexity,poor performance at low SNR and strong data dependence of feature extraction.For this reason,based on GSP theory,combining with extreme value theory(EVT),topology data analysis(TDA)and other signal processing technologies,two CR spectrum sensing algorithms and two CR signal modulation recognition algorithms are proposed,which have better performance than the existing algorithms.The main research work of this paper is as follows:(1)A graph domain-based CR spectrum sensing algorithm is proposed based on the range sequence of power spectrum.With the help of the graph domain transformation and extreme statistics,the range sequence of power spectrum of the observed signal is converted to the graph domain.Accordingly,the CR spectrum is effectively sensed by evaluating the complete connectivity of the graph using the Gini coefficient.The simulation results show that when the SNR is-2d B,the detection probability of the algorithm has reached more than 90%,and the computational complexity is 16.4% of the existing graph domain sensing algorithm,which has better detection performance and lower complexity.(2)A spectrum sensing algorithm for CR is proposed based on graph domain characteristics of autocorrelation function.The autocorrelation function of the observed signal after removing the mean value is transformed into the graph domain,and the CR spectrum is sensed by examining the number of connected components of the graph,using the number of zero eigenvalues of the Laplacian matrix as the statistic.The simulation results show that when the SNR is-10 d B,the detection probability of the algorithm is close to 100%,which is better than the existing graph domain sensing algorithm.In addition,the computational complexity of the algorithm is about 1.7times that of the existing graph domain sensing algorithm,and its overall performance is better.(3)Based on graph and persistent entropy feature,a modulation recognition algorithm for CR signals is proposed.First,the candidate signal to be recognized is reconstructed by short-time filter,then its non-linear transformation spectrum and time-frequency curve are transformed by graph transformer and continuous homology filter respectively to obtain the features as Gini coefficient,sum of degree and persistent entropy.The recognition of four common signals,binary phase shift keying(BPSK),quadrature phase shift keying(QPSK),binary amplitude shift keying(2ASK)and hexadecimal quadrature amplitude modulation(16QAM),can be achieved.The simulation results show that when the SNR is-3d B,the average recognition accuracy of the algorithm is close to100%,the computation complexity is moderate,and the overall performance is better than the existing recognition algorithm based on sinusoidal component detection.(4)A modulation recognition algorithm for CR signals is proposed based on the maximum degree feature of graph.The amplitude spectrum,square spectrum and quartic spectrum of the reconstructed signals after short-time filtering are converted to the graph domain,and the modulation types of BPSK,QPSK,2ASK and 16 QAM signals are recognized using the maximum degree of graph as a feature.The simulation results show that when the SNR is-3d B,the average recognition accuracy of the algorithm is close to 100%,and its recognition performance and computational complexity are both better than the existing recognition algorithm based on sinusoidal component detecting.