Wideb And Spectrum Sensing Based On Dirichlet Process Mixture Model

This paper mainly studies wideband spectrum sensing,which is also called multi-band signal spectrum sensing.It mainly means that there are multiple signals need to be perceived simultaneously in the detected frequency band.The narrowband spectrum sensing only detects the presence or absence of the signal,and only need to detect one signal,compared with narrow-band spectrum sensing,the wideband spectrum sensing need to detect multiple signals at a time,and the number of signals is not known when detecting.Therefore,traditional spectrum sensing methods such as energy detection,matched filtering detection and cyclostationary detection are not applicable.Some studies assume that the number of signals is known before further spectrum sensing,which is not in line with the actual situation.This paper proposes a clustering method for wideband spectrum sensing.The proposed clustering method is based on Dirichlet process mixture model,which can infer the number and carrier frequency of signals from the features extracted from the spectrum or cyclic spectrum of the signal.This algorithm does not need to know the number of existing signals clearly.In this paper,we assume that cluster parameters are extracted from the mixed model and the clustering parameters that are most suitable for the observed data are determined by Gibbs sampling.The specific research contents of this paper are as follows:First of all,there is no clear standard for the definition of the wideband spectrum sensing,and there is no unified definition of relevant parameters,in this paper,we design and illustrate the system model of wideband spectrum sensing.The bandwidth of the detection band,the signal bandwidth,the relationship between the signals,the form of the signal,and the definition of the signal-to-noise ratio in this paper are illustrated.Since this paper deals with the problem of wideband signals,the received signal is down-converted before being sampled.Secondly,this paper analyzes the wideband signals received under Gaussian channel,and compares the wideband spectrum sensing problem with the Dirichlet process mixture model.It proves that the Dirichlet process mixture model can be applied to wideband spectrum sensing problems without precise number of signals,it requires a maximum number of signals obtained by empirical statistics,and then renews the number of signals in a collapsed manner.In this paper,two kinds of signal characteristics,power spectrum characteristics and cyclic spectrum characteristics are selected.From the theoretical analysis,the anti-noise performance of the cyclic spectrum is better than the power spectrum,but the computational complexity of the cyclic spectrum is also higher than the power spectrum.In this paper,the features are extracted from power spectrum and cyclic spectrum respectively,and clustering is carried out based on Dirichlet process mixture model.The performance was simulated and analyzed,and it was verified that using cyclic spectrum for features had better performance and higher complexity.Finally,this paper analyzes the wideband spectrum sensing in fading channels,models the wireless signals received in non-frequency selective slow fading channels,and reduce the impact of signal fading by changing the way the secondary user sensing the signal from a single node to cooperative spectrum sensing.Although cooperative spectrum sensing can improve the sensing performance,but the amount of data becomes larger.This paper attempts to use the cyclic spectrum of compressed sensing sampling recovery for spectrum sensing,and simulates the performance of the cyclic spectrum and power spectrum of Nyquist sampling recovery,and simulates the impact of compression ratio on the performance of the algorithm.This paper simulates and analyzes the signal of the carrier frequency in the ISM band near 2.4G,And the algorithm of the Dirichlet process mixture model and the K-means algorithm are used for wideband spectrum sensing under Gaussian and fading channels respectively.Compared with the K-means algorithm,the algorithm does not need the number of known signals,but the performances are similar.This paper defines the success rate,error detection rate,missed detection rate and mean absolute error to measure the performance of the algorithm.