Research On Compressed Sensing Based Wideband Spectrum Sensing Method

With the rapid development of the wireless mobile communication industry,various new services are emerging.However,due to the static usage policy of the spectrum,the available spectrum resources for the new services are gradually reduced.The scarcity of spectrum has become the main bottleneck restricting the development of wireless communication.Since much of the licensed spectrum is underutilized,spectrum sharing becomes a promising technology to solve the spectrum scarcity problem.To meet the spectrum demands of so many emerging services,wideband spectrum sensing(WSS)has become an effective way to detect the available spectrum in spectrum sharing.The traditional sampling method based on Shannon-Nyquist sampling theorem requires a very high sampling rate in WSS,which is difficult for the hardware implementation of the receiver.In recent years,compressed sensing(CS)has been proposed,which can acquire high-dimensional sparse signals by low-dimensional linear measurement.Since the sparsity of wideband spectrum caused by the low utilization of licensed spectrum,CS can be used to WSS to realize sub-Nyquist sampling of wideband signals.This thesis mainly studies the WSS technology based on CS,in which the sparsity of wideband spectrum is utilized to achieve the goals of low sampling overhead,high sensing accuracy and low power consumption for WSS.Firstly,aiming at the influence of the coherence of measurement matrix on wideband compressive spectrum sensing(WCSS),a general measurement matrix optimization algorithm and a WCSS algorithm are designed,which can provide support for the following research.Then,aiming at the adaptive WCSS with low sampling rate scenario and 1-bit quantization with low samples power scenario,two WCSS method are proposed respectively.The main contributions of this thesis are as follows:Firstly,the theoretical model of wideband compressive spectrum sensing(WCSS)is proposed in this thesis.At the beginning of this part,the theoretical model of CS is introduced in detail.Then,the block-sparse CS model that can effectively reduce the algorithm complexity in WSS is introduced.Based on this,the signal of WSS is modeled as a multi-band signal model based on the block-sparse signal model.And then,the multi-coset sampling system is introduced,and its undersampling process is modeled as a standard CS problem through theoretical derivation.Finally,based on the block-sparse CS,the solution of multi-coset based WSS problem can be regarded as to search the support of spectrum block.These theoretical models provides support for the research in subsequent parts.Secondly,to solve the unstable performance problem of WSS caused by the coherence of the multi-coset’s measurement matrix,this thesis proposes a WCSS method based on the measurement matrix optimization and matching pursuit algorithm.To decrease the coherence of the multi-coset’s measurement matrix,a measurement matrix optimization algorithm based on random search is proposed.Minimizing the coherence coefficient of the measurement matrix is the optimization goal of this algorithm and the result of this algorithm is a suboptimal measurement matrix which can approximate the optimal through multiple random searches.To reduce the sensitivity of WSS to the coherence coefficient of measurement matrix,a dual-threshold WSS algorithm based on the matching pursuit algorithm is proposed.The algorithm only uses the atom selected in once iteration to update the residual,so more energy of other atoms can be retained,reducing the influence of coherence on subsequent iterations.Besides,the noise power of the reconstructed signal is obtained through theoretical derivation,and this information can be used to effectively suppress the high false alarm brought by the matching pursuit algorithm.Thirdly,aiming at the mismatch problem between the sampling rate and the signal sparsity caused by the random time variation of signal,this thesis proposes an adaptive WCSS algorithm based on leave-one-out cross-validation.The sampling rate required to reconstruct the signal when the support set is known is half that of the case when the support set is unknown.Based on this,the proposed algorithm first uses all sampling resources to estimate the support of the spectrum,and then divides the samples into a reconstruction set and a validation set,which avoids the extra cost of sampling resources for the cross-validation.To improve the accuracy of the reconstructed signal,only one coset is selected as the validation set,and all other cosets are used as the reconstruction set to reconstruct the signal.At the same time,to guarantee the accuracy of the cross-validation,different cosets are selected as the validation set repeatedly to perform cross-validation several time,which can improve the accuracy of the cross-validation.Fourthly,aiming at the problem of severe quantization noise caused by 1-bit quantized multi-coset sampling system,this thesis proposes a subspace decomposition based WCSS algorithm.Based on the Bussgang theorem,the 1-bit quantized samples are linearized,so the linear relationship between the original signal and the 1-bit quantized samples is constructed.Then,by eigenvalue decomposition of the autocorrelation matrix of the samples,the corresponding eigenvectors are decomposed into signal subspace and noise subspace.After that,a new linear system is constructed by the signal subspace for spectrum sensing,which reduce the interference of noise on WSS.To determine the dimension of the signal subspace,the exponential fitting test algorithm is introduced into the 1-bit quantization system.It can effectively estimate the dimension of the signal subspace,which provides a strong support for 1-bit quantized WCSS.