Research On Compressed Spectrum Sensing In Wide-band Cognitive Radio

Wide-band cognitive radio is capable of searching a wide frequency band for avail-able spectrum resources, providing secondary users with more opportunistic spectrumaccess. However, capturing wide-band signals requires ultra high sample rate analog-digital converter (ADC) in the radio front end. With state of art of ADC manufacturecapability, it is extreme expensive to make such a high speed ADC to implement wide-band spectrum sensing. Compressed sensing provides a novel method to capture thespectral activities within a wide frequency range using a sub-Nyquist sampling rate andto reconstruct the spectrum signal with a high probability. This method is most ef-fective when only a small fraction of the wideband spectrum is occupied by PU, i.e.,the signal is sparse in the frequency domain. This requirement is readily satisfed inCR networks, since the PUs typically do not occupy the entire licensed frequency bandat one time, thus providing sparsity in the frequency domain. However, compressedsensing is sensitive to noise. When the sensed signal has low signal noise ratio (SNR),the compressed spectrum signal reconstruction may fail. The failure of the compressedreconstruction will cause severe interference to primary users. It is very important toenable compressed spectrum sensing to be robust under noisy sensing environments.On the other hand, the sparsity of the sensed spectrum afects the performance of com-pressed spectrum sensing. When primary users occupy most parts of frequency band,the spectrum will not be sparse anymore causing failure of compressed spectrum sens-ing. Thus, the interference caused by the spectrum sensing error signifcantly violatesthe operation of PUs which is strictly forbidden. It is an essential task to detect thenon-sparsity of the sensed spectrum to provide sufcient protection for primary users.Motivated by these challenges in compressed spectrum sensing, this thesis studieshow to implement multi-user cooperative compressed spectrum sensing to improve sens-ing performance under homogeneous and heterogeneous network structure. Moreover,based on cooperative compressed spectrum signal reconstruction, we propose a collab-orative non-sparsity protection (CNSP) scheme. Compressed sensing based wide-bandspectrum sensing is one of the research frontiers on wide-band cognitive radio technology.The innovative research achievements of this thesis are briefy listed as below:1. Space-time cooperative Bayesian compressed spectrum sensingA distributive probabilistic Bayesian compressed spectrum sensing algorithm (ST-BCSS) is proposed based on Gaussian Process (GP) to improve the detection perfor-mance of compressed spectrum sensing under low SNR conditions in wide-band cognitive radio networks. A hierarchy normal distribution probabilistic model is applied to rep-resent the compressed spectrum reconstruction. The model parameters are exchangedamong the cognitive radios as cooperation information, and then utilized to implementcompressed spectrum sensing based on the local compressed sensing data. This dif-fers from existing cooperative spectrum sensing methods, which uses the fnal sensingdecisions or the raw sensing data as the cooperation information. By exchanging mod-el parameters, the proposed ST-BCSS algorithm efectively reduces the negative efectsfrom those cooperating neighbors experiencing low SNR sensing environments hence thedetection performance is improved. Simulations show that PBCS achieves a detectionrate over0.9for a false alarm of0.1when the SNR is-5dB.2. Non-sparsity ProtectionA collaborative non-sparsity protection (CNSP) scheme is proposed. In generalcase, there exists strong correlation of the probabilistic model parameters of two near-by cognitive radios. However, when the primary users occupy most sub-channels, thecompressed spectrum reconstruction of each cognitive radio tends to become random.Consequently, the correlation of the probabilistic model parameters of nearby cognitiveradios reduces. Therefore, the sparsity of the sensed spectrum can be estimated bythe correlation of the probabilistic model parameters of nearby cognitive radios. Thesimulation results demonstrate that the proposed CNSP scheme can efectively identifythe failure of spectrum sensing and avoid interference to primary users.3. Auto-clustering Collaborative Compressed Spectrum SensingAn auto-clustering collaborative compressed spectrum sensing (ACCSS) algorith-m is proposed to collaborate in a heterogenous spectrum environment which is one ofmajor challenges in wide-band spectrum sensing. Compressed spectrum sensing is apromising technology for wide-band signal acquisition but it requires efective collab-oration to combat noise. However, most collaboration methods assume that all thesecondary users share the same occupancy of primary users, which is not feasible in aheterogenous spectrum environment where secondary users at diferent locations maybe afected by diferent primary users. A hierarchal probabilistic model is proposed torepresent the compressed reconstruction procedure, and a Dirichlet process mixed modelis introduced to cluster the compressed measurements. Cluster membership estimationand compressed spectrum reconstruction are jointly implemented in the fusion center.Based on the probabilistic model, the compressed measurements from the same clus-ter can be efectively fused and jointly reconstruct the corresponding PU’s spectrum.Consequently, the spectrum occupancy status of each PU can be attained. Numericalsimulation results demonstrate that the proposed ACCSS algorithm can efectively esti- mate the cluster membership of each secondary user and improve compressed spectrumsensing performance under low SNR.4. Belief Propagation based Cooperative Compressed Spectrum SensingMost cooperative spectrum sensing methods assume that all the secondary usersexperience the same occupancy of primary users, which may not be feasible in a het-erogenous spectrum environment where secondary users at diferent locations may beafected by diferent primary users. A belief propagation based cooperated spectrumsensing is proposed in this thesis to provide efective cooperation. Firstly, a proba-bilistic graphical model is proposed to represent and fuse multi-prior information fromone hop neighboring secondary users. we model the correlation among the neighboringcognitive radios as pairwise Markov random felds (MRF) and apply Belief Propagation(BP) to fuse the prior information from neighboring nodes for local compressed spec-trum sensing. BP is applied to fuse the prior information from the cooperative cognitiveradios. Numerical simulation results demonstrate that the proposed BP based cooper-ative compressed spectrum sensing can efectively achieve cooperation in heterogenousenvironments and improve performance of compressed spectrum sensing under a lowsampling rate and low signal-to-noise ratio (SNR), compared with other distributedcooperative compressed sensing methods proposed in the literature.