Research On Distributed Wideband Spectrum Compressive Sensing In Cognitive Wireless System

With the rapid development of wireless communication technology and rapidgrowth of the demand for wireless communication service, the wireless spectrumresource has become more and more scarce. Cognitive radio (CR) is the best solutionto solve the spectrum scarcity problem, which does not affect the communications forprimary users (PU), and applies dynamic spectrum access to use the idle spectrum.Spectrum sensing is the premise of CR technology. Therefore, the reliable, accurate,and fast wideband spectrum sensing has become one of the most important factors incognitive radio technology which can be applied to reality from theory.Compressive sensing can solve the problems of high-speed sampling andmassive information processing in wideband spectrum sensing. However, because CShas changed the time relationship of original signal, we can not directly obtain thespectrum information from observation sequence. As a result, it is difficult to directlyuse the observational data to operate spectrum sensing. Under this background, we putour research focuses on distributed wideband Spectrum Compressive Sensing incognitive wireless system. In order to overcome the disadvantages in the existingwideband spectrum sensing, this dissertation has explored some fast, accuratespectrum sensing algorithms on wideband frequency range. The main work and maincontributions of this dissertation are described as follows:(1) By exploring the spectrum sparsity of cognitive radio network, the jointsparse model (JSM) has been constructed based on distributed compressive sensing(DCS). In order to slove the problem that traditional wideband Spectrum CompressiveSensing scheme is difficult to determine the optimal observation sequence length, anadaptive sensing algorithm based on sequential compressive sensing (SCS) has beenproposed. This algorithm applies the feedback mechanism between secondary users(SUs) and secondary base station (SBS) to estimate the reconstruction error of thecurrent reconstruction sparse vectors, and through reconstruction error to determinethe optimal observation sequence length. At the same time, we construct the joint sparse model (JSM) based on distributed compressive sensing (DCS), with whichmulti-dimensional data can be joint reconstruction by using joint sparsity betweenSUs to improve the sensing performance and reduce the observation sequence length.The simulation results verify the accuracy of the estimation of reconstructed error.The simultaneous reconstruction algorithm improves the reconstruction performancecompared to the single reconstruction algorithm. Meanwhile, multi-user cooperativespectrum sensing based on joint sparsity model is also obviously improve the sensingperformance.(2) For optimizing the observation matrix in Bayesian compressive sensing(BCS), relevance vector machine (RVM) Gauss prior model and Laplasse prior modelhave been constructed for jointly sparse signals. Because the matching pursuitalgorithm is not suitable for Bayesian compressive sensing scheme, distributedrelevance vector machine (DRVM) and distributed Bayesian fast Laplace (DBFL)joint reconstruction algorithm are proposed based on the two sparse models above.We analyze the performance and connection between two algorithms. Through theanalysis of the sparse signal differential entropy, the expanded observation matrix canbe optimized by using the eigenvalue decomposition. The simulation results show thatthe reconstruction performance of DBFL algorithm is better than DRVM algorithm.Compared to the random propagation matrix, the optimization observation matrix cannot only improve the reconstruction performance, but also reduce the length ofrequired observation sequence.(3) Wideband spectrum sensing does not need all the information of thewideband sparse signal. In general, the reconstructed power spectrum density (PSD)is an intermediate variable to obtain the feature parameters. In order to directly obtainthe feature parameter that can be used as decision criterion, an incompletereconstruction wideband Spectrum Compressive Sensing algorithm based on randomfilter has been proposed. The algorithm utilizes the output data of random filter torealize the compressive sensing for channel energy. In order to optimize the existingjoint reconstruction algorithm, simultaneous compressive sampling matching pursuit(SCSMP) and simultaneous sparsity adaptive matching pursuit (SSAMP) joint reconstruction algorithm are proposed by analyzing the characteristic of the matchingpursuit algorithm. The application environment and the advantages and disadvantagesof two algorithms are also described. The simulation results show that, the incompletereconstruction Spectrum Compressive Sensing can significantly reduce the number offilters required by the system. Meanwhile, the reconstruction performance of SSAMPalgorithm is much better than SOMP algorithm, but slightly lower than SCSMPalgorithm. Because the SSAMP algorithm can be adaptive to the sparsity, it has morepractical value.(4) The observation sequences after compressed sampling can completelyreconstruct the sparse signal. Therefore, it contains all the information of the sparsesignal. Based on the fact above, a multi-resolution sensing scheme based on serialscanning structure is proposed. In the low resolution environment, non-reconstructedSpectrum Compressive Sensing (NRSCS) algorithm is utilized to directly analyse thescanning frequency band without reconstruction algorithm. Meanwhile, soft-decisionfusion rules are used to decide whether the scanning frequency band is occupied ornot. The confirmed blank spectrum can be accessed directly and the others need toemploy high resolution sensing, where PSD of the signal can be joint reconstructed byjoint sparsity between multiple SUs, in order to analyse the occupancy for eachchannel. The simulation results show that, the reconstruction performance of NRSCSalgorithm is related to the compression ratio and SNR and the soft decision fusion hasa better performance of spectrum sensing than the hard decision fusion. Although thespectrum sensing performance of the multi-resolution algorithm is slightly lower thanthe completed reconstruction method, it can significantly shorten the sensing time.