Applied Research On Compressive Sensing For Wideband Spectrum Sensing

With the rapid growth of wireless communication, the conventional fixed spectrumallocation rules have resulted in both scarcity of available spectrum resources and lowspectrum usage efficiency in almost all currently deployed frequency bands. Cognitiveradio (CR), emerged as a kind of intelligent spectrum sharing technique, allows thecognitive/secondary users to access the licensed spectrum bands when they aretemporarily unoccupied by primary users. With the rapid growth of wirelesscommunications serves, CR needs to identify more potential available spectrum in orderto achieve high-speed wireless communications in future. Therefore, wideband spectrumsensing aimed at monitoring a very large contiguous or noncontiguous frequency range tofind the available spectrums is becoming a crucial issue in CR and has received muchattention. To achieve fast and accurate spectrum sensing over a wide bandwidth, typicallya corresponding large sampling rate is required, which is very challenging for practicalimplementation. Fortunately, a large part of the frequency range is vacant, that is, thesignal is frequency-domain sparse. We can use the recently developed compressivesensing/sampling to reduce the sampling rate by a large margin. It provides an alternativeapproach for fast and accurate wideband spectrum sensing with low cost of sampling.Although there have been a quantity of studies in wideband compressive spectrumsensing, more researches are needed especially for excellent wideband sensing algorithms.The main focus of this dissertation is compressive sensing and its application in widebandspectrum sensing. The main contributions are summarized as follows:1. Due to performance limitations of the L1-minimization/Basis Pursuit Denoising(BPDN) algorithm, a weighted-BPDN (WBPDN) algorithm is proposed toimprove the performance of wideband compressive spectrum sensing. TheWBPDN algorithm incorporates the statistical properties of spectrum usages toreweight the coefficients of the previous L1-minimization algorithm. There areempirical evidences of the fact that the proposed algorithm not only guaranteesaccurate spectral estimation from fewer measurements, but also outperformsBPDN and OMP in terms of measurements requirements and reconstructionerror. Moreover, WBPDN has faster convergence speed and significantly reducesthe computation time required by BPDN. Theoretical analysis demonstrates thatWBPDN also provides better performance than the original BPDN.2. The BPDN algorithm often suffers from heavy computational complexity whensolving massive recovery problems. Considering the real-time requirement ofspectrum sensing, we move forward to research the efficient greedy iterative algorithms for sparse recovery and compressive spectrum sensing. A new greedymethod, called Adaptive Sparsity Matching Pursuit (ASMP), is proposed. Unlikeanterior greedy algorithms, ASMP can extract information on sparsity of thetarget signal adaptively with a well-designed stagewise approach. Experimentsvalidate the proposed algorithm gives superior performance than BPDN andother greedy algorithms without prior knowledge of the sparsity level, whilemaintaining the low complexity of CoSaMP/SP. The applications of ASMP inwideband spectrum sensing also demonstrate that proposed algorithm givesbetter performance than the other algorithms.3. In wideband spectrum sensing, the wideband signal is of group sparsitycorresponding to the channel locations. Considering the sub-channel locationsare usually known in advance, a new greedy algorithm, called Group OrthogonalMatching Pursuit (GOMP) is proposed in order to exploit the characteristic ofgroup sparsity to reconstruct the wideband spectrum estimation. The GOMPalgorithm based on the principle of the sophisticated greedy pursuit algorithmincorporates a multipoint measurement of sub-channels to identify the signalsupport where active primary users are located. This endows the algorithm withaccuracy and robust for support identification and spectrum reconstruction.Hence, the proposed algorithm achieves fast and accurate compressive widebandspectrum sensing from far fewer measurements. A comparison with experimentshows the proposed algorithm outperforms traditional OMP algorithm and thefamous Basis Pursuit (BP) algorithm in terms of reconstruction errors anddetection accuracy with fewer measurements. Moreover, it is faster than the twoprevious algorithms. In addition, GOMP can be viewed as a variant version ofOMP when the knowledge of group information is unknown. It has potentialutilization in signal processing.These algorithms mentioned above are not limited to wideband compressivespectrum sensing. Although Several algorithm are motived by the fast and accuratewideband compressive spectrum sensing，they can also be extended to other applicationsfor sparse recovery with outstanding performance.