Research On Key Technologies For Compressed Wide-Band Spectrum Sensing

Cognitive radio (CR) is an emerging technology that can promote the spectrum utilization ratio efficiently. According to the principle of CR, the dynamic spectrum access (DSA) scheme will be introduced into the current wireless communication systems. Through allowing secondary users (SU) to perform spectrum sensing and to access the "spectrum holes" opportunistically without causing interference to the primary users (PU), CR can alleviate the contradiction between the shortage of allocable spectrum and the underutilization of wide-band spectrum under current spectrum policies effectively. In recent years, wide-band spectrum sensing receives more and more attention. It can offer more spectrum opportunities for SU in the wide-band regime, thus it is superior to conventional narrow-band spectrum sensing methods. However, current wide-band spectrum sensing schemes all follow the classical Nyquist sampling theorem, which requires sampling the wide-band signal at a rate twice of its highest frequency. Such high sampling rate potentially imposes great pressure on current hardware devices and restricts wide-band sensing to be implemented in practice. Fortunately, due to recent research advances in the signal processing field, compressed sensing (CS) combines sampling and compression together by exploiting the sparse structure of the signal, which can achieve the signal acquisition procedure at sub-Nyquist sampling rate. Viewing the underutilization status of wide-band spectrum that presents sparseness, compressed spectrum sensing is deemed as a promising solution to sample and detect the wide-band signal efficiently, which can reduce the sampling burden significantly. Based on CS, this dissertation mainly investigates the key technologies for compressed wide-band spectrum sensing that incorporate low-rate sampling, signal recovery and detection, and proposes several effective sensing algorithms, which lays the foundation for its future application in cognitive wireless communication systems. The main contents and contributions of this dissertation are summarized as follows: (1) In the research on streaming compressed spectrum sensing with invariant spectrum support set, we first analyze the drawbacks of traditional compressed spectrum sensing schemes for single frame signal, whose sensing performance is susceptible to the noise uncertainty, and then establish the streaming CS model based on analog to information converter (AIC), which samples the wide-band signal at sub-Nyquist rate continuously and achieves the signal recovery procedure by utilizing the correlation between adjacent signal frames. Specifically, we propose a probabilistic greedy pursuit (PGP) algorithm to recover the signal when its sparsity is unknown. Through dynamically updating the support confidence coefficients based on greedy pursuit with the stream of the signal, the algorithm can achieve blind streaming spectrum estimation effectively. Simulation results validate that PGP can guarantee good signal recovery accuracy and robust detection performance without the priori sparsity knowledge, and present computational advantage as well.(2) In the research on adaptive streaming compressed spectrum sensing with time-varying spectrum support set, we first discuss the open questions related to adaptive CS. When the signal sparsity is not acquired as a priori information, adaptive CS should find ways to adjust the sub-Nyquist sampling rate dynamically and to reconstruct the signal blindly. Based on the streaming CS and time-varying signal model, we propose an adaptive streaming compressed sensing (ASCS) scheme, where the sampling rate is determined by comparing the correlation between adjacent channel occupancy information (COI) with a certain threshold, and then the signal recovery is achieved by first estimating its sparsity and then utilizing the iterative greedy pursuit. Besides, ASCS is analyzed by employing restricted isometry property (RIP) in detail. Simulations show that ASCS can not only adjust the sampling rate adaptively when the spectrum support set is time-varying, but also guarantee robust detection performance.(3) With respect to the fusion schemes for cooperative compressed spectrum sensing, we first establish the system model based on distributed cognitive radio network scenario, and then highlight the common spectrum support set property of the signals sensed by different SU. It is this correlation property that can improve the cooperative spectrum sensing performance remarkably. Since the sensing channel state information (CSI) can be available by SU by their channel estimation efforts or not, we categorize the fusion schemes into spectrum estimation fusion and spectrum detection fusion, and propose differential cooperative compressed spectrum estimation (DCCSE) scheme and support set fusion-based cooperative compressed spectrum detection (SSF-CCSD) scheme respectively. Simulations have testified the effectiveness and superiority of the above two fusion schemes in the aspect of convergence speed, communication load and computational complexity.(4) With respect to cooperative compressed spectrum sensing for block sparse signals, the block-based CS theory is elaborated firstly, and then we point out that traditional block-based recovery algorithms fail to consider the imperfection of block sparse signals in practice. For example, multi-band signals with unequal bandwidths may lead to partially occupied blocks with both zero and nonzero elements. Although such problem can be solved by modifying current recovery algorithms, the sub-band edges should be acquired as a priori information. To find a general solution, we propose an atom selection scheme by employing binary tree-based iterative search, and analyze its robustness based on the null space property of the measurement matrix as well. Furthermore, according to the distributed and centralized cognitive radio network models, we present binary tree-based block adaptive matching pursuit (BT-BAMP) and simultaneous binary tree-based block adaptive matching pursuit (S-BTBAMP) respectively. Simulations show that the proposed algorithms are suitable for the recovery of general block sparse signals, and the spectrum detection performance is also promoted remarkably through cooperation.