CAJ | 학술논문

提出了一种基于因子图的分布式变分稀疏贝叶斯压缩感知算法。该算法利用因子图和变分方法将全局感知问题分解为简单的局部问题，通过认知用户邻居间的置信传播实现“软融合”，使每个认知用户能够获得全局最优估计。且充分利用邻居间传递的信息所具有的时间和空间二维相关性，提高认知用户在低信噪比下的感知性能。同时，算法在迭代过程中自适应地删除不收敛的超参数及对应的基函数，降低通信负载。实验结果表明：该方法在低采样率和低信噪比下有较好的感知性能。
제출료일충기우인자도적분포식변분희소패협사압축감지산법。해산법이용인자도화변분방법장전국감지문제분해위간단적국부문제，통과인지용호린거간적치신전파실현“연융합”，사매개인지용호능구획득전국최우고계。차충분이용린거간전체적신식소구유적시간화공간이유상관성，제고인지용호재저신조비하적감지성능。동시，산법재질대과정중자괄응지산제불수렴적초삼수급대응적기함수，강저통신부재。실험결과표명：해방법재저채양솔화저신조비하유교호적감지성능。
A distributed variational sparse Bayesian compressed spectrum sensing algorithm based on factor graph was proposed, which decomposed the global spectrum sensing problem into local problem based on factor and variation. Be-lief propagation was used for the statistical inference of the spectrum occupancy, to implement the “soft fusion”. The temporal and spatial correlation information providing two-dimensional redundancies was exchanged among cooperative cognitive users to improve the detection performance under low SNR. Meanwhile, the algorithm prunes the divergence of hyper-parameters and the corresponding basis functions for reducing the load of communication. The simulation results show that this method can effectively achieve performance of spectrum sensing under a low sampling rate and the low SNR.