基于因子图的马尔可夫压缩感知

Markov compressive sensing in Cognitive Radio using Factor Graph

压缩感知的研究对象是稀疏性的信号,这和认知无线电理论中机会主义的频谱占用环境一致,压缩感知在认知无线电中的应用可以得到比传统的探测方法更灵活的解决方案。本文假设认知无线电环境中要探测的信号为一阶马尔可夫过程,通过贝叶斯推理实现信号的探测。这种概率模型可以用因子图来表示,在因子图中通过节点之间的信息传递来实现信号的感知和重建。节点之间的联系可以在因子图中产生环,从而可以得到一种迭代更新的算法。实现结果显示这种算法可以得到更好的性能。

Because the requirement that the underlying signals should be sparse in compressive sensing is in line with the opportunistic spectrum occupancy of cognitive radio（CR）, methods have been proposed to apply compressive sensing in solving problems in CR. With this approach, an even more flexible and diversified sensing strategy instead of the conventional scheme such as filter-bank sensing mode can be casted. This paper assumes the underlying sparse signal as a first-order Markov process and models the spectrum sensing as Bayesian inference of the targeted signal. This kind of probabilistic model can be visualized by Factor Graph, which connects forward compressive sensing and backward signal rebuilding through message passing among nodes. This connection generates circles in Factor Graph and hence produces an iterative inference algorithm. Experimental result demonstrates the better performance of this algorithm.