基于广义模式耦合稀疏Bayesian学习的1-Bit压缩感知

1-Bit compressive sensing based on generalized pattern-coupled sparse Bayesian learning

在1-Bit压缩感知(compressive sensing,CS)框架下,将信号的稀疏结构先验引入广义稀疏Bayesian学习(generalized sparse Bayesian learning,Gr-SBL),研究基于Gr-SBL的1-Bit CS重构。将广义线性模型与模式耦合稀疏Bayesian学习相结合,提出了一种基于广义模式耦合稀疏Bayesian学习1-Bit CS重构算法,简称为1-Bit Gr-PC-SBL算法。该算法将1-Bit CS重构问题迭代地分解成一系列标准CS重构问题,在信号稀疏模式未知的情况下,基于模式耦合稀疏Bayesian学习实现信号重构。进而,引入阈值自适应的二进制量化,设计了自适应阈值的1-Bit Gr-PC-SBL算法,进一步提升了算法的信号重构性能。

Under the framework of 1-Bit compressive sensing(CS),the signal’s sparsity structure prior is introduced into generalized sparse Bayesian learning(Gr-SBL),and the reconstruction of 1-Bit CS based on Gr-SBL is studed.The generalized linear models are combined with the pattern-coupled sparse Bayesian learning,and the 1-Bit CS reconstruction algorithm based on generalized pattern-coupled sparse Bayesian learning is proposed,which is shortened to 1-Bit Gr-PC-SBL algorithm.This algorithm iteratively reduces the 1-Bit CS reconstruction problem to a sequence of standard CS reconstruction problems,and realizes signal reconstruction based on pattern-coupled sparse Bayesian learning,while the signal’s sparse patterns are entirely unknown.Furthermore,binary quantization with adaptive thresholds is introduced,and a 1-Bit Gr-PC-SBL algorithm with adaptive thresholds is proposed,which can further improve the reconstruction performance of the algorithm.