1-bit Compressive Reconstruction,Compressive Diffusion And Variational Approximate Message Passing Detection

In recent years,to meet demand for the explosive wireless data traffic,new technologies of data compression have being studied to support the complex multi-media services with higher compression ratio and better recovery quality.These technologies break through the limitations of traditional theories such as Nyquist Sampling Theorem by exploiting structural properties inside the data as prior information in signal processing.Besides,quantization is inevitable in the implementation of data compression.In particular,one-bit quantization can be regarded as a method to compress the real-valued data at the extreme.The dissertation gives intensive investigation of signal recovery or information extraction from one-bit quantized compressed data,and proposes some novel strategies in different application scenarios as listed follows:For the design of robust reconstruction of one-bit compressive sensing,a parameterized recovery algorithm named Soft Consistency Reconstruction is proposed.Recognizing that re-construction is essentially an optimization problem,two parameters of its objective function can be adjusted flexibly to approximate a wide range of related algorithms.Through quantitative analysis,we further investigate the heuristic method to choose the optimal parameters given measurement number,noise model and noise level.Thereby,the proposed algorithm presents excellent performance in all noise level regimes.For the design of uplink detection in one-bit massive MIMO systems,to the best of our knowledge,bilinear generalized approximate message passing(BiG-AMP)and variational Bayesian inference(VBI)are the rare algorithms whose complexity is suitable for 1-bit mas-sive MIMO systems.By exploiting the advantages of both Big-AMP and VBI.a novel vertex-message based inference algorithm called variational approximate message passing(VAMP)is proposed for the one-bit quantized massive MIMO detection.On the one hand,the proposed algorithm succeeds to complete channel estimation,data detection and noise level estimation simultaneously with only binary measurements and relatively short pilot sequences at the re-ceiver.On the other hand,inserting variational approximation into the framework of AMP as the means to deal with non-linearity,it not only inherits BiG-AMP's advantages of closed-form expressions,low complexity,low pilot overhead and excellent performance,but also makes up for its shortcoming in convergence.Moreover,asymptotic analysis based on state evolution is performed to investigate the iterative behavior,and an analytical performance bound is provided in terms of mean square error from the state evolution analysis.For the design of distributed parameter estimation,the compressive diffusion strategy that achieve better trade-off in terms of the amount of cooperation and the required communica-tion load is studied.Unlike the full diffusion configuration,the compressive diffusion approach diffuses single-bit of information to the neighborhood.Most existing distributed parameter esti-mation algorithms based on Least Mean Square method are not friendly to the hybrid data types during the compressive diffusion,so that the cooperation is not beneficial in general unless the combination rule is chosen properly.Therefore,from the perspective of Bayesian inference in-stead,a parameter estimation algorithm is proposed for the compressive diffusion strategy in dis-tributed networks.Based on the vertex-message inference,the proposed algorithm is companied by an optimal combination strategy,which ensures that all nodes in the network can benefit from the extra information in one-bit exchange data to enhance local accuracy.