Theory And Methods Of Sparsity-Based Microwave Coincidence Imaging

As a noval radar imaging technique,radar coincidence imaging(RCI)does not depend on the relative motion between radar and target with significant potentials for high-resolution and interference suppression.Thus RCI can be used in some important applications,e.g.,military reconnaissance,precision guidance,high-resolution earth observation,disaster monitoring and security check.Based on the sparse prior of target,we investigate the resolution and waveform design theory of RCI,and study the RCI with model error,e.g.,off-grid error and gain-phase error.Besides,RCI method for extended target is proposed.Finally,the theory of RCI is enriched and improved further,and the system design and practical application of RCI are supported theoretically.In Chapter 1,the classification and challenges of conventional radar imaging techniques are summarized.Besides,the origination,characteristics and potential applications are introduced.Then,the research history of imaging mechanism,imaging system,resolution,radiantion source design and imaging methods is elaborated,and the development trend of RCI is prospected finally.In Chapter 2,the resolution of RCI is investigated.The resolution theory of RCI is established primarily based on the nominal resolution limit(NRL),statistical resolution limit(SRL)and effective-rank-based super-resolution measurement theory.NRL of RCI is obtained based on the spatial ambiguity function.Then,SRL of RCI is derived based on the statistical detection and estimation theory respectively,which enriches the conventional understanding of RCI.Finally,we employ the effective rank of reference matrix to measure the incoherence of radiation field from the perspective of matrix analysis,and then propose the super-resolution measurement method of RCI.In Chapter 3,the waveform design of RCI is studied to optimize the frequencyhopping(FH)code of FH waveforms.The condition number of reference matrix is used to measure the randomness of radiation field,and minimizing the condition number is employe as the objective function of waveform design.Accordingly,the waveform design method for RCI with model error is proposed.A modified simulated annealing algorithm is used to optimize the FH code,the condition number is decreased,and the imging sensitivity to model error is depressed.Next,a joint optimization method of FH code and line-of-sight(LOS)is proposed based on the fact that the incoherence of radiation field is sensitive to LOS.Thus the optimized waveform and LOS angle can be obtained,which provides a guideline for RCI system design and parameters setup.In Chapter 4,off-grid is invstigated.First,an accurate off-grid RCI model is established using first-order Taylor expansion.Then constrained Cramér-Rao bound and Bayesian Cramér-Rao bound are derived from sparse reconstruction and Bayesian statistics perspective,respectively.Next,three off-grid imaging algorithms are proposed in sparse reconstruction framework.Band-excluded locally optimized othogonal matching pursuit(BLOOMP)is proposed based on an improved OMP algorithm,which could reduce the requirement of incoherence of reference matrix,and performs well in the case of denser grids.The second algorithm is based on variational sparse Bayesian learing(VSBL),where off-grid imaging is modeled as a joint sparse recovery problem that recovers three groups of sparse coefficients over three known dictionaries.Then the problem is solved via VSBL to obtain better imaging performance.The third algorithm is based on group-sparse variational message passing(GS-VMP),where the off-grid imaging is model as a group-sparse reconstruction problem.Then the group-sparse coefficients are assigned a Gaussian-Gamma-Gamma(G-Ga-Ga)correlation prior,and VMP approach is employed to conduct the Bayesian learning.In Chpater 5,RCI with gain-phase error is investigated and reformulated as a joint sparse optimization problem.The scattering coefficients and gain-phase error are estimated simultaneously via alternative iterations.Accordingly,we propose three autofocus algorithms,i.e.,sparse auto-calibration imaging(SACI)algorithm,a SBLbased auto-calibration imaging algorithm(i.e.,self-calibration variational message passing,SC-VMP)and an ExCoV(expansion-compression variance-component)-based autofocus imaging algotithm.They share the silimar principles where the scattering coefficients are estimated using sparse reconstruction methods,and the gain-phase error is estimated using quasi-Newton method.Besides,the reference matrix is updated and then employed to reconstruct the target.OMP is used to reconstruct the target in SACI.In SC-VMP,the scattering coefficients are assigned a three-layer G-Ga-Ga prior,and then VMP is used as the reconstruction algorithm.The ExCoV algorithm used in AC-ExCoV focuses on the target sparsity which leads to faster implementation,this improvement is outstanding particularly in large-scale signal reconstruction problems.In Chapter 6,RCI for extended target is investigated.To model the extended target,both the sparsity and cluster structure of target should be considered.Herein,the “spike-and-slab” sparse prior is assigned to the extended target in Bayesian framework,which promotes the sparseness and the cluster structure simultaneously.Then a Markov chain Monte Carlo(MCMC)method based on Gibbs sampling is used to implement the Bayesian inference.This method is sensitive to noise with a high computational complexity,thus an adaptive clustered SBL(ACSBL)algorithm is proposed.To model the statistical dependencies among neighboring scatterers of extended target,a hierarchical correlated Gaussian prior model is introduced.Then target reconstruction and parameter optimization are implemented in variational Bayesian expectationmaximizetion(VBEM)framework.ACSBL could reconstruct the extended target well,without the requirement of any structure information about the cluster prior.Chapter 7 summarizes the content of this paper and proposes some key points about RCI that need to be investigated further.