Study On Super-resolution For Radar Detection Of Group Targets

The detection of group targets by radar has always been a challenge in the field of radar signal processing.The Traditional high-resolution methods,such as MUSIC,ESPRIT and their improved algorithms,not only require a large distance between targets or require a large number of snapshots,but also face the challenge of model errors and low signal-to-noise ratio.This Thesis explores the super-resolution for radar detection of group targets based on the theory of sparse reconstruction,and obtains the following results:1.We model the radar echo signal,and convert the distance information of targets into frequency information after preprocessing（including pulse compression prepositioning,frequency dechirp,data sampling extraction,etc.）The preprocessing make the sparse reconstruction technology accurately matches the radar signal model,which provides the basis for the sparse reconstruction technology to be applied to superresolution for radar detection.2.Based on sparse Bayesian learning theory,a super-resolution processing method based on sparse Bayesian learning is proposed.Compared with algorithms such as MUSIC and ESPRIT,this algorithm requires fewer snapshots;compared with other spectral estimation methods or greedy pursuit sparse reconstruction algorithms,this method still has good resolution and stability when the signal-to-noise ratio is low.3.In order to solve the grid mismatch problem caused by the grid in the traditional fixed grid sparse reconstruction technology,the methods of gridless and grid parameterization are adopted.In the gridless method,Atomic Norm Minimization（ANM）is applied to solves the problem of grid mismatch at the root,but the ANM algorithm is related to the local coherence between the atoms that make up the real signal,The resolution performance is not good in the detection of group targets;in the grid parameterization method,techniques such as Super-Resolution Iterative Reweighted（SURE-IR）,Prior-Knowledge Aided Super-Resolution Iterative Reweighted（KA-SURE-IR）,Orthogonal Maitching Piursuit（OMP）and Dynamic Parameterized L1-Regulation（DPL1）are used to alleviate grid mismatch problem.Compared with the traditional grid sparse reconstruction technology,the resolution and stability are greatly improved.4.In order to solve the problem of poor resolution of the OMP algorithm,we combine the OMP with the DPL1 algorithm.Firstly,we use the OMP to roughly estimate the group targets,and then use the gradient descent of DPL1 algorithm to do superresolution.So we propose a group targets super-resolution processing,which based on OMP and DPL1 algorithm.Compared with the single OMP processing,this processing has a great improvement in the resolution performance.