Coherent Frequency-agile Radar Signal Processing By Solving An Inverse Problem With A Sparsity Constraint

Coherent frequency-agile radar (FAR) has low probability of intercept and excellentperformance on electronic counter-countermeasures and electromagnetic compatibility.Researches on signal processing with FAR are of great value for practical applications.This dissertation focuses on range-velocity joint estimation of multiple targets with FAR,and on how to exploit the sparse priori of the observed scene for FAR signal processing.Some results are provided as follows.The signal model of FAR returns is derived. The problem of jointly estimatingrange and velocity of targets are recast into solving a linear system of equations. Weexplain why the linear system of equations is undetermined in the case of FAR. When thetraditional matching filter is used to deal with the undetermined equation system, severesidelobes occur, which cause false alarms and mask weak targets. Since the number oftargets in the same coarse resolution range cell is usually rather small, the observed scenepossesses a natural sparse property. Compressed sensing technique can efciently solvea linear system of equations by leveraging the sparse constraint on the observed scene;thus, suppresses sidelobes significantly and reconstructs the scene correctly.The feasibility of compressed sensing technique in the case of FAR signal processingis theoretically validated. Properties of the measurement matrix in FAR are mathemat-ically analysed. A sufcient condition on the sparsity level of the observed scene andradar parameters is provided. When the sufcient condition is satisfied, compressed sens-ing achieves exact recovery of the observed scene in noiseless cases, or stable recoveryin noisy cases. Experiments with both synthetic and real data demonstrate the efciencyof compressed sensing.The model mismatch problem is solved. A practical scene is sparse in the continu-ous range-velocity two-dimensional spaces. When the continuous spaces are divided intodiscrete grid points, targets may lay of the grid points, which causes model mismatch andsaps traditional compressed sensing methods. A new algorithm, namely Adaptive Match-ing Pursuit with Constrained Total Least Squares, is proposed to solve the mismatch prob-lem. The errors of the predefined grid points are casted as unknown parameters and thenestimated by the proposed algorithm adaptively. The estimate is utilized to calibrate themeasurement matrix correspondingly, which alleviates the mismatch problem and leads to robust reconstruction of the observed scene.A novel cognitive strategy is introduced into FAR. By exploiting the prior informa-tion of the observed scene, the cognitive mechanism further improves the recovery perfor-mance of compressed sensing. A new criterion, minimizing the Cramer-Rao bound(CRB)on the reconstruction errors, is used for optimizing carrier frequencies of FAR. Reducingthe CRB makes the sensing system more informative, and relieves interferences betweenreturns from diferent targets. According to the structure of the measurement matrix inFAR, an approximation to the criterion is also developed for lower computational com-plexity. For diverse potential applications of the cognitive manner, two types of modes,sequential and batch-oriented designs of carrier frequencies, are developed, respectively.