Research On Joint Estimation Of Multidimensional And Multi-polarized Parameters In High-frequency Surface-wave Radar

In high frequency surface wave radar(HFSWR)systems,sufficiently exploiting differences in multiple dimensions is an effective way to improve the effectiveness of clutter suppression and the performance of target detection.For instance,using 2D directionof-arrival(DOA)and polarization information is shown to be substantially benefit for performance improvement.Consequently,efficiently and reliably estimating 2D DOA and polarization parameters is of great significance.In HFSWR applications,estimators are usually required to cope with highly-correlated signals and small number of snapshots.Traditional methods for array signal parameter estimation enjoy reliable statistical interpretation,but usually require sufficient snapshots and uncorrelated sources.In contrast,most sparse-based estimators are well compatible to coherent sources and small number of snapshots,whereas intractable hyperparameter tuning is sometimes required,or the computational burden in multidimensional cases is unbearable.Therefore,to enhance the practicability,combining advantages of the two algorithmic frameworks is expected in the considered application.To that end,this thesis attempts to derive powerful sparse-based estimators based on reliable statistical criteria,which are thus expected to enjoy favorable adaptivity as well as wide availability to coherent signals and insufficient snapshots.To improve the computational efficiency,developing effective dimension-reduced estimators and numerical methods for fast implementation is also considered.The main contributions of this thesis are summarized as three aspects as follows.Firstly,for 2D DOA estimation with mono-polarized arrays,dimension-reduced estimator available for arbitrary planar arrays is proposed.Sparse-based methods are considered since available snapshots after coherent accumulation are rather few.To cope with the high computational complexity resulting from huge dictionaries involving 2D parameters,the continuous approximation based dimension-reduced estimator(CADRE)is proposed.Inspired by that the two axial angles can be estimated independently based on the two orthogonal linear sub-arrays of an L-shaped array,basis matrices with decoupled2 D axial angles are constructed by the approximate linear representation of the column space of steering vectors,which thus allows for obtaining 2D DOA estimates by two independent 1D DOA estimations.To realize DOA estimation based on such basis matrices,an adaptive block-sparse recovery method in the stochastic maximum-likelihood(SML)sense,as well as group-wise cyclic minimization(GCM)based numerical algorithms for its fast implementation,are proposed.To mitigate the performance loss caused by dimension reduction,a refining scheme based on weighted signal subspace fitting(WSSF)is also provided.The proposed methodology enjoys super-resolution estimation of 2D DOA and superior performance for coherent sources and few/single snapshot(s),and is also remarkably faster than existing 2D DOA estimators with comparable performance.Secondly,for DOA and polarization estimation with multi-polarized arrays,a novel data model based on block-sparse representation is established,and estimators based on block-sparse recovery are proposed.Inspired by sparse-based DOA estimators,the received data of multi-polarized arrays is formulated as block-sparse representation with redundant dictionary to cope with coherent sources and small number of snapshots.As arbitrary polarized electromagnetic waves can be decomposed into two orthogonal components,without involving all the undetermined 4D parameters,the dictionary is formed by steering vectors corresponding to the two orthogonal components of signal from each potential DOA.The size is thus only twice that of the conventional dictionary for DOA estimation.Based on such model,the likelihood-based estimation of block-sparse parameters(BLIKES)originating from the SML criterion is proposed,and its theoretical equivalence to Lasso-type methods frequently used for regression analysis is also proved.As a consequence,the efficient adaptive group square-root/least-absolute-deviation Lasso based estimator(EAGLE)equivalent to BLIKES is further obtained.To improve the computational efficiency,the generalized adaptive preconditioned alternating direction method of multipliers(GAP-ADMM)is proposed for efficiently solving the optimization sub-problems in EAGLE,which makes EAGLE available for fast implementing BLIKES.Moreover,both BLIKES and EAGLE provide modified versions robust to nonuniform noise.The proposed algorithms are widely applicable to arbitrary polarization sensitive arrays and completely/partially polarized signals,and also enjoy significant speed advantage over existing algorithms with comparable performance.Thirdly,to provide reliability guarantee and performance evaluation benchmarks for parameter estimation algorithms,the manifold ambiguity of multi-polarized arrays and the Cramér-Rao bound(CRB)as lower bound for the root mean squared error of joint DOA and polarization estimation are analyzed.In terms of manifold ambiguity,theoretical results for multi-polarized linear arrays and spatially spread multi-polarized arrays more practicable in HFSWR systems are absent in open literatures,while this thesis explicitly provides the dimension of identifiable parameters and ambiguous parameter sets.As to the CRB for DOA and polarization parameters,existing one is established upon the stochastic signal assumption,while we further derive the closed-form CRBs for deterministic signals with both known and unknown waveforms.Furthermore,the underlying connections between the deterministic CRB and the stochastic CRB as well as the manifold ambiguity are revealed.Besides,the effectiveness and the superiority of the proposed algorithms,as well as the correctness of the provided theoretical results,are verified by both simulations and experimental data.