Research On DOA Estimation For Wideband Signals Using Sparse Reconstruction

As an important research direction of array signal processing,DOA estimation has a wide range of applications in military and civil fields.The traditional subspace-based DOA estimation methods have some shortcomings.For example,insufficient number of snapshots and source correlation will lead to unsatisfactory estimation performance.In recent years,as the theory of compress sensing(CS)theory gets mature and wildly used,researchers try to solve the DOA estimation problem by transforming it into a sparse reconstruction problem.As a result,a series of DOA estimation methods based on sparse reconstruction have been developed.Besides,with the wide application of wideband signals,many narrowband signal DOA estimation methods are also extended to broadband.This paper mainly studies DOA estimation of wideband signals based on sparse reconstruction.Two new methods are proposed by using the sparse Bayesian learning theory.The innovation points can be summarized as follows:1?To solve the common DOA estimation problem for quasi-stationary signals such as microphone sound source localization,we propose DOA Estimation for Quasi-Stationary Wideband Signals Using Block Sparse Bayesian Learning(QSW-SBL)algorithm.In this algorithm,we design a complex Gaussian prior distribution model with block diagonal covariance matrix,which is based on the spectral characteristics of the quasi-stationary signals.It can not only make use of the correlation of the signal spectrum in each frame but also control the sparsity of each frame by using the same hyperparameter matrix,which is based on the prior knowledge that all the frames share the same sparse structure.Therefore,this method can get more accurate DOA estimation results.2?Because the pattern-coupled sparse Bayesian learning(PCSBL)algorithm has a good estimation effect on the sparse reconstruction problem with unknown block prior information,that is,the element block size is not fixed,so we propose the matrix-block pattern-coupled sparse Bayesian learning(MB-PCSBL)algorithm which extends the PCSBL algorithm from one dimension to two dimensions.In this algorithm,we establish a new hierarchical Gaussian prior model that correlates the hyperparameter which controls the sparsity of the signal element with the hyperparameters of the eight adjacent neighbors in the form of a two-layer network.It can separately promote the non-zero elements and zero elements to get together to produce a matrix block structure.Moreover,the horizontal correlation and the vertical correlation are respectively controlled by different parameters.In this way,the method can be more flexibly applied to various scenarios,including DOA estimation for wideband signals.Finally,simulation results show the proposed algorithms can achieve higher estimation accuracy by using the information of all sub-bands even when the number of samples is very small.The cost is the increased computational complexity.