Research On DOA Estimation Algorithm Of Wideband Sound Source Based On Spatial Sparsity

Wideband signals have become an important research area for the estimation of the direction of arrival(DOA)of array signals due to the large amount of data and abundant frequency information.The theoretical research about the uniform linear array(ULA)is mature,and ULA has the characteristics of easy to build a model,so the DOA estimation of the wideband signal for ULA has become a research hotspot in recent years.In the thesis,DOA estimation of far-field wideband signal is explored,and the main contributions and innovations could be summarized as follows:(1)The first research content is the covariance fitting method based on sparse reconstruction.Because the second moment of the signal contains more information than the signal,a covariance fitting algorithm based on sparse reconstruction is proposed.The idea of compressed sensing is adopted in the algorithm to makes use of the sparsity of the signal in the spatial domain establishing the spatial sparsity model of the wideband signal,and it estimates DOA by sparse reconstruction algorithms reconstructing the signal covariance.The effectiveness of the proposed algorithm is verified by simulation experiments,and the algorithm of using orthogonal matching pursuit(OMP)not only can estimate the multiple sound sources,but also has smaller root-mean-square error than other sparse reconstruction algorithms under the same conditions because of the selection of optimal atoms and the introduction of orthogonality.(2)The second research is the covariance fitting method based on tensor domain decomposition.Firstly,to improve the covariance fitting method,the algorithm is improved by from the l₀ norm to the l₂ norm and the l₁ norm used to constrain.Secondly,the abundant frequency information of wideband signals is used to construct tensors,then the high-order singular value decomposition technique is used to decompose the signals in the tensor domain to extract the principal components to reduce the impact of noise.Finally,the improved covariance fitting algorithm is used to estimate DOA of signals.The performance of the algorithm is verified by simulation experiments,and it is different from the traditional algorithms that the algorithm does not need to predict the number of signal sources,which is also its biggest advantage.(3)The third research is weighted projection subspace algorithm.In projection subspace algorithms,each sub-band data is divided into signal subspace and noise subspace,and all noise subspaces are combined,and the signal subspace of the reference band is projected to other sub-bands.But the method of how the reference band is selected hasn't been shown,and the signal subspace of the reference band is only transformed by the projection.Aimed at those two points,an improved algorithm is proposed.Firstly,standards for reference frequency band selection is set,and the algorithm no longer selects a single band as a reference band,and these bands which make a clear distinction between signal subspace and noise subspace are selected as reference frequency bands.Secondly,testing the matrix structure is improved,the guide vector and signal subspace of the reference bands are used to construct weights,then the testing matrix is weighted.the simulation results verify the effectiveness of the proposed algorithm is verified by simulation experiments.At the same time the proposed algorithm can make the spectral peak more obvious and a smaller root mean square error under low SNR.(4)The fourth research is based on frequency vector algorithm.It is the essence of the subspace algorithm that the same sub-band arrays in different time delay of phase is used to estimate signals DOA.Different sub-bands also have phase difference on the same array element,and in view of it and the established wideband signal model,an algorithm based on frequency vector is proposed.In this algorithm,the frequency dimension is evenly divided into several sub-bands,so the phase difference between adjacent sub-bands is consistent.In view of this,the classical algorithm is used for DOA estimation.The simulation results show that the algorithm based on frequency vector has better performance under the condition of fewer array elements.