Near-Field Source Localization Based On Sparse Reconstruction Approaches

The acoustic source positioning is a hot research topic in speech signal processing and a very important step for subsequent processing.The most of existing source positioning approaches is dedicated to the case of far-field source positioning.However,when the source signal is in the near-field,the assumed plane wave in the far field source is no longer valid.In such case,the DOA and distance parameters need to be jointly estimated to achieve the near-field source positioning.Based on the source signal in the near-field,this paper uses the subspace-like POR algorithm and the sparse reconstruction algorithm to achieve narrowband and wideband source positioning respectively.On this basis,the estimated performance of near-field sources under 1-Bit conditions is also discussed.The specific work content includes the following aspects:1.Most near-field source localization algorithms look at the source localization problem as a two-dimensional(2-D)non-linear problem,which leads to a high computational complexity.In this chapter,we use symmetrical uniform linear arrays,and base on that,special spatial correlation sequences are constructed to decouple angle and distance.By doing so,the 2-D parameter estimation problem becomes two one-dimensional estimation subproblems.In the near-field narrowband source,a POR algorithm similar to the subspace-like is developed.Compared with the traditional subspace algorithm,the algorithm does not need to perform feature decomposition and does not need to know the prior information of the number of source signals,which greatly improves Practicality of the algorithm.In addition,based on the sparse characteristics of the signal,a sparse reconstruction algorithm is proposed.Because the algorithm needs to construct a grid dictionary,the base mismatch problem may occur,which makes the estimation performance of the algorithm decrease.In order to effectively solve this problem,this paper uses the atomic norm minimization.Simulations and actual data are conducted to show the performance of the proposed algorithm.2.In this chapter,the source positioning in the case of broadband is studied.By decomposing the broadband signal into multiple narrowband signals,the parameters of each narrowband segment are estimated using the POR and sparse reconstruction algorithms.After that,the estimation results of each narrowband segment are summed and averaged to achieve near-field broadband source localization.The numerical studies demonstrate that the proposed method outperforms other approaches and it is of low complexity.