Research On Underwater Passive Source Localization In Shallow Water

Underwater passive source localization is the focus and hotspot of underwater acoustics.It plays an important role in the fields of marine development,marine engineering and national defense security.In shallow water environment,the acoustic channel is complex and changeable,thus the characteristics of sound propagation must be considered for underwater source localization.Recently,matched field processing(MFP),which combines the sound propagation model and the array signal processing,is one of the main methods for underwater source localization.Although matched field processing has been successful and widely used in underwater source localization,its localization performance may be seriously affected or restricted in some situations,such as environmental mismatch,surface interference and model parameter uncertainty.In view of the problems mentioned above,the main works and research results of this paper are summarized as follows:In view of environmental mismatch,a robust compressive adaptive matched field localization algorithm based on convex optimization and random projection is proposed.From the perspective of improving the robustness,on the basic of the worst-case performance optimization,the proposed algorithm first imposes constraint on the uncertainty set of the replica field vector for the adaptive matched field localization algorithm.Then,the nonconvex optimization problem caused by the nonconvex constraints is converted to a convex one and solved with second-order cone programming.Finally,to gain better robustness,the proposed algorithm is generalizaed from narrowband case to broadband case using incoherent superposition.From the perspective of reducing the complexity,the random projection is applied to reduce the dimension of the replica field vector and the pressure field vector,so that the complexity is reduced while the robustness is kept.The effectiveness of the proposed algorithm is verified using both simulation data and experiment data.The results show that it can effectively maintain the mainlobe and suppress the sidelobe,and improve the robustness of the adaptive matched field localization algorithm.In view of surface interference,a matched field localization algorithm based on iterative orthogonal projection is proposed.The main idea is to construct the orthogonal projection matrix according to the localization information of the surface interference to suppress the surface interference.Under the premise that the number and locations of the surface interference and the underwater target are unknown,the proposed algorithm firstly estimates the number and the locations roughly through orthogonal projection,and then updates surface interference locations and optimizes the underwater target location by iterative processing.The simulation results show the effectiveness of the proposed algorithm.After two or three iterations,the location of the underwater target can be estimated in the presence of the surface interference.In view of model parameter uncertainty,an underwater source localization algorithm based on generalized regression neural network(GRNN)is proposed.The underwater source localization problem is modeled as a regression problem and is solved by the generalized regression neural network.The cross spectral density matrix of the pressure field and the corresponding source location are used as the input and output of the neural network.The optimal value of the only one parameter,the spread factor,is determined using cross-validation.The proposed algorithm is in theory data-driven.The characteristics of the acoustic channel are directly learned from the training data.For narrowband and broadband sources of the simulation data and experiment data,the conventional method of MFP,the classification method of feedforward neural network(FNN)and the regression method of GRNN are used to demonstrate the localization performance of the algorithm.The results show that FNN and GRNN achieve a better localization performance than MFP and GRNN performs a litter better than FNN.