Research On The Time-frequency Localization For Complicated Near-field Sources Based On Sparse Solution

As an important problem in array signal processing, near-field sourcelocalization studies how to obtain the source localization from the received data ofarray, which contains the direction of arrival and range parameters. Also near-fieldsource localization is widely applied in radar, sonar, speech enhancement, electronicsurveillance and seismic exploration etc. In recent years, aiming at the near-fieldlocalization problem, many excellent achievements have been achieved. But they donot sufficiently exploit the signals’ time-frequency domain information. And theyusually have some drawbacks in coherent signal processing and high accuracy withhigh computational complexity. We consider the source localization problem based onsensor array as a sparse representation problem, which brings a new ways to solveabove problems for localization method. So in this dissertation, combinedtime-frequency analysis with sparse representation, we research the localizationproblem of near-field sources.To overcome the weakness that traditional methods can not deal with thecoherent sources directly, the dissertation incorporates the sparse representation intothe localization methods to propose sparse localization methods. In order to avoidheavy computational load, the greedy algorithms are chosen as the reconstructionalgorithm. We conduct extensive numerical experiments analyzing the behavior ofour approach and comparing it to existing source localization methods. This analysisdemonstrates that our approach has important advantages such as low computationalcomplex, robustness to noise and robustness to coherent sources.In order to improve the accuracy of above sparse localization algorithm, thedissertation mainly studies a sparse localization method based on dynamic array,which jointly exploits the received data of different orientation array, that equal toadding the sensor and sample number. To reduce the influence of atoms in sparse baseon the resolution for localization method, the iterative hard thresholding(IHT) basedon coherence-inhibiting is studied, which prunes some support atoms to leave the lowcorrelation support atoms for the sparse reconstruction. A modified method isproposed for the situation where false DOAs are obtained due to the closely space of sources. The modified method is based on the approximately equal energyinformation of sources. The simulation results show that the improved methodsprovide improved resolution and high estimation accuracy in the low SNR, and canefficiently localize the coherent sources.To solve the localization problem, combining the time-frequency distributionswith sparse representation, we study localization algorithm of near-field sources, thetime-frequency distribution of unstationary signal has regular characteristic, whichcan be incorporated into localization method to establish the time-frequency sparselocalization model, which can be solved by greedy algorithm. The proposedtime-frequency localization method provide super resolution and high accuracy, andno requirement about the less source number than sensors.If the incoming sources are unknown unstationary signals, we firstly propose thesignal detection method based on MP algorithm and Hough transform. The simulationresults show that the method provide high correct probability even at low SNR. But itsuffers heavy computation burden due to the number of atoms in Garbor dictionary,so we propose to construct the dictionary with fixed time, and instead the WVD bythe localization of selected atoms. And then the Hough transform is used to detect theFM parameters. Simulation demonstrates the advantages and disadvantages of twodetection algorithms.The main work can be summarized as follows:(1) Based on the spatial sparsity, the sparse localization model for near-fieldsource was built, and proposed near-field localization method using greedy algorithm.The simulation results show that sparse localization algorithm provide moderatecalculation load, high estimation accuracy and resolution, at same time, it can dealwith coherent sources.(2) Using the sparsity of nostationary signal in time-frequency domain, themethods for detecting FM parameters based on received array are proposed, whichcombined greedy algorithm and Hough transform. The simulation results show thatdetection methods provide high correct probability. So they can be efficiently appliedto localization algorithm, which made some contributions to enhance the performanceof the algorithm.(3) The sparse time-frequency localization model is derived, and proposed thesparse time-frequency localization method. Due to the localization of sources can be estimated separately, the method provided high estimation accuracy and resolution,and no requirement about the less source number than sensors.The near-field localization algorithms presented in this paper can be applied toany sources that the source signals are correlated or uncorrelated, which has broadapplication prospects. The detection algorithm based using array for nonstationarysources have good performance, which has a certain reference value.