Study On Several Key Problems Of Coherent Doa Estimation

Direction-of-arrival(DOA)estimation is a problem of significance in array signal processing.In the presence of coherent sources,the signal covariance matrix is rank deficient,which results in an inaccurate estimate of signal or noise subspace.In such cases,the source number detection problem will also be challenged.It is difficult for the conventional detection algorithms,e.g.,Minimum description length and Akaike information criterion,to provide accurate source number information,and the detection probability is usually not satisfied.Furthermore,under certain challenging scenarios,such as in the sample-starved regime where relatively few snapshots are available,and in the case of highly correlated or even coherent signals,subspace methods suffer severe performance degradation which is known as threshold effect.It refers to the severe and abrupt performance degradation that happens when the SNR and/or the sample size drops below a certain threshold,and it is the characteristic of many subspace based algorithms.The above mentioned problems are all factors causing the failure of subspace methods.It should be mentioned that although the classical spatial smoothing technology is able to handle coherent sources,this method still has some shortcomings,e.g.,the decoherency is at the expense of losing array aperture which causes performance loss of DOA estimation algorithms,and it requires to know the number of coherent signals as a priori.In this thesis,we aim at dealing with the coherent DOA estimation problem by designing high performance algorithms.The associated results are included as follows:1)It is attractive to devise a coherent DOA estimation algorithm that has the largest degree-of-freedom.The problem of coherent DOA estimation is revisited through a new principal-singular-vector utilization for modal analysis(PUMA)approach,which has statistical efficiency when SNR goes to infinity.PUMA is built upon the linear prediction theory,which transform the DOA estimation problem as a root finding problem.Since PUMA has closed-form formula in each iteration and it only requires less than three iterations to converge,its complexity is much lower than the ML method,but the nice thing is that PUMA retains the ML's statistically efficiency,and this property has been theoretically proved and verified by simulations.Furthermore,PUMA does not need to known the coherent source number information.It should be mentioned that when there is only one source,the theoretical variance of PUMA is equivalent to the Cram?er-Rao bound.In order to further reduce the complexity and improve the threshold performance,a unitary PUMA algorithm is devised.Simulation results showcase the effectiveness of PUMA and verify the correctness of the theoretical proof.2)To handle the case where source number is unavailable,two novel DOA estimators that do not require the source number information have been developed for white and spatially correlated Gaussian noise,respectively.In the first algorithm,the coherency is decorrelated via forming a group of Toeplitz matrices that share the same joint diagonalization structure,and can be employed to devise a DOA estimator without knowing the source number information.The second algorithm is similar to the first one,except that before the decorrelation step,it utilizes the fourth-order cumulant to deal with spatially correlated Gaussian noise.Generally speaking,under the same scenario,the performance of the second approach is better than the first one,but at the expense of high complexity.3)In order to improve the estimation accuracy of DOA estimation algorithms,two novel efficient approaches with excellent threshold performance are proposed.Firstly,a modified PUMA(EPUMA)approach is proposed to improve the threshold performance by first generating(P + K)DOA candidates where P ? K,and then judiciously selecting K of them.Secondly,a two-step reliability test(TSRT)that is based on pseudo-noise resampling technique has been devised.This method can be easily applied to the existing DOA estimators.The main idea behind is to perturb the noise contained in the samples to change the subspace by adding pseudo-noise,resulting in a higher probability of obtaining the correct DOA estimates.The TSRT is able to identify those DOA estimates that are closed to the true DOAs.It enables to resample the data for several times to collect enough good DOA candidates,and then use a properly designed selection rule to choose K of them as the final DOA estimates.In doing so,the threshold performance can be improved.Finally,it is demonstrated by means of computer simulations that the proposed algorithms are able to provide attractive performance over the state-of-the-art approaches.