Research On Array Signal Localization Parameter Estimation Algorithm With Gain And Phase Perturbations

Signal localization parameter estimation is one of the important research contents in array signal processing.It has important application value in many fields such as radar,sonar,mobile communication,remote sensing exploration and so on.The signal parameters estimation theory under the assumption of ideal model has developed in depth.However,in practical engineering applications,such as incomplete calibration of array,non-uniformity of propagation medium and temperature or humidity.The above-mentioned non-ideal array receiving model destroys the accurate unbiased condition of receiving data array manifold.It causes the perturbations of the array received signal in gain and phase.The research show that when the gain and phase error exceed a certain threshold(e.g.the phase error exceeds 1/10 of a cycle),the performance of the parameter estimation method based on the ideal model assumption will decline rapidly or even fail.Therefore,the research on robust signal localization parameter estimation under array gain and phase perturbations is a key problem to be solved in array signal processing.In order to overcome the influence of array perturbations,many effective solutions have been proposed by scholars in recent years.However,most of the existing researches are based on the array gain and phase perturbations caused by incomplete array calibration.At this time,the perturbation is constant or slowly varying.However,when the perturbations produced by the non-uniform propagation medium is fast changing or even time-varying,the difficulty of calculation and processing of the latter is much more difficult,especially when the prior information of time-varying perturbations is unknown.Aiming at the problem of signal parameter estimation under array gain and phase perturbations,this dissertation discusses the causes of two kinds of gain and phase perturbations problems,establishes the mathematical models of perturbations respectively,and carries out the research on robust positioning signal parameter estimation based on sparse reconstruction,subspace method and deep learning method.The main innovations are as follows:(1)Aiming at the gain and phase perturbations caused by incomplete calibration of the receiving and acquisition part of the array element,an algorithm based on Log penalty under partially calibrated COLD array is proposed to obtain the joint estimation of DOA,polarization parameters and power,it still has good estimation performance when the gain and phase perturbations is large.This method extends the partially calibrated scalar array to the polarization sensitive array to obtain better estimation performance and polarization parameter estimation.The statistical characteristics of two channel received covariance matrix sum of COLD(concentrated orthogonal loop and dipole)array composed of electric dipole and magnetic dipole in polarization sensitive array are analyzed.The advantages of high estimation accuracy and high robustness of positioning signal parameter estimation under sparse reconstruction framework are used,a joint estimation algorithm of positioning parameters based on partially calibrated COLD array under array gain and phase perturbations is proposed.The initial gain and phase error is estimated through continuous multiplier function and simple algebraic operation.The direction of arrival of the signal is estimated by Log penalty and DC(difference of convex)decomposition,and the polarization parameters and power of the signal are solved.Simulation results show that the algorithm has excellent estimation accuracy,and the root mean square error does not increase with the gain and phase error,which shows that the proposed algorithm can effectively correct column gain and phase perturbation.(2)Aiming at the time-varying / fast varying gain phase perturbation caused by the non-uniform propagation medium of the signal,a gain phase perturbation estimation method based on the Hermitian structure of the covariance matrix is proposed and combined with near-field covariance matrix difference technique to suppress the influence of unknown stationary noise.Using the characteristics that the product of the anti angle elements of the covariance matrix is only related to the array gain phase perturbation and signal power,the product of gain phase perturbation and signal power is estimated.Thus,the interference of gain phase interference and additive Gaussian white noise is removed.However,the proposed method is based on the assumption of additive Gaussian white noise and is sensitive to unknown stationary noise.At that time,under the coexistence of variable gain phase perturbation and unknown stationary noise,the spatial difference matrix under the near-field source location model is constructed by using the symmetry of the stationary noise covariance matrix about the main diagonal.The spectral decomposition characteristics of the near-field difference matrix are deduced and proved.On this basis,the noise subspace is determined,and the location signal parameters are estimated by means of spectral peak search.The influence of unknown stationary noise on positioning accuracy is effectively reduced by eliminating noise components.At the same time,it is proved that the proposed algorithm can effectively avoid the pseudo peak problem of far-field difference algorithm.(3)Aiming at the problem of location parameter estimation with unknown structure of time-varying gain and phase perturbation dynamic covariance matrix,a deep learning method based on BP neural network is adopted,which improves the generalization of the algorithm.The core calculation of the deep learning theory is the multiplication and addition of the matrix and the nonlinear transformation of the matrix.When the number of array elements and the number of sources is large,there is no need for complex eigenvalue decomposition and other operations,which greatly reduces the computational complexity of the traditional estimation method.At the same time,this method is not sensitive to the effects of signal coherence,array gain and phase perturbation,array position offset and so on,it can effectively deal with the problem of array positioning parameter estimation under time-varying gain phase perturbation.Taking the covariance matrix of the array received data as the input and the timevarying gain phase perturbation and the target source azimuth information as the output,the nonlinear mapping relationship is established,and the model is trained through the data set.Simulation results show that the algorithm is more robust to gain and phase perturbations and additive noise and has lower computational complexity than the traditional location parameter estimation algorithm.