Research On Direct Position Determination Methods Based On Array Signals

Wireless localization has played an important role in the industrial production and military fields.In recent years,direct position determination(DPD)technique has received widespread attention from domestic and international scholars.The basic idea behind DPD is to locate transmitters directly from original sensor outputs without estimating intermediate parameters in a single step.Compared with conventional two-step localization methods which extract measurement parameters and then estimate the positions from them,this single-step localization technique offers superior performance in terms of higher accuracy and greater resolution capability,and it avoids data association as well.In addition,DPD can take advantage of certain signal properties(e.g.noncircular property),which is helpful in enhancing the localization performance.Considering that spatial multi-sensor arrays can provide more abundant information about the locations of the targets of interest,this dissertation concentrates on the study of DPD based on array signals.As different positioning systems have their own advantages and application scenarios,our research consists of three aspects,i.e.,single-station DPDs,multi-station DPDs,and reflector-aided DPDs.The main works and innovations of the dissertation are outlined as follows:1.The problem of single-station DPD for noncircular sources is investigated.According to the property of noncircular signals with maximum noncircularity rate,we establish an angle-based observation model for noncircular sources observed by a moving array in several time slots,and derive the corresponding Cramér-Rao bound(CRB)formula for the source positions.A single-station DPD algorithm is proposed based on extended subspace data fusion(SDF),which locates multiple noncircular sources in a decoupled manner.In addition to the proposed algorithm,we provide a necessary condition for identifiability of parameters and derive the closed-form expression for the localization mean square errors.Compared with the existing SDF-based DPD,our algorithm improves both the localization accuracy and the number of distinguished targets.2.The problem of DPD based on angle and Doppler using a single moving array is investigated.We establish a space-time signal model which combines angle and Doppler information through temporal data extending,and derive the CRB formula for the source positions.Two DPD algorithms based on angle and Doppler are proposed.The first algorithm formulates a real-valued cost function based on unitary space-time subspace orthogonality,leading to the decoupled locations of multiple targets.The second algorithm introduces the weighted subspace fitting idea to the DPD problem,and then an alternating projection procedure is used to solve the positions of multiple targets.These two proposed algorithms exhibit higher localization accuracy and greater resolution capacity than the spatial SDF-based DPD.The first one is more computationally efficient whereas the second one has higher accuracy and can approach the corresponding CRB.3.The problem of DPD for noncircular sources using separated stationary arrays is investigated.According to the time-domain and frequency-domain properties of noncircular signals with maximum noncircularity rate,we establish the time-domain and frequency-domain noncircular signal models,where the angles and time delays are incorporated.The CRB formula for noncircular sources is derived and is proved to be upper bounded by the associated CRB for circular sources.Based on the prescribed signal models,two DPD algorithms for noncircular sources observed by separated stationary arrays are proposed.The first algorithm formulates a cost function in the form of the smallest eigenvalue of a symmetric real-valued matrix by exploiting the subspace orthogonality of all frequency components.Then,a Newton-type iterative method is devised based on the matrix eigen-perturbation theory,thus reducing the computation load of the exhaustive grid search for solving the cost function without losing accuracy.The second algorithm establishes an maximum likelihood(ML)-based multidimentional nonlinear constraint optimization problem according to the time-domain noncircular signal model.A hybrid method combining alternating iterative idea and Newton-type iteration is developed to solve the prescribed ML estimators for the positions of multiple noncircular sources.These two proposed algorithms achieve higher localization accuracy and lower resolution threshold as opposed to the existing multi-station SDF-based DPD.The complexity of the first one is lower,whereas the the second one has greater performance robustness under low signal-to-noise ratio and small number of samples.4.The problem of DPD based on angle,time delay,and Doppler using multiple moving arrays is investigated.In the absence and presence of observer location errors,we derive the CRB formulas for the source position and propose the corresponding ML-based DPD algorithms.With the precise knowledge of observer locations and velocities,a Newton-type iterative method and a Taylor-series iterative method are developed to solve the ML-based cost functions for unknown and known signal waveforms,respectively.The proposed algorithm outperforms the two-step localization methods and obtains the asymptotically optimal estimation.As few efforts have been devoted to solving the DPD problem accounting for observer location errors,we quantitatively analyze the effect of observer location errors on the performance of the precedingly proposed DPD algorithm.Then combining the models of the observed signal and the observer location errors,the ML-based cost functions are formulated and solved relying on alternating iteration schemes,where two sets of unknowns are updated iteratively.This algorithm is capable to mitigate observer location errors,and it attains the corresponding CRBs in the presence of observer location errors.5.The problem of reflector-aided DPD is investigated.In the cases where the accurate locations of reflectors are available and where the small reflector location biases exist,we derive the CRB formulas for source positions are derived for the known signals.Additionally,a quantitative analysis of the effect of small reflector location biases on the DPD performance is presented.Furthermore,we propose two reflector-aided DPD algorithms for multiple targets with known signal waveforms.With accurate locations of the reflectors,the first algorithm makes use of the known uncorrelated waveforms to decouple the ML locations of multiple targets.This algorithm is superior to the traditional two-step localization methods and the existing subspace-based DPD algorithm,and it can achieve the asymptotically optimal locations.As the existing reflector-aided localization has always required very precise locations of reflectors,our second algorithm assumes that the nominal locations of the reflectors are available whereas small biases in the reflector locations exist.Taylor-series expansion is used to separate the unknown biases from the source positions,and thus the locations of multiple targets are decoupled into several location problems for each target by exploiting the information of uncorrelated waveforms under the assumption that the correlation structure of noise is unknown.This algorithm can remove spurious paths,and it demonstrates higher accuracy than the existing reflector-aided localization methods in the presence of small reflector location biases.In addition,its performance is not sensitive to the model of noise.